Current Server Time: 04:43 (Central Europe)

#apertus IRC Channel Logs

2016/07/25

Timezone: UTC


01:04
Bertl_oO
off to bed now ... have a good one everyone!
01:04
Bertl_oO
changed nick to: Bertl_zZ
01:44
Spirit532
joined the channel
05:52
Napejts
joined the channel
06:54
se6astian|away
changed nick to: se6astian
07:09
Napejts
does steingate has lot fights like hunter x?
07:13
pusle
left the channel
07:42
Spirit532
hey, apertus dev people
07:42
Spirit532
have any of you tried to push the cmv sensor clock?
07:43
Spirit532
CMOSIS don't mention the possibility of overclocking ANYWHERE in their datasheets, so I'm curious how it would perform at higher clocks
07:44
Spirit532
nothing crazy, 50-100MHz higher at most
07:46
Spirit532
though 100MHz is quite insane in terms of image sensor clock
07:49
Napejts
left the channel
07:49
Napejts
joined the channel
08:19
Spirit532
anybody? :(
08:40
se6astian
nope, we didnt do it
08:41
Spirit532
can you try it? :)
08:42
Spirit532
I'm debating on whether or not I should take the CMV2k and make something low-end, but overclocked
08:43
se6astian
not with the current hardware
08:43
Spirit532
why not?
08:44
Spirit532
even with 32 pairs instead of 64 on the cmv12k you have right now, it should be possible to just bump the clock and see if you get an image
08:45
se6astian
sure but it might behave quite differently
08:45
se6astian
also we are running the CMV12k at 250mhz instead of 600 currently
08:46
Spirit532
oh.
08:52
Bertl_zZ
changed nick to: Bertl
08:52
Bertl
morning folks!
08:53
Bertl
Spirit532: what do you hope for when you overclock the sensor by 10% or so?
08:54
Spirit532
higher framerate
08:54
Bertl
so instead of let's say 600FPS, 660?
08:55
Spirit532
that's a tiny overclock
08:55
Bertl
10%
08:55
Spirit532
most sensors are capable of 50% and higher
08:55
Bertl
well, I doubt that at 600MHz
08:56
Spirit532
no reason to not try
08:56
Bertl
except for the high sensor price .. but yes
08:58
Spirit532
you won't kill the sensor if it's cooled
08:58
Spirit532
maybe lock it up, but nothing a reset won't fix
09:09
se6astian
changed nick to: se6astian|away
09:11
Bertl
for the CMV12k I wouldn't expect much overclocking to be possible
09:11
Bertl
they designed the sensor for 600MHz, but had to release the first version with 300MHz because they couldn't reliably reach the 600MHz
09:12
Bertl
version two now is specified at 600MHz though ...
09:22
Spirit532
680 IIRC
09:50
ItsMeLenny
joined the channel
09:59
Spirit532_
joined the channel
10:02
Spirit532
left the channel
10:10
Bertl
off for now ... bbl
10:10
Bertl
changed nick to: Bertl_oO
11:23
se6astian|away
changed nick to: se6astian
12:09
ItsMeLenny
left the channel
12:48
Spirit532_
changed nick to: Spirit532
12:52
se6astian
we just moved into our new office: https://twitter.com/ApertusOSCinema/status/757545933794140160/photo/1
12:54
Spirit532
se6astian, that's neat
12:55
Spirit532
now you need to hire 30 underage chinese workers and reduce the price of the beta to $3
12:55
Spirit532
what's the funny-looking metal table for?
12:57
se6astian
why not 60 - ah yes because thats not matching our philosophy: https://www.apertus.org/opensource
12:58
se6astian
the table is missing the tabletop
13:01
Alex_Chooks
joined the channel
14:02
se6astian
changed nick to: se6astian|away
14:09
se6astian|away
changed nick to: se6astian
14:09
se6astian
changed nick to: se6astian|away
14:17
Napejts
left the channel
14:43
Spirit532
So to go from 936 to 1000 fps @ 4096x1080 with the CMV12k the clock boost needs to only be 41MHz
14:43
Spirit532
I think that's well within the reach, if not operating optimally
14:44
Spirit532
gets you 1479fps @ 4096x720
14:44
Bertl_oO
it might be possible with some sensors
14:44
Bertl_oO
the problem is more that your calculation is flawed
14:44
Spirit532
is it?
14:44
Bertl_oO
I presume so, at least from the values
14:45
Spirit532
well, 1000/936 is 1.01683 recurring
14:45
Spirit532
so we need to boost the clock 1.01683 times
14:45
Bertl_oO
there are three stages which consume time, the precharge, the exposure and the readout stage
14:45
Spirit532
er, 1.0683
14:46
Bertl_oO
you cannot really speed up the precharge stage, even with higher clock rates
14:46
Bertl_oO
you also have some limits to the exposure time, but that shouldn't be that criticial around 1000FPS
14:46
Spirit532
exposure time is fine
14:46
Spirit532
1/963rd is not that more than 1/1000th
14:46
Bertl_oO
and then you have the actual readout time, which will speed up as you crank up the clock
14:47
Spirit532
(roughly speaking)
14:47
Bertl_oO
you also need to compensate for readout noise when you overlap exposure and readout
