#apertus IRC Channel Logs

who is here?

great, any news bertl worth sharing?

Mainly to see where the performance limits are and it turns out that the MIPI interfaces and data transport from them is definitely not something hitting the limits easily

And this also included frame verification via a checksum and the checksum algorithm makes up the larger part of the 35%

I tried USB drives (the obvious choice) first but those seem to be limited at around 120MB/s sustained write throughput (and I couldn't find any USB stick which would go beyond that)

Then I tested with NVMe attached via various USB to NVMe bridges and while the sustained read speed goes up to 380MB/s there, the sustained write speed caps out at 260MB/s on the best option

I also tested USB to SATA bridges with fast SSDs but the results are unfortunately similar or worse

Not sure how the hardware (USB Controller) could be to blame, but of course that could be the problem here as well

http://vserver.13thfloor.at/Stuff/AXIOM/BETA/RPI5/

If you have any further questions regarding the setup or how to reproduce, feel free to contact me

excellent!, many thanks!

we capture a buch a of testclips with/without blackframe subtraction which seems to work rather well directly in the axiom recorder but not well in davinci resolve (nobody knows what the software does to the raw data...

that also captured fine but we dropped a bunch of frames as its beyond the datarate of 25 FPS throughput

what's the limit there?

2048x1080@50 FPS is too much for magewell / USB or other part in the chain

but we don't know which part?

like the toshiba mipi hdmi bridge chip, I just made a request for quote in the myarrow app

I also started preparing the ninja V hdmi test with BogdanXORs device but failed to replace the gateware in FW 2.0 with confusing symlinks and PATHs, anuejn will help me simplify that process at the next tinkering session hopefully

thats it from my side I think

ah I started converting my axiom recorder gui from pysimplegui (which is now a paid software) to dearpygui which anuejn suggested

interesting! good to know!

Here is the datarate overview

we will test where the bottleneck is

the magewell should be fine as it offers that mode without issue

the USB limit should be somewhere around 400MB/s in an ideal setup it could get up to 480MB/s

I'm not sure what you mean with the max throughput on the beta output? the HDMI data rate?

what is the max the beta could supply?

well, with 148.5MHz and the 1080p60 config we have very precisely 60FPS

the 2048x1080p60 mode uses the same timing but is 2048 pixel wide

could we go 2048 but more height?

more height requires a different format

so in theory we could add a few more lines vertically and keep the same timing frame

yeah, the standard formats are enough to keep us busy right now

i.e. we could use the 60FPS output but run the capture at 50FPS

right

we will report with next tinkering session

is there a way to tell if the USB interface is getting saturated?

you probably also want to make sure that there is no audio in the capture as that will needlessly fill up the USB data stream

but how to tell if its not the SSD writing too slow?

simple, do not write to the SSD for verification, just make checksums and check the marker

great

thanks for the moderation!