23:02 | shinokubo | joined the channel | |
23:52 | gcolburn | left the channel | |
00:27 | shinokubo | left the channel | |
07:09 | Bertl | off to bed now ... have a good one everyone!
| |
09:27 | se6astian | joined the channel | |
09:28 | shinokubo | joined the channel | |
09:29 | se6astian | good morning
| |
10:20 | philippej | joined the channel | |
10:24 | philippej | left the channel | |
11:06 | Guest39465 | joined the channel | |
11:07 | Guest39465 | left the channel | |
11:08 | bootstrap | joined the channel | |
11:10 | bootstrap | greetings: I just learned of your project, and might be able to help. I'm an electronics engineer, designed similar but lower-rez camera with aptina MT9P401I12STC sensor, and have worked with large CCDs for professional astronomy projects.
| |
11:12 | bootstrap | First I'd like to learn a little more about your project - technical details.
| |
11:16 | FergusL | Hello bootstrap !
| |
11:16 | FergusL | welcome here
| |
11:17 | bootstrap | aloha.
| |
11:17 | bootstrap | Are you working on the camera?
| |
11:17 | bootstrap | I've read about the sensor, but don't know much else.
| |
11:17 | bootstrap | A note about something you said on your web-page: My design is 8-layer PCBs, and they aren't much more expensive than 2 or 4 layer PCBs, but much better performance.
| |
11:17 | FergusL | no, as in I don't own a prototype nor have I worked on low level stuff on the camera
| |
11:18 | bootstrap | Are there any "real" prototypes yet?
| |
11:18 | bootstrap | I saw a couple still images, nothing more.
| |
11:19 | FergusL | there are ! but only for engineers, I'm just helping wherever I can
| |
11:19 | FergusL | so I'm not working on the camera, but I could say I'm working on Apertus as a whole, just as every other volunteer
| |
11:19 | bootstrap | So, maybe they don't need electronics engineer types any more?
| |
11:19 | bootstrap | Yes, I understand. I'm just trying to figure out how far along you are.
| |
11:19 | FergusL | I think we do :) you'd have to talk about this with Bertl and first with se6astian
| |
11:20 | FergusL | have you got specific questions regarding the sensor ?
| |
11:20 | bootstrap | The camera I designed was less demanding (but lots cheaper to build)... only 2595x1944 resolution.
| |
11:20 | bootstrap | I downloaded the sensor datasheet, which will probably be enough for now.
| |
11:20 | bootstrap | I'm more interested in what else is inside the camera.
| |
11:21 | bootstrap | I saw someone mentioned "FPGA", but nothing else.
| |
11:21 | FergusL | indeed, that and the informations the sensor manufacturer sent us are our only sources
| |
11:22 | bootstrap | I have worked with several CCDs, especially in astronomy, so I pretty much understand them (though not this specific one).
| |
11:22 | FergusL | Okay, great
| |
11:22 | bootstrap | One of the biggest challenges is... getting all that data out of the camera and into a computer at such high speed.
| |
11:23 | FergusL | yes indeed for now it's working off an FPGA board from Xilinx
| |
11:23 | bootstrap | Most folks "give up" and make the camera body accept flash memory to store the images until they can be offloaded to the PC.
| |
11:23 | bootstrap | Getting the data into a PC in real time is extremely challenging!
| |
11:24 | FergusL | yes, you'd have to design a bit on the computer side
| |
11:24 | bootstrap | So you'll have a custom PC card that plugs into a PCIe slot in the PC ???
| |
11:25 | FergusL | Apertus ?! no !
| |
11:25 | FergusL | or rather, I should say Axiom
| |
11:25 | bootstrap | I keep expecting nvidia to have a fast video input on their GPU cards. Maybe they do (haven't checked lately), but that's the best solution, because the image is already in the GPU where the fastest processing can happen.
| |
11:25 | FergusL | I was about to say that not all camera owners want or can have a computer sitting near the camera
| |
11:26 | FergusL | what kind of GPU cards ? their standard gaming-ish gpu cards ?
| |
11:26 | bootstrap | I just found your site so I can't keep Apertus and Axiom straight in my mind.
| |
11:26 | bootstrap | Yes, high end ones like GTX680 or above.
| |
11:27 | FergusL | Okay Apertus is both the company and association running the operations and gathering the followers, volunteers and, hopefully, the manufactering (in very general terms) . Axiom is the name of the camera
| |
11:28 | bootstrap | Okay, I'll try to remember that.
| |
11:28 | FergusL | I hope too, and what kind of fast video input ? are you thinking about digital like SDI or consumer HDMI or Displayport ?
| |
11:29 | bootstrap | Well, the trick is finding a video input into the card. video output is not so rare.
| |
11:29 | bootstrap | What kind I don't care... I can design hardware! :-)
| |
11:29 | FergusL | I don't think they do and if they ever do, the market for gpu cards will have changed a lot from now
| |
11:30 | FergusL | I mean, it's a crucial point to me, it makes it a very different product
| |
11:30 | bootstrap | Yes, it would change everything if nvidia or amd accepted video input directly into the GPU card... but I don't think they have ... yet.
| |
11:30 | FergusL | for now, all computers can have (unless one can design a pci card :-) are a very selective range of often ecpesnive products that seem to have thought to fit in very specific niches
| |
11:31 | FergusL | indeed
| |
11:31 | bootstrap | Yes. I assume you want to make the price of this camera reasonable (for a 4K rez device).
| |
11:31 | bootstrap | Like maybe $4K ???
| |
11:31 | FergusL | hm, that info is on the website, that's for sure
| |
11:31 | bootstrap | I didn't see that (or forgot). What's the price goal?
| |
11:31 | FergusL | http://axiom.apertus.org/
| |
11:32 | FergusL | below 10k
| |
11:32 | bootstrap | Okay, so $4K would make you smile. :-)
| |
11:33 | bootstrap | Is that PCB shown on that page a real PCB you made for the product?
| |
11:35 | bootstrap | Some places on the page say "this is a concept", and other places say "we've shot footage". What's the actual status?
| |
11:35 | bootstrap | And do you need an electronics engineer or not? Sounds like you might have one (or several) already.
| |
11:35 | FergusL | I don't think an initiative of our magnitude can achieve $4k
| |
11:35 | bootstrap | Depends on the sensor, which I read somewhere is $1000 to $1500 range.
| |
11:36 | bootstrap | But that wasn't the manufacturer, so it could be wishful thinking.
| |
11:36 | FergusL | the status is a working prototype that we use to calibrate the sensor, there are... 3 I believe
| |
11:36 | FergusL | there is one electronics engineer actively working, he's Bertl, present here but away it seems
| |
11:36 | bootstrap | Are the prototypes supposed to be the final design, or just "something" to get images out?
| |
11:37 | FergusL | something to get images out
| |
11:37 | bootstrap | Okay.
| |
11:37 | FergusL | https://www.apertus.org/alpha_prototype few pictures there
| |
11:38 | bootstrap | One thing I was confused about. Is this camera supposed to hold images inside (in flash memory chips), or immediately flow the data into a PC ?
| |
11:39 | FergusL | that design was considered for a long while : very high speed connection to a computer but was considered a flaw by a lot of potential users
| |
11:40 | bootstrap | Yes, i understand. My latest design flowed directly to the PC (each horizontal pixel row at a time), but my sensor was only 2592x1922.
| |
11:40 | bootstrap | For your sensor, flowing straight through would be quite a challenge!
| |
11:40 | bootstrap | You'd either need 10Gbps ethernet or maybe USB3 but not sure.
| |
11:40 | FergusL | and now yes it is supposed to hold images and video inside
| |
11:41 | FergusL | 10GbE was the suggested solution
| |
11:41 | bootstrap | Yes. I developed a lossless image compression scheme that helped me make mine work.
| |
11:41 | bootstrap | Unfortunately it only compresses about 25%, but that was enough for my system.
| |
11:42 | bootstrap | It is much easier to store the data in standard flash memory chips, which can then be removed and inserted into a PC.
| |
11:42 | bootstrap | For cinema, that is MUCH better, but I am interested in robotics vision applications, for which immediate flow-through is necessary.
| |
11:43 | bootstrap | In the end, I decided the best way is to create two separate sets of electronics, one for each application.
| |
11:44 | bootstrap | Especially in your case (with higher bandwidth).
| |
11:44 | bootstrap | Would you say the large majority of people interested want this for cinema?
| |
11:45 | FergusL | it depends, it means you have to implement a full file handling based on a given spec like raw video or a still format and all the electronics that goes with it
| |
11:45 | FergusL | absolutely yes
| |
11:45 | FergusL | and absolutely agreeing too with the idea of two sets of electronics
| |
11:46 | bootstrap | Though you probably won't hear from that many people, the robotics people would go crazy for a cheap 4K camera with real time image flow through.
| |
11:46 | bootstrap | But you will hear from lots of wannabe movie makers, and obviously have.
| |
11:46 | FergusL | what are the exact usecase ?
| |
11:47 | FergusL | and more importantly, why that much res ? what is to you an ideal framerate for robotics ?
| |
11:47 | bootstrap | The most common one is in factories, where the actions robots are taking must be corrected every 1/60 second (or 1/30 second) based upon what the robot sees.
| |
11:48 | bootstrap | Well, there is NEVER enough resolution.
| |
11:48 | FergusL | hahaha
| |
11:48 | bootstrap | And NEVER enough speed either.
| |
11:48 | bootstrap | My scheme sent the horizontal rows of pixels out as the CCD generated them, so the algorithms could start thinking about what to tell the robotics motors before the whole frame even got sent out.
| |
11:49 | bootstrap | The algorithms in the PC.
| |
11:49 | bootstrap | You just cannot wait for the image in robotics!
| |
11:50 | FergusL | and that is was useful ahead time for calculations ?
| |
11:51 | FergusL | I hope you can stay here for a while, Bertl and se6astian will be able to answer your questions with more details
| |
11:51 | bootstrap | Let's say some moving object (or the moving "hand" of the robot) is near the top of the frame. After a couple dozen horizontal rows of pixels come out, you can see where that stuff is, and immediately make motion decisions.
| |
11:51 | bootstrap | I can stay for a while.
| |
11:51 | bootstrap | BTW, where are the most central people located in the world?
| |
11:52 | FergusL | Europe
| |
11:52 | bootstrap | So... must be about 10 or 12 hours later than Hawaii.
| |
11:53 | FergusL | 13:53 here
| |
11:54 | FergusL | I'm in France, people are also in Belgium, UK, Austria
| |
11:55 | bootstrap | Okay... so I'd be the "odd man out" as usual. hahaha.
| |
11:56 | bootstrap | Even in your prototype, you must have some intermediate electronics of your own, right?
| |
11:56 | FergusL | exactly
| |
11:56 | FergusL | did you look at the picture ?
| |
11:56 | bootstrap | Can't interface the sensor to a standard Xilinx prototype PCB.
| |
11:57 | bootstrap | The photo that shows a PCB in a housing? Yes, I see that.
| |
11:57 | FergusL | https://www.apertus.org/sites/default/files/alpha-prototype-2d.jpg the purple board behind the lens is what we've designed
| |
11:57 | bootstrap | yes, i saw that too.
