09:36 DottorLeo: Hi!
09:43 DottorLeo: Is there any documentation about the meaning of every stat on fossildb reports? Like InvThroughput, MaxWaves, etc... and if it's is better when they increase/decrease? Thanks!
09:44 DottorLeo: Just to understand the stats
10:27 pendingchaos: radv_GetPipelineExecutableStatisticsKHR() in radv_pipeline.c and statistic_infos in aco_interface.cpp has descriptions
10:27 pendingchaos: it's better if they decrease unless more_is_better=True in report-fossil.py's stat list
10:34 DottorLeo: @pendingchaos thanks
10:40 javier0: Hello. I have an rdna2 amd graphics card and I was playing around with pp_od_clk_voltage yesterday. I was able to find stable values for OD_MCLK and OD_VDDGFX_OFFSET, but I'm not sure how the latter works - I was looking at https://www.kernel.org/doc/html/latest/gpu/amdgpu/thermal.html#pp-od-clk-voltage but it's not described there. Anyone have a link to docs describing fields in
10:40 javier0: /sys/class/drm/card0/device/pp_od_clk_voltage?
11:19 javier0: Sorry I have to disconnect, will be back later
11:40 javier0: I'm back, yet I realized it's Sunday so probably lots of you are afk...
11:40 javier0: s/lots/a lot/
12:59 fililip: javier0: https://wiki.archlinux.org/title/AMDGPU, section 5.3.2
13:07 javier0: fililip: thanks! yes, I was on archlinux wiki also, but if you will ctrl+f and look for OD_VDDGFX_OFFSET - there's nothing there, also in the official kernel docs they linked to
13:08 javier0: here: https://docs.kernel.org/gpu/amdgpu/thermal.html#pp-od-clk-voltage
14:18 fililip: that's just voltage
14:18 fililip: the voltage offset, i mean
14:19 fililip: personally i prefer using something like corectrl/lact, these gui apps are just more readable in general
14:20 fililip: lact has a systemd unit for maintaining these settings without the gui client so if you're running something headless it might be useful
14:20 javier0: fililip: I'm just using amdgpu-clocks
14:23 javier0: I have set OD_VDDGFX_OFFSET to -105mV, anything above that value causes GPU to crash, does this depend from silicon to silicon? (and of course gpu model)
14:23 javier0: s/above/below/
14:31 fililip: yeah, probably
14:32 fililip: from what i know, a 100mv undervolt is usually what works
14:33 javier0: fililip: thanks!
14:46 javier0: OD_MCLK I was able to settle at 900MHz, anything below also caused a crash, FORCE_POWER_CAP won't let me go down below 239W as it returns error that provided value is wrong
14:46 javier0: but I recall reading an article saying there is a new mechanism in kernel preventing to lower wattage too much
14:47 javier0: I mean, in the amdgpu driver
14:57 javier0: I'll be back later