I have not much experiences with Einstein.
Until before yesterday I had unchecked the thread title question.
Since yesterday I have it checked.
I have (inter alia) an Intel Celeron J1900 CPU with iGPU ('Intel HD Graphics' (without a number ;-) ) and NV GT730.
With the:
Binary Radio Pulsar Search (Arecibo) v1.34 (opencl-intel_gpu-new)
windows_x86_64
- app the WUs lasts around 1 hour (on the Intel HD Graphics).
Then with the:
Binary Radio Pulsar Search (Arecibo) v1.34 (opencl-intel_gpu-Beta)
windows_x86_64
- app the WUs lasts 1 hour and 30 minutes (on the Intel HD Graphics).
Now with the:
Binary Radio Pulsar Search (Arecibo) v1.34 (opencl-intel_gpu-new)
windows_x86_64
- app the WUs lasts 1 hour and 30 minutes also (on the Intel HD Graphics)?
In past with the 'new' app the GPU Load was continuously over ~ 95 %.
After with the 'beta' and with the 'new' the GPU Load is just jumping between ~ 60 - 80 %.
Is this because something is broken now, or this is just because of the currently available WUs?
All Binary Radio Pulsar Search (Arecibo) tasks for computer 12298250:
http://einsteinathome.org/host/12298250/tasks&offset=0&show_names=0&state=0&appid=19
(In both cases the apps use ~ 400 MB system RAM.
BOINC say the Intel HD Graphics have 1304 MB.
In BIOS I set 'use max possible', in Windows 8.1's Task-Manager it say 554 MB reserved for hardware.
So in real the GPU have just ~ 512 MB 'VRAM'?
So 2 WUs/GPU aren't possible, because not ~ 800 MB available?)
If I looked correct, my PC could get the following apps:
http://einstein.phys.uwm.edu/apps.php
Binary Radio Pulsar Search (Arecibo) - BRP4
i-1.34 (opencl-intel_gpu)
i-1.34 (opencl-intel_gpu-new)
i-b-1.34 (opencl-intel_gpu-Beta)
Binary Radio Pulsar Search (Parkes PMPS XT)
c-1.52 (BRP6-cuda32-nv301)
c-b-1.57 (BRP6-Beta-cuda55)
i-b-1.52 (BRP6-Beta-opencl-intel_gpu)
Binary Radio Pulsar Search (Arecibo, GPU)
c-1.39 (BRP4G-cuda32-nv301)
c-b-1.52 (BRP4G-Beta-cuda32-nv301)
i-b-1.52 (BRP4G-Beta-opencl-intel_gpu)
[i = Intel, c = CUDA (NV), b = beta]
If the thread title is checked, my PC get just the follwing (beta) apps?
Binary Radio Pulsar Search (Arecibo) - BRP4
i-b-1.34 (opencl-intel_gpu-Beta)
Binary Radio Pulsar Search (Parkes PMPS XT)
c-b-1.57 (BRP6-Beta-cuda55)
i-b-1.52 (BRP6-Beta-opencl-intel_gpu)
Binary Radio Pulsar Search (Arecibo, GPU)
c-b-1.52 (BRP4G-Beta-cuda32-nv301)
i-b-1.52 (BRP4G-Beta-opencl-intel_gpu)
(BTW. AFAIK, Einstein have the Beta-Project Albert, there the apps will be tested before they are released here? So the Beta apps at Einstein aren't buggy and should run smoothly?)
Thanks.
Copyright © 2024 Einstein@Home. All rights reserved.
I got a new test application
)
I got a new test application just today:
Application
Gamma-ray pulsar binary search #1 on GPUs 1.24 (FGRPopenclTV-nvidia)
Name
LATeah3011L00_620.0_0_0.0_9691074
State
Ready to start
Received
5/19/2021 9:24:18 AM
Report deadline
6/2/2021 9:24:18 AM
Resources
1 CPU + 1 NVIDIA GPU
Estimated computation size
525,000 GFLOPs
Executable
hsgamma_FGRPB1G_1.24_windows_x86_64__FGRPopenclTV-nvidia.exe
mikey wrote:I got a new test
)
No you didn't! :-).
It's not a test app (no beta in the name) and it's not new. Check the applications page. It was released back on January 29 to solve problems for recent nvidia cards (Turing, Volta, Ampere series), hence the 'TV' in the name. There was a lot of discussion at the time. Bernd did post about it but since it didn't affect me I didn't make a proper note of exactly what happened. Search for the app name. You'll probably find the full story if you're interested. The discussion was a couple of days before release if I remember correctly. I don't remember it taking very long to 'fix' the problem :-).
My guess for the download of this app now is that perhaps you have recently installed a GPU that requires it?
Cheers,
Gary.
Gary Roberts wrote: mikey
)
I'm still allowing test apps and am running the O3AS tasks that way instead of by clicking on the box to run them and then all of a sudden over half a dozen on these new for me tasks showed up.
mikey wrote:... all of a
)
The task you listed is a GRP task. It's nothing to do with your settings for O3ASE or for test apps.
Have you not done GRP on that host before? To get them now, perhaps you have GRP GPU tasks enabled or perhaps you have allowed 'non-preferred' apps. Maybe there was a temporary shortage of O3ASE and the scheduler sent you some GRP tasks instead.
Cheers,
Gary.
Gary Roberts wrote: mikey
)
I think the very last idea is the reason it happened because I have BRP cpu tasks selected and Yes to run test applications and was happily doing doing the O3AS tasks and then all of a sudden I got over half a dozen of those GRP tasks.
Sutaru Tsureku написал:In
)
Hi Sutaru
I would look what another computational devices doing when iGPU starts to show sporadic load.
Your host has NV device also and all GPU apps use increased priority over CPU ones so NV device could compete with iGPU part for CPU cycles. Hence underload for iGPU.
Just one of possible reasons other than "broken state" due to pre-history.