| |
11:58 | bootstrap | so i guess i'm seeing a small 60mm square PCB (yours) above a xilinx demonstration PCB ???
| |
11:58 | bootstrap | maybe more like 50mm square?
| |
11:59 | FergusL | yes
| |
11:59 | FergusL | there is a poll on the right about your desired application
| |
12:01 | bootstrap | ah, so about 88% for cinema-like apps, and maybe 12% for robotics.
| |
12:01 | bootstrap | i'm surprised it isn't more like 98% versus 2%.
| |
12:02 | bootstrap | but actually, you'll get a huge number of orders for factory automation and other robotics if you have a suitable model.
| |
12:02 | bootstrap | those folks have MONEY out the wazoo.
| |
12:02 | bootstrap | they don't talk much, they don't spend much time on the internet or anything, but they spend money like crazy.
| |
12:04 | bootstrap | So clearly you are now settled on making the first version for cinema, and will accept conventional flash memory chips (several, cuz you need parallel access to get enough speed).
| |
12:07 | FergusL | are you thinking about this as buffer memory ?
| |
12:07 | FergusL | before writing to an actual support like SSD drives
| |
12:08 | guesst | FergusL: industrial vision is insane... i was asked if I can supply my camera for filming train crashes :) that is a demanding app
| |
12:08 | bootstrap | hahaha... not all industrial/robotics apps are that crazy... or demanding.
| |
12:09 | FergusL | I do realise that
| |
12:09 | bootstrap | frankly, 150fps is outstanding.
| |
12:09 | bootstrap | i think i saw that number somewhere (maybe lower rez though).
| |
12:09 | bootstrap | or maybe that was 8-bit A/D.
| |
12:10 | FergusL | yes
| |
12:10 | FergusL | 150 is with lower res
| |
12:10 | guesst | yes?
| |
12:11 | FergusL | don't remember the exact number
| |
12:11 | bootstrap | maybe lower depth too (8-bits versus 12-bits).
| |
12:11 | bootstrap | for industrial, usually 8-bit depth is okay.
| |
12:13 | bootstrap | Oh, I missed your question about buffer memory. Yes, I was thinking about writing directly into [very high speed versions of] those little flash memory thingies that DSLRs and video cameras accept.
| |
12:14 | bootstrap | Each one can accept perhaps 1Gbps rate, which is too little, so at least 4 to 8 would be required to write in parallel.
| |
12:15 | guesst | usually all storage devices fake their write speed as it is not stated with: incompressible over whole capacity
| |
12:15 | bootstrap | The alternative is to record on DRAM, then sit around for a while after taking a shot for the data to download to a PC via 1Ge.
| |
12:16 | bootstrap | Yes, I understand that, which is why it is necessary to either select certain special brands, or downgrade expected bandwidth.
| |
12:16 | bootstrap | And depending on what happens, we can get another 25% by lossless compression.
| |
12:17 | bootstrap | Personally, I totally HATE lossy compression.
| |
12:17 | guesst | i hate even lossless (due to power efficiency)
| |
12:17 | FergusL | you're def the odd man out !
| |
12:18 | se6astian | hi bootstrap
| |
12:18 | se6astian | just got back to the channel
| |
12:18 | bootstrap | on compression? well, i have a lossless compression technique that takes VERY little processing power.
| |
12:18 | se6astian | let me read up what you discussed so far :)
| |
12:18 | bootstrap | But the downside is, it only compresses about 25%. That's not much for a compression technique!
| |
12:18 | bootstrap | hi se6astian!
| |
12:20 | bootstrap | A few years ago I was writing some image processing software, and it just LOVED to enhance "image details" that were actually artifacts from lossy compression! DROVE ME NUTS.
| |
12:21 | bootstrap | Hence I swore off lossy compression for anything I make. Hahaha.
| |
12:22 | guesst | i think people do not understand RAW and lossless
| |
12:22 | guesst | not vendors do :)
| |
12:22 | guesst | BM is stating lossless dng for pocket cam, when you look to the file it states: JPEG lossy ;)
| |
12:23 | bootstrap | Huh. Is the file wrong, or it really is JPEG lossy?
| |
12:23 | guesst | not sure if lossy/lossless is optional
| |
12:23 | bootstrap | Lossy compression can get enormous compression ratios... but... well... drives me nuts.
| |
12:24 | bootstrap | I figure, why have a big honking sensor, and then throw away detail? Duh.
| |
12:26 | bootstrap | Which is better, 5 megapixels lossless, or 10 megapixels lossy? Personally, I'd rather have the 5 megapixels.
| |
12:28 | se6astian | currently there are several options we consider
| |
12:29 | se6astian | Dual 3G-SDI
| |
12:29 | se6astian | SATAIII from FPGA High speed transceiver
| |
12:29 | se6astian | or USB3
| |
12:29 | se6astian | or SFP+ (10G)
| |
12:29 | se6astian | SATAIII would be the most convenient one for film makers
| |
12:29 | bootstrap | for direct streaming of image data to PC ???
| |
12:30 | se6astian | there is no PC
| |
12:30 | bootstrap | or for later transfer?
| |
12:30 | se6astian | well in SFP+ case there would have to be a PC
| |
12:30 | bootstrap | is this for cinema we're talking about now?
| |
12:30 | se6astian | yes
| |
12:30 | bootstrap | so this is for AFTER taking shotgs.
| |
12:30 | bootstrap | so this is for AFTER taking shots.
| |
12:31 | bootstrap | start filming... stop filming... then what? transfer to ?????
| |
12:31 | se6astian | no, its the recording media interface
| |
12:31 | bootstrap | Okay, so buy solid state drives?
| |
12:31 | se6astian | record video -> dump to SATA SSD for example
| |
12:31 | se6astian | while recording
| |
12:32 | bootstrap | without compression, or with?
| |
12:32 | se6astian | uncompressed
| |
12:32 | se6astian | raw
| |
12:32 | se6astian | up to 12 bit
| |
12:33 | bootstrap | okay, let me do a quickie here: 4K x 3K = 12M * 1.5 bytes per pixel * 60 fps == bandwidth?
| |
12:34 | se6astian | 4096x3072 x 12bit x 25FPS = 3.6Gbit/s
| |
12:35 | bootstrap | 1,132,462,080 bytes per second
| |
12:35 | bootstrap | 9,059,696,640 bits per second
| |
12:35 | bootstrap | but yeah, at 24 fps i need to multiply by 24/60.
| |
12:36 | bootstrap | so i agree.
| |
12:36 | bootstrap | so those higher rates on the webpage (full rez at 75FPS) are not what you are really shooting for.
| |
12:37 | bootstrap | woops! my mistake. I'm looking at the sensor website! :-o
| |
12:38 | se6astian | well first priority is making sure it works at standard framerates (24,25,48, etc.) but with enough overhead planned into the available bandwidths to go higher
| |
12:38 | bootstrap | Okay, so your plan is to stream out the images onto an external solid-state hard disk drive.
| |
12:38 | se6astian | cmosis recently announced a new version of the sensor that doubles the frame rates once more
| |
12:39 | shinokubo | left the channel | |
12:39 | bootstrap | so 150fps @ 12-bits @ 4Kx3K ?
| |
12:39 | bootstrap | I guess the question I'm asking myself is... can we practically design for those higher frame rates?
| |
12:39 | se6astian | but 4096x3072 x 10 bit x 300 FPS = 36Gbit/s which is rather insane :)
| |
12:39 | bootstrap | yes, that's quite a bit of data.
| |
12:40 | bootstrap | i have a super-simple lossless compression scheme i developed for my camera, but it only buys us about 20%.
| |
12:40 | se6astian | we might be able to sustain that for a very short duration with a big amount of RAM
| |
12:40 | se6astian | like for 1 second
| |
12:40 | bootstrap | So it isn't worth the effort unless you are just short of "making it".
| |
12:40 | se6astian | then it takes 10 seconds to read out the RAM
| |
12:40 | se6astian | like how all current high speed cameras operate
| |
12:40 | bootstrap | Are you planning to have RAM ?
| |
12:40 | se6astian | sure
| |
12:41 | se6astian | DDR3
| |
12:41 | se6astian | light compression could make sense
| |
12:41 | bootstrap | I was always thinking of supporting 8 flash memory devices in parallel to achieve that bandwidth (10Gbps).
| |
12:41 | se6astian | the simpler the method the better tbh :)
| |
12:41 | bootstrap | it was designed for inside the FPGA.
| |
12:41 | bootstrap | it really is trivial.
| |
12:41 | bootstrap | it is especially designed for 12-bit pixel depth.
| |
12:42 | se6astian | Bertl once had the idea to build our own solid state memory
| |
12:42 | bootstrap | converts most 12-bit pixels into 8-bits, and the rare "failure to compress" pixel into 16-bits.
| |
12:42 | bootstrap | not seriously... you don't mean the chips themselves, do you?
| |
12:42 | se6astian | which is not as difficult as one could think as the memory blocks from Samsung, etc. are all the same in current SSDs, usb sticks, etc
| |
12:42 | se6astian | but it would still be a huge project on its own
| |
12:43 | bootstrap | oh, right. we could buy the die and bond them.
| |
12:43 | se6astian | even though we could achieve insane write speeds that way
| |
12:43 | bootstrap | well, that is not as impractical as you may think.
| |
12:43 | bootstrap | not trivial either, but... doable.
| |
12:43 | bootstrap | depends on how much value gets attached to supporting high frame rates.
| |
12:44 | se6astian | its not a VERY high priority at the moment
| |
12:44 | bootstrap | I understand. But it is worth looking forward in order to avoid having to develop a dozen completely separate implementations.
| |
12:44 | se6astian | in our current website survey only 7% say high speed is their primary application
| |
12:45 | se6astian | which does not mean nobody else would shoot high speed of course
| |
12:45 | bootstrap | I understand. But if you can achieve it in a practical way, people would love it.
| |
12:45 | se6astian | exactly
| |
12:45 | se6astian | we will try to do it
| |
12:46 | bootstrap | So, where would you plan to put the solid-state hard disk?
| |
12:46 | se6astian | only if we hit problems that will make everything drastically more expensive or more complicated we would go for the lower speed/bandwidth options
| |
12:46 | bootstrap | Or "that's not your problem"?
| |
12:46 | se6astian | how do you mean "where"?
| |
12:47 | bootstrap | I mean physically where? I guess the customer buys this piece, huh?
| |
12:47 | bootstrap | I mean, it is just a standard product they buy in addition to the camera.
| |
12:47 | se6astian | about your 12 bit -> 8 bit compression, where/how do you store the metadata which pixels get 8 and which 16 bits?
| |
12:47 | se6astian | yes, any SSD should work
| |
12:47 | se6astian | and we just have a standard SATA connector and a drive bay
| |
12:48 | bootstrap | if the first 4-bits are 0s, then the next 12-bits are the actual value, otherwise the 8-bit value is a delta (difference in intensity from last pixel of same color, assuming bayer RGBG).
| |
12:48 | se6astian | I see
| |
12:48 | bootstrap | very simple.
| |
12:48 | se6astian | indeed!
| |
12:49 | bootstrap | but 90-some percent of pixels are within +/- 120 of the previous pixel of the same color, which makes the delta within the 240 values available.
| |
12:49 | guesst | how you encode difference of less than 8 ?
