<cphealy>
Ahh, so the cros ec probably has a current sensor on the board somewhere that provides the overall current consumption I assume.
<macc24>
there is a difference
<macc24>
500mhz g72, whole device consumes around -114xxxx<something>, 800mhz: -116xxxx
<macc24>
when rendering glxgears on half of 1920x1200 screen
<cphealy>
is the framerate with glxgears the same between 500 and 800MHz?
<macc24>
i forgot to check ,_,
<macc24>
no big difference in framerate
<macc24>
i'd say that further testing is needed
<cphealy>
ack
<HdkR>
You'd want it vsync locked for testing :P
<macc24>
HdkR: it is by default
<cphealy>
I was just going to say that. ;-)
<cphealy>
If vsync locked, wouldn't both cases be 60fps?
<macc24>
yeah they were
<cphealy>
Or at least no more than 60fps.
<macc24>
the results might be different on midgard, other bifrosts and on valhall
<HdkR>
It will end up being different everywhere :D
<macc24>
i can check midgard tomorrow if i have energy to do so
<cphealy>
macc24: That sounds like a pun :-)
<macc24>
oh this wasn't intentional
raster has joined #panfrost
endrift has quit [Ping timeout: 264 seconds]
zkrx has quit [*.net *.split]
macc24 has quit [*.net *.split]
macc24 has joined #panfrost
zkrx has joined #panfrost
raster has quit [Quit: Gettin' stinky!]
endrift has joined #panfrost
popolon has quit [Quit: WeeChat 3.0]
stikonas has quit [Remote host closed the connection]
karolherbst has quit [Ping timeout: 272 seconds]
vstehle has quit [Ping timeout: 260 seconds]
Green4 has joined #panfrost
robertfoss_ has joined #panfrost
zkrx has quit [Disconnected by services]
Green has quit [*.net *.split]
gcl has quit [*.net *.split]
uis has quit [*.net *.split]
robertfoss has quit [*.net *.split]
griffinp has quit [*.net *.split]
endrift has quit [*.net *.split]
Green4 is now known as Green
zkrx has joined #panfrost
gcl has joined #panfrost
xperia64 has joined #panfrost
_whitelogger has joined #panfrost
bbrezillon has quit [Remote host closed the connection]
bbrezillon has joined #panfrost
endrift has joined #panfrost
q4a has joined #panfrost
vstehle has joined #panfrost
<q4a>
Hi all. I tested panfrost on supertuxcart - got 60 fps on lowest (1st preset), 38-40 on low settings (2st preset), but when I enable dynamic light (2nd preset + dynamic light or 3rd preset) - only 5 fps
Green has quit [Quit: ...]
<q4a>
i'm using rk3288 with mesa from git. It's a bit strage for me, but I see 18-20% cpu usage with 38-40 fps on low graphics settings, but on low graphics settings + dynamic light - I got only 7-8% cpu usage with 5 fps
Green has joined #panfrost
kaspter has quit [Ping timeout: 264 seconds]
kaspter has joined #panfrost
<HdkR>
You could just be GPU bottlenecked :)
<icecream95>
q4a: The SuperTuxKart "Advanced pipeline" is not well optimised for tiling GPUs because the frame is rendered in multiple passes, with the GPU having to write the entire frame to main memory each time
<icecream95>
Enabling it makes the game a lot slower even with the blob driver
<q4a>
yes) I tried `GALLIUM_HUD=GPU-load,cpu,fps supertuxkart`, but I don't see gpu load
<HdkR>
That gallium-hud flag only works for r600 and radeonsi is why
<icecream95>
Using framebuffer fetch or pixel local storage, it should be possible to significantly improve performance, and I have a WIP branch doing that which I should resurrect...
nlhowell has quit [Remote host closed the connection]
davidlt has quit [Ping timeout: 264 seconds]
Elpaulo has quit [Ping timeout: 260 seconds]
urjaman has quit [Read error: Connection reset by peer]
urjaman has joined #panfrost
nlhowell has joined #panfrost
karolherbst has joined #panfrost
karolherbst has quit [Ping timeout: 272 seconds]
alpernebbi has joined #panfrost
karolherbst has joined #panfrost
_whitelogger has joined #panfrost
stikonas has joined #panfrost
stikonas has quit [Remote host closed the connection]
icecream95 has quit [Ping timeout: 256 seconds]
stikonas has joined #panfrost
warpme_ has joined #panfrost
davidlt has joined #panfrost
karolherbst has quit [Ping timeout: 260 seconds]
stikonas has quit [Remote host closed the connection]
stikonas has joined #panfrost
nlhowell has quit [Ping timeout: 264 seconds]
warpme_ has quit [Quit: Connection closed for inactivity]
chewitt has quit [Quit: Zzz..]
kaspter has quit [Quit: kaspter]
karolherbst has joined #panfrost
stikonas has quit [Remote host closed the connection]
<macc24>
cphealy: it looks like frequency matters on t860 on my kevin (rk3399) too
<macc24>
and i uses cros ec to measure current on it too
<cphealy>
Cool, thank you for checking.
<macc24>
bbl
<cphealy>
I've been researching how overall (not per-process) GPU load can be read from userspace with Linux and I'm not finding a lot. With panfrost I see devfreq is used for frequency scaling which makes it's decisions using a computed "busy_time". Is there a way for this busy time to be accessed from userspace?
karolherbst has quit [Ping timeout: 260 seconds]
karolherbst has joined #panfrost
chewitt has joined #panfrost
chewitt has quit [Quit: Zzz..]
stikonas has joined #panfrost
stikonas has quit [Remote host closed the connection]
q4a has quit [Remote host closed the connection]
tomboy64 has quit [Read error: Connection reset by peer]
<cphealy>
macc24: this depends on the GPU HW counters and probably by itself doesn't provide details on the percentage of wall clock time the GPU is busy working. I was thinking of something more along the lines of exposing devfreq busy time or GPU scheduler busy time. Perhaps this doesn't exist though.
q4a has joined #panfrost
popolon has joined #panfrost
alyssa has quit [Remote host closed the connection]
alyssa has joined #panfrost
davidlt has quit [Ping timeout: 240 seconds]
raster has quit [Quit: Gettin' stinky!]
alpernebbi has quit [Quit: alpernebbi]
raster has joined #panfrost
cphealy has quit [Ping timeout: 260 seconds]
cphealy has joined #panfrost
karolherbst has quit [Ping timeout: 260 seconds]
archetech has joined #panfrost
raster has quit [Quit: Gettin' stinky!]
archetech has quit [Client Quit]
archetech has joined #panfrost
karolherbst has joined #panfrost
tomboy64 has quit [Remote host closed the connection]