From all the chatter on the AMD thread topics with drivers, the AMD RX6000 series of cards have greater compatibility with the drivers and seem to install easier.
The RX7000 series are still a work in progress with the drivers.
The FGRP1G-tasks are running with only 6 to 10% of CPU-Time from Run-Time. So I don't believe there is really a benefit of running a second task in parallel, as the processor needs time for swapping between the tasks. I don't believe the 6950XT will run by a factor of about 10 faster than my 6600. That's different to O3MDF. If I remember correctly, they are running about equal long on CPU and GPU.
I agree, it's really peculiar, that so much more GPUs from NVIDIA than AMD are used also here. From my longtime observation on E@H server status page I know, the benefit of AMD is by a factor 2,62 +/- 1,02 higher than NVIDIA, taken from the last 392 measurements.
I edited the BOINC systemd .service file, replaced "ProtectSystem=strict" with "ProtectSystem=off".
Performance is OK in my opinion, most Einstein@Home GPU tasks complete in ~2 minutes. I believe I once reached 1 million RAC, but now my RAC is around 600k and falling. I crunch all the available Einstein@Home GPU tasks without exception.
Sam, you should enable beta tasks on the project to get the faster 1.28 version of the app for AMD cards. Should be much faster than the 1.18 app version.
Also would like to know your task concurrency since it isn't printed out in stderr.txt file for AMD apps it seems unlike what is shown for the CUDA apps. Maybe it will be for the 1.28 version since Petri rewrote much of that and he included that data in the CUDA versions.
I'm curious if anyone in
)
I'm curious if anyone in Einstein@Home uses the AMD Radeon RX 6950 XT GPU?
Because I'm curious, a.k.a. Curious George (and I had 2 bananas this morning), I would like to know:
I know that NVIDIA has out paced AMD in the GPU sales roughly 8:1. But I would still like to know.
Thanks in advance for your time in answering these questions.
Proud member of the Old Farts Association
From all the chatter on the
)
From all the chatter on the AMD thread topics with drivers, the AMD RX6000 series of cards have greater compatibility with the drivers and seem to install easier.
The RX7000 series are still a work in progress with the drivers.
Hallo George!The
)
Hallo George!
The FGRP1G-tasks are running with only 6 to 10% of CPU-Time from Run-Time. So I don't believe there is really a benefit of running a second task in parallel, as the processor needs time for swapping between the tasks. I don't believe the 6950XT will run by a factor of about 10 faster than my 6600. That's different to O3MDF. If I remember correctly, they are running about equal long on CPU and GPU.
I agree, it's really peculiar, that so much more GPUs from NVIDIA than AMD are used also here. From my longtime observation on E@H server status page I know, the benefit of AMD is by a factor 2,62 +/- 1,02 higher than NVIDIA, taken from the last 392 measurements.
Kind regards and happy crunching
Martin
George, One of the
)
George,
One of the confounding factors of your question is most of the late model Radeon systems are not running Linux.
I am not sure how compatible Darwin is to Linux is either.
Out of the top 100 users list, how many are running late model Radeon gpu's?
We need a list so we can get at more data.
Radeon VII's appear to be processing about 50% the speed of Rtx 3080 ti's.
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor)
https://www.hardwaretimes.com
)
https://www.hardwaretimes.com/amds-999-rx-7900-xtx-is-faster-than-the-1599-rtx-4090-in-jedi-survivor-even-with-ray-tracing/
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor)
Dear all, Keith asked me
)
Dear all,
Keith asked me to chime in, so here I am :)
I have an AMD Radeon RX 7900 XTX running on Ubuntu 22.04 x86_64. I made the following changes to make it work:
Performance is OK in my opinion, most Einstein@Home GPU tasks complete in ~2 minutes. I believe I once reached 1 million RAC, but now my RAC is around 600k and falling. I crunch all the available Einstein@Home GPU tasks without exception.
This is my host: https://einsteinathome.org/host/13105647 , feel free to have a look :)
Best regards,
Sam
+42 Thank you.
)
+42
Thank you.
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor)
magic_sam wrote:This is my
)
Sam,
Are you running 1 task or 2 per gpu on your grp#1 tasks?
The times would be impressively low if you were running 2 tasks at a time.
If you switch to pure grp#1 it looks like you could boost past 2M on your RAC.
Thank you again!
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor)
Sam, you should enable beta
)
Sam, you should enable beta tasks on the project to get the faster 1.28 version of the app for AMD cards. Should be much faster than the 1.18 app version.
Also would like to know your task concurrency since it isn't printed out in stderr.txt file for AMD apps it seems unlike what is shown for the CUDA apps. Maybe it will be for the 1.28 version since Petri rewrote much of that and he included that data in the CUDA versions.
Hi Tom, I'm only running
)
Hi Tom,
I'm only running one GPU task at a time. I tried to run two tasks some time ago but I had stability and heating issues.
Cheers, Sam