| |
12:49 | bootstrap | only at edge transitions do the pixel differences get larger, in which case it takes 16-bits to send a 12-bit value.
| |
12:50 | bootstrap | i don't bother with that.
| |
12:50 | bootstrap | it is just a super-simple technique easy to implement in FPGA.
| |
12:50 | bootstrap | and in my case, that extra 20% made all the difference (cuz i was squeezing 2592x1944 12-bits through 1Ge).
| |
12:51 | guesst | but diff less than 16 or 8 is 0000xxxx... stream corruption?
| |
12:53 | bootstrap | no. if the most-significant bits are 0001, 0010, 0011, 0100, 0101, 0110, 0111, 1000, 1001, 1010, 1011, 1100, 1101, 1110, 1111... then those are the most-significant 4-bits of the delta value (followed by 4 more bits).
| |
12:53 | bootstrap | in other words, all delta values from -120 to +120 are sent as 8-bit values... but must be offset upwards by 16.
| |
12:53 | guesst | ok
| |
12:54 | bootstrap | in other words, to encode -120 the value 00010000 is sent : -119 the value 00010001 is sent, and so forth up to +120 == 11111111.
| |
12:55 | bootstrap | however, when the delta is more than +/- 120, then the first 4 bits are 0000 (which only happens when 16-bits are sent), and the next 12-bits are the absolute value of the pixel intensity.
| |
12:55 | bootstrap | actually, i think i settled for -119 to +119 in the end, to have a couple extra "special code" values.
| |
12:56 | bootstrap | or maybe it was even -118 to +118... i'd have to go back and look.
| |
12:56 | bootstrap | or -119 to +118... 2's complement is strange, sometimes one more value in the negative direction.
| |
12:57 | guesst | have you done any research whether the 8/12 bit is the best case? :)
| |
12:57 | bootstrap | no. and obviously it depends on the sensor! hahaha.
| |
12:57 | bootstrap | if you're sensor only generates 10-bits, or generates 16-bits... the whole issue needs to be re-factored.
| |
12:58 | bootstrap | in my case, i was lazy, didn't want to deal with non-bytes, so i just implemented this one scheme.
| |
12:58 | guesst | understandable
| |
12:58 | guesst | especially if you need to decode in sw
| |
12:58 | bootstrap | and the 20% was sufficient to save my ass.
| |
12:59 | bootstrap | as you can see, the encode and decode is ultra-trivial.
| |
12:59 | bootstrap | as i recall, to decode takes about 4 assembly-language instructions!
| |
12:59 | guesst | until you shoot trees and grass... and it will actually enlarge the frames :)
| |
12:59 | bootstrap | well, a few more, but it only executes 4 as i recall.
| |
12:59 | bootstrap | in that case i just send the raw data.
| |
13:00 | guesst | have you a per line buffer to choose which to send?
| |
13:00 | bootstrap | i collect both in the FPGA, and at the end of the horizontal row of pixels, i just send the shorter one.
| |
13:00 | guesst | same thinking :)
| |
13:00 | bootstrap | the compression is done first thing as the pixel value enters the FPGA from the sensor.
| |
13:00 | se6astian | sounds good
| |
13:00 | bootstrap | actually, i almost forgot!
| |
13:01 | bootstrap | hahaha. i have a terrible memory.
| |
13:01 | se6astian | so back to your original question: we do need help with electronics, hardware, etc. :)
| |
13:01 | bootstrap | one reason i kept a couple extra values was to toggle back and forth between uncompressed and compressed... in the middle of the scan line.
| |
13:01 | bootstrap | so a value 11111111 toggles back and forth.
| |
13:02 | se6astian | Bertl is currently working on FPN measurements, correction software
| |
13:02 | bootstrap | so if you go from "sky" to "tree leaves", it might switch.
| |
13:02 | se6astian | when he will be here in the next 2-3 hours maybe
| |
13:02 | bootstrap | i saw the preliminary fixed pattern noise tests. the sensor has quite a bit of that.
| |
13:02 | se6astian | please talk to him as he knows best what a first/next step could be
| |
13:03 | se6astian | yes due to the global shutter nature its a lot more than what we knew from rolling shutter ones
| |
13:03 | se6astian | as each coloumn has its own ADC
| |
13:03 | bootstrap | well, i can maybe help with circuit design and PCB layout if you guys need help. plus, of course, brainstorm all these many tradeoffs!
| |
13:03 | se6astian | sounds great
| |
13:03 | se6astian | can you PM me your email address
| |
13:03 | bootstrap | yes. BTW, you'll find that you get better results if you measure and store the fixed pattern noise at 16 different intensity levels.
| |
13:04 | guesst | se6astian: even rolling shutter got each column an ADC... you need that once you pass some MPx/s
| |
13:04 | bootstrap | *email address removed* or *email address removed* or *email address removed*
| |
13:04 | bootstrap | the first continues to work if my email server goes down.
| |
13:05 | bootstrap | BTW, you might as well assume 8-layer PCBs. the savings for less layers is so little.
| |
13:07 | bootstrap | i haven't looked at the information i've downloaded about the sensor yet, but what pins they present will make a big difference in what electronics is best.
| |
13:08 | bootstrap | okay, what i downloaded doesn't say much, except it appears 32 LVDS outputs.
| |
13:08 | bootstrap | oh wait... maybe that's 64 LVDS outputs.
| |
13:08 | se6astian | thanks
| |
13:09 | bootstrap | okay, let me see here. 10Gbps / 64 outputs... hey, that's not bad.
| |
13:09 | se6astian | 64 LVDS
| |
13:09 | se6astian | yes :)
| |
13:09 | bootstrap | it sure would be wonderful (cost wise) to be able to adopt a cheapo FPGA like cyclone5 or something.
| |
13:10 | se6astian | where do you produce your 8 layer PCBs
| |
13:10 | bootstrap | man, the faster (and fancier) FPGAs are totally absurd in price.
| |
13:10 | bootstrap | let me check where i got them done.
| |
13:10 | se6astian | we used services from OSHpark which does group panelization of small orders for a very good price
| |
13:10 | se6astian | but only up to 4 layers
| |
13:10 | se6astian | its easy of course once you go high volume...
| |
13:11 | bootstrap | i had like 5 voltages (5 power planes)... but only 4 power-plane layers.
| |
13:11 | bootstrap | i bought low volume, for prototypes. let me check.
| |
13:13 | bootstrap | www.pcbinternational.com
| |
13:13 | bootstrap | i'm trying to find price, but it was something like $30 each.
| |
13:13 | bootstrap | prototype volumes.
| |
13:13 | se6astian | not bad
| |
13:14 | se6astian | we payed $10 per square inch for 3 boards
| |
13:14 | se6astian | so actually 3.3$ per square inch
| |
13:15 | bootstrap | well, i bought 25 boards, so maybe that helped.
| |
13:16 | bootstrap | price is much higher at 1 to 5 boards.
| |
13:17 | bootstrap | example: http://www.iceapps.com/ice_vision_pcb_6704.jpg : http://www.iceapps.com/ice_vision_pcb_6706.jpg : http://www.iceapps.com/ice_vision_pcb_6721.jpg
| |
13:17 | bootstrap | you won't understand those PCBs unless I explain.
| |
13:17 | bootstrap | the big PCB is "controller" for 1 to 4 cameras. the small PCB is the sensor PCB for one camera.
| |
13:18 | bootstrap | so for many industrial applications, only one gigabit ethernet cable could handle 4 cameras.
| |
13:18 | bootstrap | like... pointing in 4 directions (a very common requirement).
| |
13:19 | guesst | no memory there?
| |
13:19 | se6astian | ah ok, so not that cheap for prototyping
| |
13:19 | se6astian | anyway
| |
13:19 | se6astian | gotta go to the supermarket
| |
13:19 | se6astian | anything you want to ask before I leave?
| |
13:19 | se6astian | otherwise we can just continue when I am back
| |
13:19 | se6astian | should not take that long
| |
13:20 | bootstrap | okay, if i am still up, otherwise maybe i can catch you tomorrow or later.
| |
13:20 | bootstrap | who are the most important people for me to talk with?
| |
13:20 | se6astian | Bertl
| |
13:20 | se6astian | he lives more in the US timezone :)
| |
13:20 | bootstrap | he must be the hardware guru.
| |
13:21 | se6astian | even thogh he is located like one hour away from where I live :)
| |
13:21 | se6astian | yes
| |
13:21 | bootstrap | i'm usually on phobos time! hahaha.
| |
13:21 | bootstrap | okay, hopefully i'll catch you later.
| |
13:21 | bootstrap | about the memory, those PCBs have a little static ram memory, but not enough to do sophisticated image processing.
| |
13:22 | bootstrap | i think it was something like 8MB... let me check.
| |
13:22 | se6astian | ok see you later maybe then
| |
13:24 | bootstrap | great
| |
13:24 | bootstrap | oh, actually the ram is only 1MB.
| |
13:24 | bootstrap | yeah, that's right. not even one frame.
| |
13:25 | bootstrap | also, there are a fair number of components on that large PCB that aren't normally stuffed, including those RAMs.
| |
13:25 | bootstrap | and like 10 binary counters to twiddle the addresses of those RAms.
| |
13:25 | bootstrap | and like 10 binary counters to twiddle the addresses of those RAMs.
| |
13:38 | danhanes | joined the channel | |
13:45 | bootstrap | @se6astian: i found what i paid for the PCBs. 25 of the 150mm square controller PCB == $1200 : 50 of the 75mm square sensor PCB == $900 : 8-layer PCBs with 4-mil traces, 0201s, 0.35mm contact pitch, BGAs, some filled holes for BGAs, gold plated (all conductive surfaces).
| |
14:02 | bootstrap | so $48 each for 150mm square PCBs and $18 each for 75mm square PCBs.
| |
14:04 | bootstrap | not sure whether this will matter, but i have a 4-trace 5 giga-sample-per-second oscilloscope that might be useful for circuit verification and debugging.
| |
14:06 | bootstrap | Lately I've been tempted to buy a pick-and-place machine (plus automatic stencil printer and reflow oven), which would let us assemble the PCBs too (in modest but not large volumes... maybe 1000s/year).
| |
14:28 | philippej | joined the channel | |
14:29 | se6astian | back
| |
14:30 | bootstrap | hi. i've been looking through some FPGA specs at the LVDS specs. can i assume the sensor puts out an LVDS clock as well as the data channels?
| |
14:30 | se6astian | yes
| |
14:31 | bootstrap | good. i'm curious to figure out whether we can handle this sensor with a cheap FPGA (well, relatively cheap).
| |
14:32 | bootstrap | BTW, I typed in a little information for you above, about what i actually paid for those PCBs.\
| |
14:32 | bootstrap | BTW, I typed in a little information for you above, about what i actually paid for those PCBs.
| |
14:33 | bootstrap | off hand, looks good.
| |
14:33 | se6astian | thanks, yes
| |
14:33 | se6astian | well its easy to get good prices one you buy a decent size/amount
| |
14:34 | se6astian | I was hoping you had found a way to get cheap prototypes as well for 8 layers
| |
14:34 | se6astian | *once
| |
14:34 | bootstrap | well, i thought those were pretty cheap. oh well.
