Let me explain a little better (sorry Tom) how you should go about 'testing' your GPUs for Einstein FGRPB1G tasks, assuming this is what you are trying to do.
[EDITED to shorten my reply]
HTH!
It's all good but I do find this to be terribly obvious. My problem has not been knowing how but has been going back to collect the times and knowing which I was running at the time and getting it all into a sheet. Purely a matter of not keeping detailed tracking at the time because I started out doing it based on GPU temp when I ASSUMED that temp would track directly to times. It doesn't.
I just ran a couple more sets of 3x and have reverted to 2x now to collect fresh times from those. What so far is not making sense is the 3x is running longer than 5x or 4x. Based on old #s I think 2x is going to end up being the best choice for my card / box. The 3x thing still bothers me though.
I just ran a couple more sets of 3x and have reverted to 2x now to collect fresh times from those. What so far is not making sense is the 3x is running longer than 5x or 4x. Based on old #s I think 2x is going to end up being the best choice for my card / box. The 3x thing still bothers me though.
Skip
If your 3x time is running longer than 5x or 4x, have you looked into the invalids or errors? Take into consideration that what is in the 'tasks pending' file could very well include the ones that you are specifically waiting for to verify whether or not your 3x, 4x, or 5x is running too long or not at all because of a failure.
What I'm trying to say is you need to be a bit more cognizant of the fact that it is a bit more complex than simply testing 3x, 4x, or 5x and checking times.
I just ran a couple more sets of 3x and have reverted to 2x now to collect fresh times from those. What so far is not making sense is the 3x is running longer than 5x or 4x. Based on old #s I think 2x is going to end up being the best choice for my card / box. The 3x thing still bothers me though.
Skip
If your 3x time is running longer than 5x or 4x, have you looked into the invalids or errors? Take into consideration that what is in the 'tasks pending' file could very well include the ones that you are specifically waiting for to verify whether or not your 3x, 4x, or 5x is running too long or not at all because of a failure.
What I'm trying to say is you need to be a bit more cognizant of the fact that it is a bit more complex than simply testing 3x, 4x, or 5x and checking times.
I filtered out anything that wasn't "completed and validated"
I'm pulling all valid GW tasks for a single box off the website and into a spreadsheet.
any feedback about the new amd gpu 7900 xt and 7900 xtx on einstein applications? FP32 seems good in theory, but I'd like to understand how these two GPUs works in reality.
any feedback about the new amd gpu 7900 xt and 7900 xtx on einstein applications? FP32 seems good in theory, but I'd like to understand how these two GPUs works in reality.
Just ordered a 7900xt to replace my old RTX 2060. Will post results in the coming days. Power efficiency is clearly on NVidias side this generation and the 20GB memory won't really help here with E@H. But although it's mostly for Gaming, that never prevented me to let it work for E@H in the past. We'll see.
Nothing better than a burn-in stress test for new hardware than running E@H :)
I just hope Mesa 23.0, amdgpu and 6.2.9 Kernel will do the job. I heard 6.3 Kernel will have major AMD stuff, but that's still weeks ahead. Never ran Linux with Radeon before.
any feedback about the new amd gpu 7900 xt and 7900 xtx on einstein applications? FP32 seems good in theory, but I'd like to understand how these two GPUs works in reality.
Just ordered a 7900xt to replace my old RTX 2060. Will post results in the coming days. Power efficiency is clearly on NVidias side this generation and the 20GB memory won't really help here with E@H. But although it's mostly for Gaming, that never prevented me to let it work for E@H in the past. We'll see.
Nothing better than a burn-in stress test for new hardware than running E@H :)
I just hope Mesa 23.0, amdgpu and 6.2.9 Kernel will do the job. I heard 6.3 Kernel will have major AMD stuff, but that's still weeks ahead. Never ran Linux with Radeon before.
Hi EXARD3K and thanks. I'm very interested to see the results when you share them in the next few days.
Specifically . . . Description: Update to ROCm 5.4.1
:
: Notice: GFX 11 hardware might not work yet. <---!!!!!!
Severity: None
The one thing that you can count on with Nvidia cards is that the driver installation is a extremely simple process and they always work out of the box.
Update: Got the card, done a fresh OS install (Tumbleweed, running with 6.2.10 Kernel).
The 7900xt is a beast in gaming. Works out of the box, I ran benchmarks 15 seconds after first Plasma session started. So far I'm impressed, had a lot more trouble with setting up NVidia driver for linux.
But yeah, BOINC doesn't detect any GPU. So ROCm, OpenCL or whatever doesn't work yet. AMD just annouced new Radeon PRO cards and there are some major updates coming with 6.3.