14:48
Spirit532
theoretically though, "making the sensor go" 1.06 times faster should be doable
14:48
Bertl_oO
very likely
14:49
Spirit532
I don't mind 936fps, but 1000 just ticks the OCD mark for me, heh
14:49
Bertl_oO
6% is typically within the safety margin, especially if you cool the sensor
14:49
Spirit532
I plan on doing active TE cooling the sensor in my camera
14:49
Spirit532
that is, after I get enough money to do it, right now I'm sticking with the CMV2000, because it allows for overall much cheaper OTS hardware
14:51
Spirit532
right now I'm planning on using the MYIR Z-turn board, since it's only $119 for the 7020, and it's got 39 LVDS pairs
14:51
Spirit532
but I'm waiting on clarification on their 200MHz limit via the connectors
14:52
Bertl_oO
yes, looks like a good choice for this purpose
14:52
Bertl_oO
the connectors are unfortunately not very performant
14:52
Spirit532
yeah, I really want to be able to run it full-speed
14:52
Spirit532
plus I'd need that for the SSD to record raw
14:53
Bertl_oO
you plan to connect an SSD directly to the zynq?
14:54
Spirit532
possibly
14:54
Spirit532
or with some glue
14:55
Spirit532
not sure about it yet, nothing is concrete - I haven't read much zync doku
14:56
Spirit532
possibly CFast cards, if it's easier
14:56
Bertl_oO
sounds interesting ... let me know when you got a design which you think should work ...
14:58
Bertl_oO
the 1.27mm connectors on the back are an unfortunate design decision when it comes to high speed interfaces
14:59
Bertl_oO
you have a lot of cross talk, not much shielding or ground and the pins/pads are relatively large
15:01
Bertl_oO
you might get away with the 600MHz interface though, if you place the pairs properly ... but I don't see that you get up to 1GHz or higher on this connector, but please let me know how it goes
15:02
Spirit532
I'm not going for 600
15:02
Spirit532
the cmv2k has 16 pairs @ 480
15:02
Spirit532
I want to push it to 500-550 to see if the sensor pops or starts displaying extraterrestrial alien images
15:04
Spirit532
though it'll probably just stop functioning past 10% OC
15:05
Bertl_oO
looking forward to the alien images :)
15:06
Spirit532
as am I :>
15:06
Spirit532
hopefully it'll be soon, since there's not much to do
15:06
Spirit532
make a simple length-matched breakout for the sensor and whatever IO I want
15:29
se6astian|away
changed nick to: se6astian
15:50
se6astian
changed nick to: se6astian|away
15:50
se6astian|away
changed nick to: se6astian
16:01
Bertl_oO
inter pair length matching for the sensor is something you can safely forget about
16:02
Bertl_oO
the LVDS pairs have a timing offset anyway, so you need to delay each of them separately
16:02
se6astian
changed nick to: se6astian|away
16:03
Bertl_oO
i.e. it is not a source synchronous output as the datasheet suggests ... i.e. each LVDS pair has a specific offset/phase
16:03
se6astian|away
changed nick to: se6astian
16:04
se6astian
changed nick to: se6astian|away
16:05
se6astian|away
changed nick to: se6astian
16:06
Napejts
joined the channel
17:21
Alex_Chooks
left the channel
17:28
arpu
joined the channel
17:46
Spirit532
Bertl_oO, but I like wiggly lines :(
17:46
Spirit532
and it's 3 buttons
17:47
Bertl_oO
well, if you don't care about signal quality, go ahead :)
17:47
Spirit532
is that... bad?
17:47
Spirit532
I'm thinking of length matching it down to the package
17:48
Spirit532
I have an x-ray machine that can give me the internal lengths :^)
17:48
Spirit532
if you're overengineering, overengineer to the max! :D
17:49
Bertl_oO
the package lengths are specified by the manufacturer
17:49
Spirit532
does cmosis specify theirs?
17:50
Spirit532
I know xilinx does for the zynqs, but not sure about the cmvs
17:50
Bertl_oO
but it should be obvious, that length matching with meanders degrades the signal
17:50
Spirit532
well, I won't be adding 50mm to each line
17:50
Bertl_oO
every bend will cause EM
17:50
Spirit532
meh, I'll see
17:51
Spirit532
the PCBS will be dirt cheap, since there won't be any BGAs or very tight tolerances
17:51
Spirit532
I can probably even get away with 4 layers
17:55
slikdigit
joined the channel
18:03
Spirit532
left the channel
19:24
Napejts
left the channel
20:52
se6astian
changed nick to: se6astian|away
20:56
slikdigit
left the channel
20:56
slikdigit
joined the channel
21:21
Bertl_oO
off to bed now ...
21:21
Bertl_oO
changed nick to: Bertl_zZ
21:38
Napejts
joined the channel
22:11
se6astian|away
changed nick to: se6astian
22:13
se6astian
I am off to bed
22:14
se6astian
changed nick to: se6astian|away
22:36
Napejts
left the channel