| |
14:34 | se6astian | yes its a good price but you also bought 25 pieces
| |
14:34 | bootstrap | yup. how many prototypes do you envision?
| |
14:35 | bootstrap | i guess that depends on how confident you are the first cut won't require revisions! hahaha.
| |
14:35 | bootstrap | i have a pretty amazing track record for needed zero revisions.
| |
14:35 | bootstrap | but i hate to say that, because Murphy might be listening.
| |
14:36 | shinokubo | joined the channel | |
14:36 | bootstrap | okay, looks like 320 Mbps is sufficient to handle 150 FPS.
| |
14:36 | bootstrap | on each of the 64 LVDS channels.
| |
14:36 | bootstrap | is that what the new faster sensor supports (at 12-bits per pixel)?
| |
14:36 | se6astian | we have two assembled prototypes of the 3 pcbs we ordered
| |
14:37 | shinokubo | left the channel | |
14:37 | se6astian | since the image sensor is rather expensive we cant easily make more
| |
14:37 | bootstrap | i understand. i'm thinking about the next ones.
| |
14:37 | bootstrap | yeah. is the sensor about $1500 ???
| |
14:37 | se6astian | yes new sensor does 300mhz on each LVDS line instead of 150hz
| |
14:37 | se6astian | 1800$
| |
14:38 | se6astian | *150Mhz
| |
14:38 | bootstrap | $1800 is the new [faster] one, or the old one?
| |
14:40 | bootstrap | and the signals are LVDS or mini-LVDS ?
| |
14:41 | bootstrap | oh, i guess LVDS.
| |
14:45 | se6astian | lvds
| |
14:45 | se6astian | price is the same for both versions
| |
14:46 | bootstrap | the new faster one is no more expensive? hmmmm.
| |
14:46 | bootstrap | let me think... which to buy? hahaha.
| |
14:47 | bootstrap | well, i see conflicting information, but looks like maybe the cheap FPGA can handle 320 Mbps LVDS.
| |
14:47 | bootstrap | one place says up to 875 Mbps LVDS.
| |
14:48 | bootstrap | ah, that really sucks! # supported (on cheaper parts) == 60. the more expensive parts 120. talk about rip-off time!
| |
14:51 | bootstrap | question: how many of these do you GUESS you'll sell in the first year?
| |
14:52 | se6astian | we will target a minimum amount of 100-150 presold cameras in our crowdfunding campaign
| |
14:54 | bootstrap | is the sensor cheaper in 100 or 1000 unit quantities?
| |
14:55 | se6astian | yes, a little
| |
14:55 | se6astian | but nothing like the price drops when ordering higher volume PCBs
| |
14:56 | philippej | left the channel | |
14:56 | bootstrap | but maybe 25% cheaper?
| |
14:57 | bootstrap | also, has anyone considered mounting a peltier cooler to the chip? --- whoops, no can do with pin-grid array package.
| |
14:57 | bootstrap | would be nice to suck some heat out of that baby to reduce noise.
| |
14:57 | se6astian | yes its something in the range of 20% max
| |
14:58 | bootstrap | okay, so sensor is $1500 even in pretty big quantities.
| |
14:58 | se6astian | yes, thats the plan with a peltier element, heat grill and we are currently evaluating a piezoelectric "microblower" to move the air through the heatsink
| |
14:58 | se6astian | without making noise or vibrations
| |
14:59 | bootstrap | yeah, but too bad we can't mount it right on the chip. doesn't work with pin grid packages though.
| |
14:59 | troy_s | joined the channel | |
14:59 | rexbron | joined the channel | |
14:59 | rexbron | left the channel | |
14:59 | rexbron | joined the channel | |
14:59 | bootstrap | but maybe we could try to pull heat right out of the pins (that go into ground and power-plane layers).
| |
14:59 | bootstrap | hmmmm... need to think about that one. tricky.
| |
15:00 | bootstrap | looks like the cheapest FPGA that can handle 64 channels of LVDS costs about $160.
| |
15:01 | se6astian | we considered cutting a hole into the PCB under the sensor and attaching a heat conductor right through that hole onto the back surface of the sensor
| |
15:01 | se6astian | but its all a concept for now
| |
15:01 | rexbron | Hey!
| |
15:01 | se6astian | the zynq we target will be around 200-250$ so thats almost in the same ballpark already
| |
15:01 | se6astian | hi rexbron
| |
15:02 | bootstrap | does the sensor maker advise about cooling, or just ignore the issue?
| |
15:03 | bootstrap | in their spec sheet, all i can see is the top side of the chip. not sure what is that metal-frame like thing with two slots in it.
| |
15:03 | bootstrap | is that a heat-sink?
| |
15:03 | bootstrap | something we could suck heat out of?
| |
15:03 | rexbron | I'm working on a kids show, we're scheduled to shoot until February 2015. All of the sets are built to kid scale. A small camera head definitely has its place here.
| |
15:05 | rexbron | We're looking at purchasing a 2nd scarlet for the Movi but if I ever get my hands on a prototype, I'd like to try rocking it on the right
| |
15:05 | rexbron | Rig rather
| |
15:06 | bootstrap | @se6astian: wait --- i just read your comment about attaching the cooler. isn't this a pin-grid package? if so, how can that work with 237 pins sticking out the back of the package (that need to go in through-holes or a socket).
| |
15:08 | se6astian | cmosis has no advice regarding cooling
| |
15:08 | se6astian | rexbron, that would be great!
| |
15:08 | se6astian | I hope we will have something useable until then
| |
15:09 | se6astian | bootstrap: a socket
| |
15:09 | bootstrap | why socket rather than direct solder into PCB ?
| |
15:09 | bootstrap | i guess "the price" is your answer.
| |
15:10 | rexbron | Repair ability
| |
15:11 | bootstrap | hmmmm... that will make mechanical issues (alignment of sensor face perpendicular to optical axis) more difficult, as well as cooling.
| |
15:12 | rexbron | The work flow for the show is kinda funny. We're shooting red but recording the hdsdi feel to atomos samurai prores, except for vfx and high speed shots, then it's internal r3d
| |
15:13 | rexbron | Well, the lens mount and flange is more likely culprit for perpendicular issues
| |
15:13 | bootstrap | i hope not! that's all cnc machined stuff, which should be easily within .001" == 0.025mm
| |
15:14 | rexbron | Lol!
| |
15:14 | rexbron | It's the person who installs it
| |
15:14 | rexbron | For example, I don't own a torque wrench
| |
15:14 | bootstrap | i suppose it might be possible to try to ?spring-load? the glass face against a machined surface.
| |
15:15 | se6astian | repairability and I guess you cannot put the sensor into a reflow oven :)
| |
15:15 | bootstrap | but that's a bit gross, mechanically.
| |
15:15 | bootstrap | did they say that? no reflow oven for the sensor?
| |
15:15 | bootstrap | or you are just too afraid on general [financial] principles?
| |
15:15 | rexbron | Also, when I shim a lens mount for flange focal distance, the tolerance needs to be within 0.01mm for acurate focus marks
| |
15:16 | rexbron | You won't notice with stills lenses though, only cinema lenses with very fine focus adjustment
| |
15:16 | bootstrap | well, good machining should be good to 0.01mm.
| |
15:17 | se6astian | I would need to read up in the datasheet but I am pretty sure it wont be healthy for the sensor
| |
15:17 | bootstrap | but i don't know how perfect the sensor package is, so it might be necessary to provide for a little adjustment.
| |
15:18 | bootstrap | actually, with a pin-grid package, it can be soldered manually with a hot-air soldering pencil.
| |
15:18 | bootstrap | the pin spacing could be pretty large in such a big package... maybe even 2mm apart (though problem not quite).
| |
15:18 | rexbron | See how the epic does it
| |
15:18 | bootstrap | i don't know how. do you?
| |
15:19 | rexbron | There is a screw that adjusts forward and backwards
| |
15:19 | rexbron | Not sure ho
| |
15:20 | rexbron | how arri does it
| |
15:20 | bootstrap | one problem with adjustability (especially on the lens side), is that some people will attach HEAVY lenses, and perhaps bend and shift things.
| |
15:20 | bootstrap | i'd rather have any adjustability be on everything behind the sensor (light weight and not tweaked by lens weights).
| |
15:21 | bootstrap | but i don't even like it there.
| |
15:21 | bootstrap | frankly.
| |
15:21 | bootstrap | but might be necessary... boo hoo.
| |
15:21 | rexbron | Yes but you need to design for that. Cooke s5 is as heavy as they make primes, beyond which you need lens support
| |
15:22 | bootstrap | if you look at the case design, you'll see there are limits.
| |
15:23 | rexbron | http://www.cookeoptics.com/l/fiveilens.html
| |
15:24 | bootstrap | those lenses look within reason.
| |
15:24 | rexbron | It needs to be able to support the lenses cinematographers use or it will never have any impact with them
| |
15:24 | bootstrap | well... 5kg, that's a bit on the edge.
| |
15:27 | bootstrap | @se6astian: looking at the side view of the camera (showing modules)... is this really what you're hoping to do (support and endless stream of 20mm ~ 30mm thick add-on modules)?
| |
15:28 | bootstrap | @rexbron: Actually, the photo I'm looking at has a bracket that appears to be able to provide support for the front end of lenses.
| |
15:28 | bootstrap | that is, assuming i'm understanding the photo.
| |
15:29 | se6astian | bootstrap, we are working on version two of the open modules concept
| |
15:29 | se6astian | that uses slots instead o stacks of modules
| |
15:29 | se6astian | I will hopefully have new concept images ready soon
| |
15:30 | bootstrap | so like a long motherboard along the bottom, and push PCBs in from the top.
| |
15:30 | bootstrap | or the motherboard gets extended each time?
| |
15:31 | bootstrap | this problem always drove me crazy too! not an easy one to do optimally for every situation!
| |
15:34 | bootstrap | is the back of each module closed? or there are some connectors (that you can push rubber/plastic covers over if the last module)?
| |
15:34 | bootstrap | i guess they're open, and you just have an end cap.
| |
15:35 | bootstrap | BTW, are there a lot of people who want the width kept small so they can put two side-by-side for 3D work?
| |
15:36 | bootstrap | it is such a royal pain in the PCB to try to make the PCBs (and case) so damn small.
| |
15:37 | se6astian | currently we think about different options (different number of slots) for one long motherboard
| |
15:38 | se6astian | the lens mount we use is already so big that a side-by-side 3d rig makes no sense even if you put them as close together as possible
| |
15:39 | bootstrap | it was my decision to blow off side-by-side for 3D... unless people could accept a wider separation than eyeball distance.
| |
15:40 | bootstrap | of course some impossible-to-please types will always complain.
| |
15:42 | bootstrap | another problem with 3D is that lots of lenses have stuff sticking out (and shade cups) that will hit each other.
| |
15:43 | bootstrap | what diameter is that square body now? i'd love to bloat it out to 100mm diameter myself.
| |
15:44 | bootstrap | oh, it IS 100mm already! yes!
| |
15:44 | bootstrap | i vote for 100mm x 120mm... or larger! hahaha.
| |
15:45 | bootstrap | i made my cases 80mm x 80mm x 25mm... and later decided i was an idiot.
| |
15:45 | Bertl | morning everyone!
