Current Server Time: 19:50 (Central Europe)

#apertus IRC Channel Logs

2014/01/25

Timezone: UTC


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