14:46ric96: karolherbst: looks like you mutter copr built fot f33, I'm using it now, its usable gnome3 and all work.
14:46ric96: Some glitches.
14:46ric96: However glmark2 performance is weirdly underwhelming with anywhere between 54-5fps, i s this expected?
14:46ric96: Thats on the Jetson Nano
14:47karolherbst: ric96: ahh, yeah, expected on the nano
14:47karolherbst: it's the lowest end and I think eMC is stuck at low clocks as well
14:47karolherbst: eMC = embedded memory controler
14:47karolherbst: so performance won't be great
14:47karolherbst: I think an updated kernel might help. I think something landed in 5.9
14:48karolherbst: but not sure
14:48karolherbst: the glitches are also known to me, lower resolution seem to help. Or having a more powerful board..
14:48karolherbst: I think it's some syncing issue somewhere, but didn't had time to investigate
14:50ric96: Gotcha, all good.
14:50ric96: I think even arm core clocks are effected as well to some degree, cpufreq doesn't seem to be working and that might be slowing it down as well.
14:51karolherbst: the jetson nano didn't get as much love as the other more powerful boards, and I think that's fine, just annoying for users :)
14:52karolherbst: I suspect it will get better over time though
14:53karolherbst: ric96: but maybe you could give the latest rc a try? I couldn't see any improvements on my end, but I also wasn't really running any benchmarks
14:53karolherbst: I think you can just install the rawhide kernel
14:54ric96: Hmmm, the last I tried rawhide wasn't booting on the nano, oddly enough.
14:54ric96: I'll give just the kernel another go
14:55karolherbst: ric96: btw, did you verify that nouveau is used with glxinfo or something? I know that I need a firmware level workaround to even get nouveau to load on newer kernels/l4t versions
14:57ric96: GL_RENDERER: tegra
14:57ric96: GL_VERSION: ... Mesa 20.2.0-rc4
14:57ric96: GL_VENDOR: NVIDIA
14:58karolherbst: ahh, okay
14:58karolherbst: should be fine then
18:00ric96: karolherbst: rawhide 5.9.0-rc4 enabled cpufreq, but now nouveau doesn't load, nothing on dmesg as well. firmware issue?
18:05karolherbst: ric96: https://gist.github.com/karolherbst/0bb8747ba620d6926e640cb6e7e12b84
18:06karolherbst: sudo mkimage -A arm -O linux -T script -C none -a 0 -e 0 -n "Tegra Boot script" -d boot.txt /boot.scr
18:06karolherbst: that should do the trick
18:06karolherbst: uboot will pick this file up and apply the workarounds specified
18:07karolherbst: this is somehow caused by a missmatch of the dts stuff from the firmware and the kernel.. and the firmware just reports something broken back or so
18:08karolherbst: the file has to be in / of the last partition, which I think contains everything from the OS.. not sure what kind of layout you are using
18:09karolherbst: but over the serial console it will tell if it picks the file up or not
18:10karolherbst: the bug seen from the kernel is, that some dts entries have the same handle and the kernel gets all confused
18:11karolherbst: ric96: another issue is, that even though we can reclock the GPU with nouveau, it has to be done manually and the stuff is quite unstable
18:12karolherbst: I think there is something wrong with applying the voltage or something as this is out of the control of nouveau anyway
18:12karolherbst: but I never really looked into why it doens't work all that great
18:16ric96: so my layout is not using the nvidia partitioning scheme on the sdcard, i have all the preboot partitions including uboot on the spi flash. this can be done with the latest l4t release.
18:16ric96: The sdcard is just a generic f33 aarch64 server image.
18:16ric96: So it boots grub.efi
18:17karolherbst: I see
18:17ric96: I tried running the boot.txt commands manually in uboot before issuing the boot command but that didn't help
18:17karolherbst: maybe it's a new issue then
18:17karolherbst: didn't try 5.9 yet
18:18ric96: although nouveau seems to be loaded
18:18ric96: oddly nothing on demesg, let me try increasing the loglevel
18:20karolherbst: ric96: yeah.. nouveau loads, but it should fail to bind the device because something is missing
18:20karolherbst: some fail on the dts level
18:20karolherbst: at least that's what I know from 5.8...
21:23swick: I was just watching XDC day 3 stream and Lyude mentioned that it's possible to get access to vesa specs. how does that work exactly?
22:08gentlewonder: It does not really much matter what hardware people use, they formulate group of maybe 5-6 different architectures, regardless of cpu or gpu model all of them can be made to perform decently enough for human being.
22:10gentlewonder: Hence i am definitely mobile equipment guy, and mostly cheap entry level of that.
22:43gentlewonder: This tech world and those lead implementations and base code is still very interesting to work on, computer algorithms is one of my most favorite real interest, it is fascinating to me and to everyone else too i think.