| |
15:46 | bootstrap | hello Bertl. they've been telling me i need to talk with you.
| |
15:46 | se6astian | hi Bertl
| |
15:47 | bootstrap | @Bertl: not sure whether you can read the past hour or so of conversation or not, but we've been talking design.
| |
15:50 | bootstrap | I'm an electronics designer, designed cameras before, and worked with big CCDs in astronomy (big telescope) context for about 25 years (off and on). I might be of some use... maybe.
| |
15:51 | danhanes | morning Bertl
| |
16:06 | Bertl | bootstrap: yep, I've read up on the entire day now
| |
16:06 | mars_ | bootstrap: i guess, your knowledge is very welcome here
| |
16:06 | bootstrap | sorry to put you through that!
| |
16:06 | Bertl | not so bad, was kind of fun :)
| |
16:06 | mars_ | and good morning Bertl ;)
| |
16:07 | bootstrap | well, do you have anything you want to tell me to straight me out?
| |
16:07 | Bertl | yes, as mars_ pointed out, help and know-how is always welcome
| |
16:07 | bootstrap | don't worry about shooting down my thoughts. doesn't bother me.
| |
16:08 | Bertl | bootstrap: as you already figured, we created a sensor frontend which connects via FMC to the zedboard
| |
16:08 | bootstrap | which makes you able to run your preliminary tests.
| |
16:08 | Bertl | this FE uses 32 of the 64 sensor channels to transfer data at 300MHz (150MHz DDR)
| |
16:09 | Bertl | yes, this is our (now boxed) alpha prototype
| |
16:09 | rexbron | bootstrap: link, because I've never used lens support with cooke s5s
| |
16:10 | rexbron | the PL mount can support 5.1Kg without a lens support but the rest of the body needs to as well :)
| |
16:11 | bootstrap | this is one issue i don't know a lot about. what is required? 5kg, 10kg, ??? (without support).
| |
16:11 | Bertl | bootstrap: so, after folks already tried (and mostly succeeded) to answer most questions, what are the open ones or what details do interest you?
| |
16:11 | bootstrap | well, i've been self-employed developing [mostly electronics, computer, software, optics] products since i got out of school, so this is right down my alley, and very interesting to me.
| |
16:12 | rexbron | bootstrap: also, side by side 3d will always be inferior to mirrors
| |
16:12 | bootstrap | i've been thinking about something along these lines, but when i started my last camera this sensor didn't exist.
| |
16:12 | bootstrap | yeah, i'd love to totally forget 3D in the design of this sucker (if it was up to me).
| |
16:12 | rexbron | agreed :)
| |
16:13 | bootstrap | okay, here is one point of view: big case diameter is fine with me. even larger than 100mm x 120mm would be okay with me.
| |
16:14 | bootstrap | i think trying to make a 4K camera fit in a shirt pocket is... mixing two different things.
| |
16:14 | rexbron | bootstrap: as a cinematographer and camera assistant, I like cameras that are designed for the human body
| |
16:14 | bootstrap | meaning... for holding while filming?
| |
16:15 | bootstrap | did i say filming? i mean shooting.
| |
16:15 | bootstrap | or whatever is the term now.
| |
16:15 | bootstrap | i saw a pretty sweet "steadycam" type produce on a crowdsourcing site a few days ago.
| |
16:16 | bootstrap | makes the size of the camera not so important.
| |
16:16 | bootstrap | unless it was to get out of hand.
| |
16:16 | rexbron | bootstrap: stabilizers are their own beast
| |
16:16 | bootstrap | yup. but anything worth shooting at 4K is worth shooting well.
| |
16:16 | rexbron | and god knows the number of times I work with directors who want "That handheld feel"
| |
16:17 | rexbron | bootstrap: lol, it will still be mostly cats and kids
| |
16:17 | bootstrap | so either well mounted or with some kind of "steadycam" gizmo. or you disagree?
| |
16:17 | rexbron | bootstrap: I hate cameras I have to fight
| |
16:17 | rexbron | I have to fight the epic constantly
| |
16:18 | bootstrap | well, you know more about that than me. so tell us how the diameter and depth of the case impact this for you.
| |
16:19 | rexbron | bootstrap: well, tbh I've been pushing for something like an Aaton form factor and so have any well known DPs that have been attracted to the project but it is like at odds with the other design goals of the project
| |
16:19 | rexbron | ergonomics cost because the human body doesen't mesh with right angles and boxes well
| |
16:19 | rexbron | but they are easier to machine :{
| |
16:19 | rexbron | :P
| |
16:19 | Bertl | that's what plastic surgery is for! :)
| |
16:20 | rexbron | lol!
| |
16:20 | bootstrap | hahaha... yeah, i guess we're too much like hardcore engineers to design something like that!
| |
16:20 | rexbron | well it's definately a cost thing
| |
16:20 | rexbron | having irrgular pcb shapes pushes up the cost
| |
16:20 | bootstrap | but it opens a whole extra market for people making mechanical wrappers for the camera!
| |
16:21 | rexbron | bootstrap: and I hate it
| |
16:21 | bootstrap | you hate what? rectangles?
| |
16:21 | rexbron | Having to rig out cameras
| |
16:21 | rexbron | because things don't always mesh well together
| |
16:21 | Bertl | rexbron: do you have a problem with an ergonomic housing which 'contains' the rectangular camera?
| |
16:21 | rexbron | I'd rather have a small camera, that stays a small camera
| |
16:22 | bootstrap | yeah, and rig out something else.
| |
16:22 | Bertl | or to be precise, the rectangular parts of the camera
| |
16:22 | rexbron | Bertl: depends on the design of the container
| |
16:22 | bootstrap | mount the camera ON something, then rig THAT out. or maybe that doesn't work?
| |
16:22 | rexbron | bootstrap: you want your CoG as low as possible for handheld
| |
16:22 | Bertl | rexbron: well, whatever you like, as long as it fits around the camera parts :)
| |
16:23 | bootstrap | i don't have a wide enough variety of experience with modern high-end video equipment.
| |
16:23 | rexbron | so on never works as well as wrapped around a shoulder
| |
16:23 | rexbron | basically, the evf and top handel should be at or around eyelevel and the rest below that
| |
16:24 | rexbron | http://www.hurlbutvisuals.com/blog/wp-content/uploads/2012/10/arri_235_on_shoulder.jpg
| |
16:24 | Bertl | the reason I ask is, that I do not see a problem with having both, a camera block and an ergonomic shoulder fitting whatever
| |
16:24 | bootstrap | i'm probably full of it, but i always imagined i'd like to totally separate the camera and ME (and the viewfinder/display, and the controls).
| |
16:24 | rexbron | Bertl: neither do I, provided the ergo rig works as well as it should :)
| |
16:25 | rexbron | bootstrap: how so? The art of camera operating to to become one with the camera
| |
16:25 | rexbron | to make it an extention of your body and effortlessly controlled
| |
16:25 | Bertl | glass: please record this :)
| |
16:25 | bootstrap | in fact, the more i've thought about video, the more i want to mount the camera on a servo driven system that records the exact orientation (and moves in programmed ways).
| |
16:26 | rexbron | bootstrap: there are pleanty of people doing moco
| |
16:26 | bootstrap | moco cameras? or moco accessories?
| |
16:26 | rexbron | from small diy timelapse dollies to the new technocrane
| |
16:27 | rexbron | I mean moco as programable moves, not motion capture of human data points
| |
16:27 | bootstrap | frankly, i think within a couple years, people will attach cameras to drone quad-copters and fly them around.
| |
16:27 | rexbron | bootstrap: lol, it already is done
| |
16:27 | bootstrap | they do now, but i mean commonly.
| |
16:27 | bootstrap | yes.
| |
16:28 | rexbron | bootstrap: those have a time and a place
| |
16:28 | rexbron | Cinematography is storytelling
| |
16:28 | bootstrap | yes.
| |
16:28 | rexbron | I don't belive that any camera is "THE ONE TRUE GOD" camera, despite what manufaturers will try and tell you
| |
16:28 | bootstrap | and to me, to NOT constrain the camera to any one set of limitations is the best move.
| |
16:28 | bootstrap | which means modular.
| |
16:29 | rexbron | bootstrap: modular is a constraint
| |
16:29 | bootstrap | which means not building 1000 gizmos into the camera.
| |
16:29 | rexbron | modular constrains the assistant
| |
16:29 | bootstrap | at least, not 1000 mechanically controlled gizmos.
| |
16:29 | bootstrap | maybe. but does it have to?
| |
16:30 | rexbron | modular is a pain in my ass when I order a camera body and it shows up with the wrong lens mount or missing SDI because the owner couldn't afford gas for his new Lambo
| |
16:30 | bootstrap | what I mean is, if the cinematographer and assistant can fiddle anything they want without touching the camera, isn't that a lot easier?
| |
16:30 | rexbron | bootstrap: you've never used a redmote
| |
16:30 | bootstrap | true.
| |
16:30 | rexbron | fuckign thing loses signal _ON THE DAMN CAMERA_
| |
16:30 | rexbron | because we have to put the body in a metal cage
| |
16:31 | rexbron | to attach all the gear
| |
16:31 | rexbron | that that should be on the inside!
| |
16:31 | bootstrap | i guess part of what i am asking is... which gear MUST be physically on the camera, and which gear can be "elsewhere"?
| |
16:31 | bootstrap | obviously the lens has to be on the camera, but... otherwise?
| |
16:31 | rexbron | all of it
| |
16:32 | rexbron | try working as an AC
| |
16:32 | bootstrap | no, CAN BE, but doesn't have to be.
| |
16:32 | rexbron | :)
| |
16:32 | rexbron | true, I'd say SHOULD be
| |
16:32 | rexbron | less points of failure
| |
16:32 | bootstrap | well... that means the cinematographer and his assistants must fly around on the hovercraft drone too.
| |
16:33 | Bertl | drones for everyone!
| |
16:33 | rexbron | bootstrap: if you are designing a camera for that application, sure
| |
16:33 | rexbron | the problems arrise when _I_ get your drone camera
| |
16:33 | rexbron | and the DP and director want to shoot handheld
| |
16:33 | bootstrap | well, i better turn this over to Bertl, cuz i have zero "authority" to decide what this camera is designed for!
| |
16:33 | rexbron | or in studio mode
| |
16:33 | rexbron | and it's not designed for that
| |
16:33 | bootstrap | @Bertl: please enlighten us!
| |
16:34 | rexbron | it designed to optimise one aspect, size
| |
16:34 | bootstrap | i'm not sure about that.
| |
16:35 | Bertl | well, while it seems to be simple and trivial, it also seems to be hard to communicate
| |
16:35 | Bertl | the main design focus is on 'openness'
| |
16:36 | Bertl | so while most cameras focus on this or that but never on being open (interface or design wise)
| |
16:36 | bootstrap | does that mean lots of people will try to build this hardware?
| |
16:36 | Bertl | we try to focus on that, which means we can do a lot of things others simply can't
| |
16:36 | Bertl | (sec door bell)
| |
16:36 | bootstrap | i really don't know your plans yet! i only just learned about this project a few hours ago.
| |
16:40 | bootstrap | but it just occurred to me that it seems IMPOSSIBLE for this project to create a fully/massively integrated camera with every little feature and convenience that multi-billion dollar manufacturers have developed in the past decades!