I could dig down the rabbithole and spend the weekend to get it to work, but I'll probably just wait for developers to catch up with the new cards. A shame "new" hardware always takes months to work flawlessly. And I'm on a leading edge distro, not some Ubuntu LTS from 2018.
Guess we'll have to wait for proper ROCm a bit longer. I'll post some WU stats once proper support kicks in.
The good thing with AMD, I can set power limits, fan curve and shit without persistence-mode,artificial limits or using the god-aweful nvidia-settings GUI. And I can see all the sensors that Nvidia never gave me. That's quite the relief.
Tom M wrote: George, No
)
It should be a simple testing process, but somehow I think Skip MAY have been slightly confused. If not, no harm and no foul.
Proud member of the Old Farts Association
GWGeorge007
)
It's all good but I do find this to be terribly obvious. My problem has not been knowing how but has been going back to collect the times and knowing which I was running at the time and getting it all into a sheet. Purely a matter of not keeping detailed tracking at the time because I started out doing it based on GPU temp when I ASSUMED that temp would track directly to times. It doesn't.
I just ran a couple more sets of 3x and have reverted to 2x now to collect fresh times from those. What so far is not making sense is the 3x is running longer than 5x or 4x. Based on old #s I think 2x is going to end up being the best choice for my card / box. The 3x thing still bothers me though.
Skip
Skip Da Shu
)
If your 3x time is running longer than 5x or 4x, have you looked into the invalids or errors? Take into consideration that what is in the 'tasks pending' file could very well include the ones that you are specifically waiting for to verify whether or not your 3x, 4x, or 5x is running too long or not at all because of a failure.
What I'm trying to say is you need to be a bit more cognizant of the fact that it is a bit more complex than simply testing 3x, 4x, or 5x and checking times.
Proud member of the Old Farts Association
GWGeorge007 wrote: Skip Da
)
I filtered out anything that wasn't "completed and validated"
I'm pulling all valid GW tasks for a single box off the website and into a spreadsheet.
Greetings to you all, any
)
Greetings to you all,
any feedback about the new amd gpu 7900 xt and 7900 xtx on einstein applications? FP32 seems good in theory, but I'd like to understand how these two GPUs works in reality.
Cristian Frigerio
)
Just ordered a 7900xt to replace my old RTX 2060. Will post results in the coming days. Power efficiency is clearly on NVidias side this generation and the 20GB memory won't really help here with E@H. But although it's mostly for Gaming, that never prevented me to let it work for E@H in the past. We'll see.
Nothing better than a burn-in stress test for new hardware than running E@H :)
I just hope Mesa 23.0, amdgpu and 6.2.9 Kernel will do the job. I heard 6.3 Kernel will have major AMD stuff, but that's still weeks ahead. Never ran Linux with Radeon before.
I have a Vega56 undervolted
)
I have a Vega56 undervolted and overclocked, and set -20% power limit when running E@H.
With these settings, it takes around 5 minutes to crunch 1x GRP WU (3465 credits) and the card consumes 148W.
In theory, this card would produce around 990k credits per day @ 148W. Not too bad efficiency.
Exard3k wrote: Cristian
)
Hi EXARD3K and thanks. I'm very interested to see the results when you share them in the next few days.
Greetings
Might regret that decision.
)
Might regret that decision. Look at this thread. RADEON RX 7900 XTX LINUX (FEDORA) BOINC DOESN'T DETECT USABLE DRIVER
Specifically . . . Description: Update to ROCm 5.4.1
:
: Notice: GFX 11 hardware might not work yet. <---!!!!!!
Severity: None
The one thing that you can count on with Nvidia cards is that the driver installation is a extremely simple process and they always work out of the box.
Update: Got the card, done a
)
Update: Got the card, done a fresh OS install (Tumbleweed, running with 6.2.10 Kernel).
The 7900xt is a beast in gaming. Works out of the box, I ran benchmarks 15 seconds after first Plasma session started. So far I'm impressed, had a lot more trouble with setting up NVidia driver for linux.
But yeah, BOINC doesn't detect any GPU. So ROCm, OpenCL or whatever doesn't work yet. AMD just annouced new Radeon PRO cards and there are some major updates coming with 6.3.
I could dig down the rabbithole and spend the weekend to get it to work, but I'll probably just wait for developers to catch up with the new cards. A shame "new" hardware always takes months to work flawlessly. And I'm on a leading edge distro, not some Ubuntu LTS from 2018.
Guess we'll have to wait for proper ROCm a bit longer. I'll post some WU stats once proper support kicks in.
The good thing with AMD, I can set power limits, fan curve and shit without persistence-mode,artificial limits or using the god-aweful nvidia-settings GUI. And I can see all the sensors that Nvidia never gave me. That's quite the relief.