| |
16:40 | bootstrap | not only would the camera have to include lots more hardware and controls, but... all the integration (making things work together conveniently) would need to be figured out.
| |
16:41 | se6astian | time to cook dinner, bbs
| |
16:41 | bootstrap | okay, talk later (probably another day, getting towards my bedtime).
| |
16:42 | rexbron | bootstrap: do you know of the Yolk Y2 concept camera
| |
16:42 | bootstrap | no.
| |
16:42 | bootstrap | but now i'm looking at some photos of it.
| |
16:43 | bootstrap | BTW, does LCD viewfinder work for you, or you need optical?
| |
16:45 | bootstrap | what is that? 2 hard drives in the shoulder piece? hahaha!
| |
16:47 | bootstrap | well, i think someone (you maybe) could make a wrapper for the straightforward camera that makes it look like that (or even better, if you can figure out the best configuration).
| |
16:48 | se6astian | lcd or evf is fine
| |
16:49 | bootstrap | that would certainly be one way to handle the "open" aspect... by making a "component" that others can wrap other stuff around to make special purpose implementations.
| |
16:49 | bootstrap | what is evf?
| |
16:49 | Bertl | back
| |
16:50 | bootstrap | great. read a bit then kick butt (especially mine).
| |
16:50 | Bertl | so the idea is not so much to cover all possible variations or features
| |
16:50 | Bertl | the idea is to allow for all of those features to happen eventually
| |
16:51 | Bertl | let me try to ilustrate that on a simple example:
| |
16:51 | bootstrap | make the absolutely required functionality... than add bits and pieces (and wrappers) as necessary to customize.
| |
16:51 | Bertl | we plan on using the CMV12k, so there will be a CMV12k frontend
| |
16:52 | Bertl | the sensor can do up to 300FPS, so we plan on not artificially limiting it to less
| |
16:52 | Bertl | (it can even do more at lower resultions)
| |
16:53 | Bertl | now consider a typical camera you can buy today, using this very same sensor
| |
16:53 | bootstrap | outta my price range!
| |
16:54 | Bertl | for various reasons, it will be limited in many ways, but let's ignore that for now
| |
16:54 | Bertl | let's assume you found your perfect CMV12k camera (from sony, red, whatever)
| |
16:55 | Bertl | now cmosis offsers a v2 sensor, same product name CMV12000, same interface, but higher datarates
| |
16:55 | Bertl | it might be that your perfect camera will provide an update as well, but that's rather unlikely
| |
16:56 | Bertl | more likely is that they will do a new camera, which is essentially the old one with the new sensor :)
| |
16:56 | Bertl | that's something we do not really consider 'good'
| |
16:56 | Bertl | so, we design the frontend/interface to be as flexible as possible
| |
16:57 | Bertl | now let's look at the same concept from the price PoV
| |
16:57 | Bertl | we target 4k at 12bit and high frame rates, so it is definitely not consumer
| |
16:58 | Bertl | OTOH, there will be a lot of folks out there which cannot afford this camera and the modules
| |
16:58 | Bertl | still, with open interfaces they can use the same or similar building blocks
| |
16:58 | Bertl | for example, create a frontend with a cheaper, simpler sensor
| |
16:59 | Bertl | or folks interested in still photography at high resolution can add a 60 megapixel sensor which doesn't do more than 15FPS
| |
16:59 | Bertl | and so on and so forth
| |
17:00 | Bertl | naturally we will have to create certain 'products' which basically consist of a default set of modules
| |
17:00 | Bertl | because many folks do not want to build their camera like lego
| |
17:00 | bootstrap | i like that approach in theory, but i suspect in practice it won't work quite as perfectly as that.
| |
17:01 | Bertl | but even those folks might consider upgrading or updating one or the other module
| |
17:01 | bootstrap | for example, if some OTHER company comes out with the next "killer sensor", the electronics will need to be somewhat different.
| |
17:01 | Bertl | there is no perfect in modular, but there is sufficient
| |
17:01 | bootstrap | true.
| |
17:01 | bootstrap | i guess this partly depends on where people get their hardware.
| |
17:02 | bootstrap | if in practice few people try to copy the whole design, you guys will be supplying the hardware (but not an infinite selection of accessories that other people hopefully make).
| |
17:02 | bootstrap | is that how you see this? that not many (or any) others make the core modules?
| |
17:03 | bootstrap | you're not planning to offer bare PCBs (or stuffed PCBs) are you? or are you?
| |
17:03 | Bertl | personally, I would not mind if many folks produce whatever modules
| |
17:04 | bootstrap | sure, but i'm asking for your opinion of what will happen in practice.
| |
17:04 | Bertl | but out of experience, there won't be that many folks copying the core modules if they have a reasonable price
| |
17:04 | bootstrap | that's what i think. i mean, i've got a 4-trace 5-giga-sample-per-second oscope... but not many people do.
| |
17:04 | Bertl | and of course, a good quality
| |
17:05 | bootstrap | the fact it breaks up signals into 64 LVDS channels helps A LOT (with signal speed), but it still isn't a hobby level circuit!
| |
17:05 | Bertl | for the community it doesn't really matter who produces what, as long as the components fit together
| |
17:06 | bootstrap | of course.
| |
17:06 | bootstrap | BTW, where do you envision the image storage is located? In a backpack... or what?
| |
17:07 | bootstrap | I was thinking the data would stream right into a PC, but se6basian said no.
| |
17:07 | Bertl | first stage, storage is external, i.e. SDI recorder
| |
17:07 | Bertl | second stage, storage via small sized SSDs inside the camera
| |
17:07 | Bertl | they are plugged in like SD cards and data is transferred via docking stations
| |
17:07 | bootstrap | what is an SDI recorder?
| |
17:07 | bootstrap | yeah, i understand the SSDs.
| |
17:08 | Bertl | there are a number of readily available recorders which take N (3G) SDI inputs and record the data
| |
17:08 | bootstrap | i was even thinking of storing on a bank of 8 standard insertable flash modules.
| |
17:08 | Bertl | often they are combined with preview or monitoring equippment
| |
17:09 | bootstrap | huh, seems like somehow i never noticed these SDI recorder gizmos.
| |
17:09 | Bertl | modern flash cards start to approach reasonable speeds, but having a complete array of them makes it rather complicated
| |
17:09 | Bertl | i.e. the SSD approach is much simpler for the user
| |
17:10 | Bertl | bootstrap: just google for SDI recorder/monitor
| |
17:10 | bootstrap | yeah, and if we push to the max speed of the camera, probably 8 wide is not nearly wide enough.
| |
17:10 | bootstrap | i did that, just never noticed them before.
| |
17:10 | bootstrap | apparently they contain SSD inside.
| |
17:11 | Bertl | the first stage can also be used with SDI PCIe cards
| |
17:11 | Bertl | (they are also readily available)
| |
17:11 | bootstrap | what do you prefer for storage, if you could go right to your favored configuration?
| |
17:12 | Bertl | currently 1.8" SSD with SATA 6G seems reasonable and doable
| |
17:12 | Bertl | two of them in a striping configuration
| |
17:13 | bootstrap | fast SSD seem to support about 500MBps.
| |
17:13 | Bertl | yes, that gives us about 1Gbyte/s
| |
17:13 | Bertl | which is reasonable and should be sufficient for 'normal' filming
| |
17:13 | bootstrap | not sure whether that number is advertizing, or realistic sustained write speed.
| |
17:13 | bootstrap | yes, for normal frame rates, definitely.
| |
17:14 | bootstrap | for slo-mo... don't think so.
| |
17:14 | Bertl | as SSD easily cross the 500MB/s barrier, it is more the SATA interface imposing the limit
| |
17:14 | bootstrap | 150 FPS seems to require about 23Gbps sustained.
| |
17:14 | guesst | Bertl: there is no such ssd which can do 500MB/s on incompressible stream
| |
17:14 | Bertl | depends on the resultion, bit depth, compression, etc
| |
17:15 | se6astian | back
| |
17:15 | bootstrap | oh, one thing that might help (we shall see) is that i invented a super-simple to implement in FPGA lossless compression. only 20% compression, but... in my camera that made all the difference.
| |
17:16 | bootstrap | if you end up against the wall, and 20% will save your butt, remember this.
| |
17:16 | bootstrap | so it sounds like you prefer to store the images INSIDE the camera.
| |
17:16 | Bertl | yeah, I read that, but I don't see how it is different from normal delta run length encoding?
| |
17:17 | guesst | there is no RL in it :)
| |
17:17 | bootstrap | not sure where you first heard about that. maybe that's where it came from.
| |
17:17 | bootstrap | i guess you read how it works up above.
| |
17:17 | Bertl | anyway, compression cannot be relied on, as you have to plan for the worst case for lossless recording
| |
17:17 | bootstrap | yes indeed.
| |
17:18 | bootstrap | though i never found any examples in which the compression was less than about 16%.
| |
17:18 | Bertl | as I said, there will be options
| |
17:18 | bootstrap | but that's a minor side issue.
| |
17:18 | Bertl | i.e. even if we design the 2nd stage to store data on SSDs
| |
17:18 | bootstrap | i'm just trying to figure out how to implement the most convenient storage system.
| |
17:19 | Bertl | there is still the option to stream them via SDI or 10G ethernet
| |
17:20 | bootstrap | yeah, i was gonna go 10Ge.
| |
17:20 | bootstrap | but then again, i hadn't seen this sensor at 150FPS then!
| |
17:20 | bootstrap | that requires about 23 Gbps!
| |
17:20 | bootstrap | unless i screwed up my math.
| |
17:22 | bootstrap | you know what? maybe we should make our own memory module. i see an SSD that has a write speed of over 4GBps called "fusionIO ioDrive Octal".
| |
17:22 | bootstrap | if we find out what flash parts they have, we just lay out our own module.
| |
17:22 | guesst | last time i've seen flash chips, they were quite slow
| |
17:22 | bootstrap | things change!
| |
17:23 | bootstrap | this one streams (it says) at 4400MB/s.
| |
17:23 | bootstrap | it appears to have 8 or 16 flash memory chips on it (on a PCIe PCB).
| |
17:23 | Bertl | yeah, 2GByte/s is quite common for SSD storage nowadays
| |
17:24 | guesst | one chip did like 20MB/s... today if it does double per die
| |
17:24 | Bertl | guesst: just google for OCZ Z-Drive or Fusion-io
| |
17:25 | guesst | it has hundreds of chips with volume higher than would fit your camera
| |
17:25 | bootstrap | unless they have proprietary chips, we can buy the same chips, solder them to a PCB, and that's our internal memory.
| |
17:26 | bootstrap | i'm looking at a small PCIe card.
| |
17:26 | Bertl | http://ocz.com/images/product/r4_r_series_full_lrg.jpg
| |
17:26 | bootstrap | that claims those speeds.
| |
17:26 | bootstrap | right. that's no monster.
| |
17:26 | guesst | ocz is sandforce, 500M burst, 150MB cont :)
| |
17:26 | Bertl | for example has 4 controller and 4x8 flash chips
| |
17:26 | bootstrap | do they have components on both sides of the PCB?
| |
17:26 | Bertl | the throughput is sustained 2GByte/s read or write
| |
17:26 | guesst | Bertl: on the photo is a daugther board, so double that
| |
17:27 | Bertl | correct
| |
17:27 | guesst | maybe double for both side flash chips :)
| |
17:27 | bootstrap | okay. the point of lots of chips is parallelizm, not storage capacity.
| |
17:27 | guesst | it is basically a SAS HBA and then just sata ssds put on one pcb
| |
17:27 | guesst | not the way to go
| |
17:28 | bootstrap | no, but the chips may be. if they can put flash chips in parallel, so can we.
| |
17:28 | guesst | there is an openssd project
| |
17:28 | bootstrap | i don't think it is sophisticated, just PARALLEL.
| |
17:28 | guesst | you can ask there
| |
17:28 | bootstrap | is the point massive throughput?
| |
17:29 | guesst | with flash you have to find the correct ratio of dies per channel for the best efficiency
| |
17:29 | bootstrap | yes, i'm sure it will take a few minutes of thought. maybe even hours, or days.
| |
17:30 | guesst | it can be calculated :) but if you decide to use different chips, its different
| |
17:30 | bootstrap | but you need to realize, the point for this application is NOT as a hard disk drive, but to capture the camera output.
| |
17:30 | guesst | how does that differ?
| |
17:30 | bootstrap | i don't know yet.
| |
17:31 | guesst | i record my 4k raw dngs to pair of conventional drives :) found it easier than tuning SSDs for performance
| |
17:31 | bootstrap | but if it was up to me, i'd prefer to find a way to shove 23Gbps down an thin little optical cable to the PC, and let the PC capture the data.
| |
17:32 | guesst | use external PCIe
| |
17:32 | bootstrap | part of the question is frame rate. if the new sensors work at 150 FPS as seems to be the case, that's serious throughput!
| |
17:32 | bootstrap | what is that?
| |
17:32 | bootstrap | a PCIe PCB with an external connector?
| |
17:33 | guesst | btw the PC wont handle much bandwidth if you decide to use the chipsets sata ports (limited by DMI to 4x 5Gbps)
| |
17:33 | se6astian | https://twitter.com/ApertusOSCinema/status/427147068176486400
| |
17:33 | guesst | yes, with equalization to pass the few meter cable
| |
17:33 | bootstrap | i know. but frankly i've been itching to make a custom super-high-speed i/o card for PCs for ages! what an excuse!
| |
17:34 | guesst | leave out the card. no longer needed
| |
17:34 | guesst | go direct by pcie
| |
17:34 | bootstrap | but... i also want to make the worlds smallest "PC motherboard" for the fastest 64-bit chips. and so the distance the data would have to travel is insignificant.
| |
17:35 | bootstrap | i dunno. if we're talking 23Gbps, that's not PCIe... is it? well, maybe 16-channel would.
| |
17:35 | bootstrap | actually, i guess 8-channel would.
| |
17:35 | guesst | 8x 5G would do :)
| |
17:35 | guesst | or 4x 8G.. pcie gen3
| |
17:36 | bootstrap | BTW, what is the minimum "reasonable" storage capacity?
| |
17:36 | bootstrap | speaking "in camera" now.
| |
17:37 | guesst | depends on use and how much gear you got behind the camera :)
| |
17:37 | bootstrap | and what is the slowest "reasonable" rate to transfer data out of the camera into a PC or whatever.
| |
17:37 | bootstrap | yeah, but that's not an answer that helps create a schematic or lay out a PCB.
| |
17:37 | guesst | it is not a pcb issue, it is conceptual issue
| |
17:38 | bootstrap | a camera is a physical device, not a conceptual one.
| |
17:39 | bootstrap | hell, my conceptual camera weighs 1 gram, has a 65536x32768 sensor, runs at up to 65536 frames per second, and has 10 years of storage (continuous operation).
| |
17:39 | bootstrap | but i can only actually build a real, physical camera (and only out of existing components).
| |
17:40 | guesst | the good concept is about making possible a wide range... not just the most common or reasonable
| |
17:40 | rexbron | bootstrap: an EVF is an electronic viewfinder. Think small hi res screen with a diopter
| |
17:40 | bootstrap | let's see. 10 minutes == 600 seconds @ 2.5GB/s == 1500GB? ouch!
| |
17:41 | guesst | you do not shoot 10min at full fps :)
| |
17:41 | Bertl | but that's the possible, not just the common or reasonable :)
| |
17:41 | bootstrap | very true, that would be unusual.
| |
17:42 | bootstrap | so maybe 1/3 of that == 50 FPS or 1/6 of that == 24 FPS.
| |
17:42 | guesst | unusual things can have a volume handicap
| |
17:42 | guesst | unless serious money is put in :)
| |
17:42 | bootstrap | so that would be 250GB... more reasonable.
| |
17:43 | guesst | i use a 1TB/hr estimations (25p 4k/12b)
| |
17:44 | bootstrap | that's pretty close.
| |
17:44 | Bertl | as I was told, most 'shoots' are not longer than a few minutes, maybe up to half an hour
| |
17:44 | guesst | usually it was like 10 min (look for film roll length)
| |
17:44 | Bertl | after that, it isn't a problem to switch the recording medium
| |
17:44 | bootstrap | you mean the whole shoot, or each continuous shot?
| |
17:44 | guesst | each cont.
| |
17:45 | bootstrap | true. in fact, these days, seems like nothing stays on the screen for more than half a second! hahaha.
| |
17:45 | bootstrap | nonetheless, there are exceptions.
| |
17:46 | bootstrap | i like to watch that 12 (or was it longer) minute continuous shot near the beginning of the movie Serenity.
| |
17:46 | bootstrap | so... there are times and places for that kind of thing.
| |
17:46 | Bertl | agreed, and I think there is no need to put artificial limits on that
| |
17:46 | bootstrap | the other question is, how long to transfer 250GB or 1000GB out of the camera to a computer.
| |
17:47 | guesst | depends on the interface of course :)
| |
17:47 | bootstrap | yeah, i'm just trying to figure out the minimum amount of memory we could put in the camera (on one of our own PCBs) and not piss too many people off.
| |
17:47 | bootstrap | whoops... sorry. i meant "you". hahaha.
| |
17:48 | guesst | what kind of memory?
| |
17:48 | Bertl | bootstrap: well, it is a good sign that you see it as 'we' already :)
| |
17:48 | bootstrap | i don't want to be presumptuous!
| |
17:48 | bootstrap | you guys have a lot of work in this already.
| |
17:48 | Bertl | the basic design idea might go like this:
| |
17:49 | Bertl | 1TB is enough for most cases
| |
17:49 | bootstrap | yes, i suppose DRAM is another option.
| |
17:49 | Bertl | let's say, 512GB for a start, so design for this, but keep solutions with 2+TB in mind
| |
17:50 | bootstrap | though you might be killed if someone accidentally pulls the power plug and looses his last hour of shots!
| |
17:50 | bootstrap | okay, go ahead.
| |
17:50 | Bertl | result, we have something that works for most cases and doesn't cost an arm an a leg
| |
17:50 | Bertl | and if there is demand for more, we can easily extend that
| |
17:50 | Bertl | or if we do not want to extend it, somebody else can do so
| |
17:50 | rexbron | Be warry of advertised SSD speeds
| |
17:50 | Bertl | to cover the 'special' cases
| |
17:51 | bootstrap | which makes DRAM sounds a lot better all the time.
| |
17:51 | rexbron | BMD keeps the list small because most drives cant sustain the data rate
| |
17:51 | guesst | BM is not enough demanding app
| |
17:51 | rexbron | also, MLC nand can be tricky
| |
17:51 | Bertl | rexbron: yes, this will require some testing, but I'm sure half a year from now they will be there (at least at the SATA 6G speeds)
| |
17:51 | bootstrap | if the battery is in the camera, then unlikely the DRAM will lose power.
| |
17:51 | rexbron | maybe, maybe not
| |
17:52 | rexbron | MLC is the devil
| |
17:52 | Bertl | rexbron: and if not, we could easily use 3 or 4 drives
| |
17:52 | rexbron | Bertl: that is what the delta had to do but up goes the price :)
| |
17:52 | guesst | Bertl: 3-4 drives with unexpected behaviour is not a solution
| |
17:52 | rexbron | one mag is now $800
| |
17:52 | rexbron | and it's raid 0 :P
| |
17:53 | rexbron | Didn't have enough media to backup on set? Lose a drive? Poof, there goes your days footage
| |
17:53 | guesst | rexbron: at red?
| |
17:53 | rexbron | guesst: I don't understand your question
| |
17:54 | guesst | what mag are you talking about
| |
17:54 | rexbron | guesst: 4x $200 off the shelf SSDs
| |
17:54 | guesst | ok
| |
17:55 | bootstrap | 8GB of seriously fast DRAM costs $120.
| |
17:55 | Bertl | rexbron: again, there is an option for tradeoff, e.g. you can do error correction/redundancy coding so that even if one drive gets lost, the data is still there
| |
17:55 | rexbron | bootstrap: and its still DRAM
| |
17:55 | bootstrap | what do you mean?
| |
17:56 | bootstrap | i mean... is it necessary that people be able to wait several hours to drive home and download to their computer?
| |
17:56 | bootstrap | if not, what's the problem with DRAM?
| |
17:56 | bootstrap | i mean, i understand the problem, but it sure has a lot going for it too.
| |
17:57 | Bertl | DRAM as storage is not feasable
| |
17:57 | bootstrap | cost, density, power consumption, easy interface.
| |
17:57 | bootstrap | why is that?
| |
17:57 | guesst | bootstrap: in pro world, slower than realtime is not too accepted
| |
17:57 | guesst | e.g. wait at most the same time as for shooting
| |
17:57 | bootstrap | why is DRAM slow?
| |
17:57 | bootstrap | you mean to download into storage (a PC) after shooting for 30 minutes?
| |
17:57 | Bertl | you saud $120 for 8GB, you can get 120GB of SSD storage for the same price
| |
17:57 | Bertl | *said
| |
17:58 | guesst | if you do 30min shooting, then you probably go to external rec.
| |
17:58 | bootstrap | i know, but the DRAM is much faster.
| |
17:58 | bootstrap | and the interface is easier.
| |
17:58 | Bertl | it also consumes a lot more power
| |
17:58 | guesst | and you fill it in 3 secs (2.4G/s)
| |
17:59 | bootstrap | oh, i think i momentary slipped GB and TB. minor detail, only a factor of 1024x.
| |
17:59 | bootstrap | don't pay attention to that moron behind the curtain!
| |
18:00 | rexbron | bootstrap: that is a feasable solution for highspeed. That his how the phantoms do it
| |
18:00 | Bertl | so you can get high speed DRAM with 8TB for 120 USD?
| |
18:00 | bootstrap | yes... that's PC2666.
| |
18:00 | rexbron | ring buffer in ram, trigger locks the buffer and it's played out
| |
18:00 | bootstrap | nah. gotta keep up, so better to go straight into whatever is the storage.
| |
18:02 | Bertl | bootstrap: DDR3 memory typically costs about 5USD/Gigabyte
| |
18:03 | Bertl | your calculation is like 0.1USD/Gigabyte, any specific chips you are looking at?
| |
18:03 | guesst | Bertl: he made a mistake, stop picking on it
| |
18:04 | Bertl | I know there was a mistake, but I'm not sure where ...
| |
18:04 | bootstrap | hey, i was only off by a factor of 1024x. minor detail, right?
| |
18:05 | bootstrap | back to flash memory.
| |
18:05 | Bertl | okay :)
| |
18:05 | bootstrap | i really should have bought ECC when i purchased my brain.
| |
18:06 | guesst | one + for own flash: crash recovery (especially if the chips store the values in special way)
| |
18:06 | guesst | (my wife killed a sandforce OCZ ssd in 3 months... thanks to encryption it is not reconstructible)
| |
18:07 | guesst | e.g. if accidentally a chip fails, one might appreciate a 2K dataset instead of 4K
| |
18:07 | Bertl | everything can be reconstructed with flash chips
| |
18:07 | Bertl | there are many companies making good money with drive reconstruction
| |
18:08 | Bertl | anyway, I don't think that failing SSDs are a real problem for recording
| |
18:08 | bootstrap | i figured out a simple to implement 64-bit error correction scheme that works on 4KB (32Kb) streams. it doesn't take much overhead on long streams like that (640 bits).
| |
18:08 | Bertl | but maybe I'm wrong, and film makers would honor a raid setup
| |
18:09 | bootstrap | i think because we can control where we write the stream, we can get away with a very efficient ECC like this.
| |
18:09 | guesst | in demanding apps they would do (i asked about daisy chained recorders as backup/raid: not acceptable)
| |
18:09 | guesst | so usually it goes by two completely separate paths
| |
18:09 | guesst | same for live broadcast
| |
18:10 | guesst | single point of failure being reduced to the camera itself
| |
18:10 | bootstrap | i think the likely problem is inside a given flash chip, not the circuitry.
| |
18:27 | Bertl | bootstrap: so, to get back to development, did you have a look at the sensor frontend yet? if so, any comments on that? or maybe the FPGA code?
| |
18:27 | bootstrap | you mean the circuitry?
| |
18:28 | bootstrap | i don't know where to see the circuit design or FPGA.
| |
18:28 | Bertl | okay, do you know eagle?
| |
18:28 | bootstrap | the PCB software?
| |
18:28 | Bertl | yup
| |
18:28 | bootstrap | no, i work with a different program.
| |
18:29 | Bertl | which one?
| |
18:30 | bootstrap | diptrace. they have schematic software and PCB layout.
| |
18:30 | Bertl | okay, anyway, you can download eagle for free for Linux, MacOS and Windows (the free version)
| |
18:31 | Bertl | and it doesn't require any complicated install or so
| |
18:31 | Bertl | https://github.com/apertus-open-source-cinema/alpha-hardware/tree/master/SFE-PCB/cmv12k-adapter-v1.2
| |
18:31 | Bertl | here is the schematic and board layout
| |
18:32 | Bertl | for the FPGA part, the software can be found here: https://github.com/apertus-open-source-cinema/alpha-software/tree/master/cmv_io2
| |
18:32 | bootstrap | those are eagle files? is there a [semi]-free version of eagle to view these with?
| |
18:32 | Bertl | yes, as I said, just download it and run it as freeware
| |
18:33 | bootstrap | okay, i'll do that. but i've been working for about 20 hours now, and getting a little loopy, so that's have to be tomorrow.
| |
18:33 | bootstrap | maybe that's why i can't tell the difference between GB and TB any more! hahaha.
| |
18:33 | Bertl | no problem, get some sleep first :)
| |
18:34 | bootstrap | the FPGA will probably make my brain explode anyway. i've only worked with altera cyclone devices.
| |
18:35 | Bertl | don't worry, it's not that complicated
| |
18:35 | guesst | Bertl: was Altera considered or xilinx is the only way (in europe? :)
| |
18:35 | bootstrap | gads, these flash datasheets are opaque!
| |
18:35 | Bertl | we considered both, we also contacted altera, but they were not interested in working with us
| |
18:36 | bootstrap | in what way? they have great design software (free).
| |
18:36 | guesst | so do xilinx :)
| |
18:36 | Bertl | and the zynq platform was selected because of the zedboard
| |
18:36 | Bertl | bootstrap: xilinx has the webpack (i.e. free as in beer) as well
| |
18:36 | danhanes | jumping in here - I have a good bit of Altera experience, but the Xilinx part seems the best suited for this applications
| |
18:37 | danhanes | *application
| |
18:37 | guesst | btw how hot is the zynq part (or how much W it consumes - when leaving the arm core idle) ?
| |
18:37 | bootstrap | well, one thing that totally torqued me was finding that the cheaper altera parts cyclone5 have 60 LVDS --- four short! the next step up is 120 LVDS.
| |
18:38 | bootstrap | OTOH, probably we need more i/o anyway.
| |
18:38 | Bertl | I agree, upcoming altera chips might be a good choice hardware wise, so maybe at some point we switch or offer an alternative baseboard
| |
18:38 | bootstrap | but the price difference was $60 ==> $160... so quite a difference.
| |
18:39 | bootstrap | i haven't heard about the cyclone6 yet, but it is overdue.
| |
18:40 | bootstrap | it says the LVDS is capable of 875Mhz, which is more than plenty.
| |
18:40 | bootstrap | but... no problem. my biggest concern is the hardware.
| |
18:41 | bootstrap | i'm still trying to wrap my head around the whole system concept.
| |
18:43 | Bertl | as I said, if you have questions, just ask, we will try to answer :)
| |
18:45 | bootstrap | i will. i was just looking at a flash chip that is 16-bits parallel with (it appears) 5ns per write.
| |
18:46 | bootstrap | so that's 200Mbps * 16 == 400MBps for one chip.
| |
18:46 | guesst | to internal buffer.
| |
18:46 | bootstrap | not counting overhead to start up the process.
| |
18:46 | guesst | look at page erase / block program
| |
18:46 | bootstrap | i thought i did, but might have missed something (it is a big datasheet).
| |
18:47 | Bertl | well, erase is not that relevant, can nowadays be done in the background and/or for our purpose at the beginning
| |
18:47 | bootstrap | and very obscure. man, people don't know how to write specs any more!
| |
18:47 | Bertl | block write cycles are relevant
| |
18:47 | bootstrap | yup.
| |
18:47 | troy_s | left the channel | |
18:47 | rexbron | left the channel | |
18:47 | bootstrap | i thought that's what i was seeing.
| |
18:47 | guesst | bootstrap: those datasheets are fairly good.. i have met some very censored ones :)
| |
18:47 | troy_s | joined the channel | |
18:47 | rexbron | joined the channel | |
18:47 | rexbron | left the channel | |
18:47 | rexbron | joined the channel | |
18:48 | bootstrap | anyway, if that is even close to what it can do, then a handful of those might be sufficient (say, 16 or 32 of them).
| |
18:49 | bootstrap | but yeah, i definitely need to study the damn spec sheet, cuz it is massively obscure.
| |
18:49 | bootstrap | i'm looking at a micron datasheet at the moment.
| |
18:49 | guesst | which part?
| |
18:50 | bootstrap | MT29C5G96MAYBACJG-5
| |
18:50 | bootstrap | plus a bunch of other part numbers.
| |
18:50 | bootstrap | stupid piece of crap doesn't even say storage capacity on the cover page!
| |
18:51 | bootstrap | supposedly 512M x 16 though.
| |
18:51 | guesst | made a typo there, 5G capacity does not exist:)
| |
18:51 | bootstrap | freaking LPDRAM... another new term for me.
| |
18:52 | guesst | btw that is not a typical flash
| |
18:52 | guesst | but a PoP flash+dram combo
| |
18:52 | bootstrap | so, if 32 were required, that would mean 8GB * 32 == enough.
| |
18:53 | guesst | have a sleep and recalc tomorrow :)
| |
18:53 | bootstrap | well, i thought it just had a dram interface (as an alternate way to access the flash).
| |
18:53 | bootstrap | hahaha... yeah. not the best time to start looking at something sorta unfamiliar.
| |
18:53 | guesst | it has separate interfaces
| |
18:53 | bootstrap | right. but it doesn't actually have dram in it... does it?
| |
18:53 | guesst | has
| |
18:53 | guesst | two stacked dies
| |
18:53 | bootstrap | really?
| |
18:53 | guesst | (at least two)
| |
18:54 | bootstrap | sheesh.
| |
18:54 | guesst | that is a chip for phone... put it on the cpu itself
| |
18:54 | guesst | so you get a cpu core+ram+flash on one place
| |
18:55 | bootstrap | wonder why it showed up as the fastest flash memory chip.
| |
18:55 | bootstrap | in digikey.
| |
18:58 | bootstrap | the serial flash (which i was somewhat familiar with) seem to be ~100Mbps (but of course much cheaper and smaller packages).
| |
18:58 | guesst | and capacitites in mbit
| |
18:58 | bootstrap | yeah, like 256Mb
| |
18:58 | bootstrap | 32MB
| |
18:58 | bootstrap | 512Mb
| |
18:58 | bootstrap | 64MB
| |
18:59 | bootstrap | so... tiny.
| |
18:59 | bootstrap | would need thousands of them.
| |
18:59 | bootstrap | i think i haven't found the relevant flash memory chips yet.
| |
18:59 | bootstrap | easy to find the modules, but... not finding the chips.
| |
19:00 | bootstrap | yeah, another day i guess.
| |
19:00 | bootstrap | thanks everyone. time for me to go unconscious for a few hours.
| |
19:01 | bootstrap | so good morning and good night and adios.
| |
19:02 | guesst | bootstrap: look at pictures of ssd's and then lookup the parts
| |
19:02 | guesst | or just go to micron site and select the nand flash section and the largest capacity
| |
19:02 | bootstrap | i'll give that a try. maybe i can find a review that reveals the flash part #s in the products.
| |
19:03 | bootstrap | is micron the best fast flash supplier?
| |
19:03 | guesst | MT29F1T08CUCABH8-6 : 1Tbit
| |
19:03 | guesst | get 8 for a TB drive
| |
19:03 | bootstrap | the problem is, we need to put at least 32 in parallel to get enough bandwidth.
| |
19:04 | bootstrap | i think, anyway. maybe 16.
| |
19:04 | guesst | you have to read the DS.. there are multiple dies in it
| |
19:04 | guesst | so you need to calculate the actual write speed
| |
19:04 | guesst | and then cap that by the TDP, to not fry the thing
| |
19:05 | bootstrap | SLC, MLC, TLC ???
| |
19:05 | guesst | MLC
| |
19:19 | bootstrap | left the channel | |
19:34 | troy_s | left the channel | |
19:34 | rexbron | left the channel | |
19:34 | rexbron | joined the channel | |
19:34 | troy_s | joined the channel | |
19:38 | rexbron | left the channel | |
19:38 | troy_s | left the channel | |
19:38 | rexbron | joined the channel | |
19:38 | troy_s | joined the channel | |
19:38 | rexbron | left the channel | |
19:38 | rexbron | joined the channel | |
20:53 | shinokubo | joined the channel | |
21:12 | shinokubo | left the channel | |
21:24 | intracube | joined the channel | |
22:02 | se6astian | time for bed, good night!
| |
22:02 | se6astian | left the channel | |
22:54 | shinokubo_ | joined the channel |