Something must be wrong with this application ... although I get work for application OpenCL for AMD GPUs 9.15 (mt), this application is only using my 8 CPU cores.

In Status column, it says Compiting (8 CPUs) but it doesn't requires any ATI GPU ...

This app is at the moment appearing to BOINC as a multithreaded CPU only application.

I'll have to wait until your machine's returned some work before I can comment on exactly what resources were used.

Matt

Captainjack - are you getting work now?

MJH

Yep, I just got 16 tasks. One is running now.

It is running on all 16 CPU threads and shows 98% GPU utilization on an AMD 6950.

As reported earlier, it shows 100% complete after about 90 seconds of run time, but it is going ahead with its calculations.

Let me know if you need more information.

As reported earlier, it shows 100% complete after about 90 seconds of run time, but it is going ahead with its calculations.

Yes, it's not reporting progress yet.

Matt

Greetings MJH

Running a WIN 7 x64 Quad8400 with an AMD HD 5850 1GB

1 am reporting I have 4 tasks. One is running now.

It is running on all 4 CPU threads and shows 68% GPU utilization on an AMD 5850.

Also as reported earlier, it shows 100% complete after about 90 seconds of run time but continues to run until it is completed.

I have completed a couple of validated tasks as well.

Hope this helps with the indication another AMD user is having some success.

Regards

Something must be wrong with this application ... although I get work for application OpenCL for AMD GPUs 9.15 (mt), this application is only using my 8 CPU cores.

This also happened to me. But at the time mt GPUGrid app started to run another CPU app was still running. It acted like this even though I set 100% of CPU time and processors in BOINC preferences. When CPU WU (Universe@home) finished the GPUGrid mt app kicked in and GPU was at 60%.

Something must be wrong with this application ... although I get work for application OpenCL for AMD GPUs 9.15 (mt), this application is only using my 8 CPU cores.

In Status column, it says Compiting (8 CPUs) but it doesn't requires any ATI GPU ...

This app is at the moment appearing to BOINC as a multithreaded CPU only application.

I'll have to wait until your machine's returned some work before I can comment on exactly what resources were used.

Matt

Thanks for your answer ... it seems to be running fine now but it's now fully using the GPU : I see spikes from 0 to 100% every couple of seconds ... and the GPU is only pulling 100W where it can pull up to 250W when fully used.

Thanks for your answer ... it seems to be running fine now but it's now fully using the GPU : I see spikes from 0 to 100% every couple of seconds ... and the GPU is only pulling 100W where it can pull up to 250W when fully used.

Yes. Getting the balance right between CPU and GPU is going to be a challenge to do automatically. f you have a look at the output
http://www.gpugrid.net/result.php?resultid=13773682 you'll see a message:

NOTE: The GPU has >25% less load than the CPU. This imbalance causes performance loss

Try dialing down the amount of CPU you are allocating to the the application until that message goes away. Go too far and you'll get a message saying the GPU has too much load.

Note that the CPU use is assigned when the job is sent to your client, so you can't change it for any jobs you have already received, even if they are still queued.

Matt

The trial rig I used ask's for work but is told there is "No tasks are available for OpenCL for AMD GPUs" although there seems to be plenty available by the server status. I'm guessing my problem is that I have mixed rigs and they are not supported yet? Is that right?
I have no rigs with single AMD's in them all I have is 3 rigs with mixed GPU's both ATI & Nvidia in the same box which work extremely well with mixed boinc projects but apparently not here ATM. I sure hope they will in the future.
I guess like others i'm out of luck until things progress further.

Your most definitely to be commended for trying something new. Good luck with it.

In my logs, Gormacs says also that you should use SSE 4.2 on my CPU instead of SSE 2 ... I don't know if you can do that automatically.

In fact, you're going through what I've alread known in Folding@Home when they experimented with Gromacs, SMP and GPUs :

- I've seen a MPI Gromacs core which was using MPI and processes approach which required the installation of OpenMPI which was causing many troubles (performances issues on Windows, computations errors when a netowrk device was plugged/unplugged, ...)
- They then moved to a threaded approach in Gormacs which is much more reliable and faster on a signle computer environment. These cores are still in production and didn't change for almost a couple of years ... the only flaw of this core is that it's still 32 bits and using quite an old codebase now but I don't know if they plan on using Gromacs in the future core versions ...
- Gromacs has been the base of many GPU cores too, starting with DirectX implementation which turned to be unusable outside a lab and going through CAL/Brook+ ATI cores and CUDA cores. ATI versions are already deprecated and no longer supported, and NV ones support is progressively being dropped.
- Since the move to using OpenCL on ATI and NV platforms, Stanford is now using their own code called OpenMM. These applications are GPU only and don't compute on CPU (although they might need one free CPU thread to feed the GPU efficiently).

This is fun to see the history being reapeated here in this project ... I hope you won't have as much troubles as they had while developing these tools.

Try dialing down the amount of CPU you are allocating to the the application until that message goes away. Go too far and you'll get a message saying the GPU has too much load.

Note that the CPU use is assigned when the job is sent to your client, so you can't change it for any jobs you have already received, even if they are still queued.

Matt

Matt,

I'm not quite sure how you're advocating that people "dial down the amount of CPU you are allocating". BOINC doesn't really have a mechanism for users to control how much CPU is allocated to a particular application.

There is the website (or local) preference "On multiprocessors, use at most xxx% of the processors", but this controls how many CPU cores BOINC can use as a whole, across all projects - within that limit, any MT task will be set (on assignment) to use every core available to BOINC. There's no way, using the web or GUI tools, to adjust the number of cores available to this AMD Beta app, while still allowing BOINC to use other cores for other applications or projects.

It is, however, possible to use an Application configuration (app_config.xml) file to control this, providing you are running at least v7.2.42 of the BOINC client.

You can set both

       <avg_ncpus>x</avg_ncpus>
       <cmdline>--nthreads n</cmdline>

for an mt plan_class (recommended that x == n), and even change the values in mid-run by stopping and restarting the application - though this will mess up Matt's test statistics, and it might be preferable to change parameters between tasks, rather than during a run.

Do you expect that in the future ATI app will work exactly like app for NV? I mean low CPU usage, high GPU usage, multi GPUs?

I'm seeing some unexplainable discrepencies in points attributions ...

Here are some examples :

13765549 10604022 197813 1 Feb 2015 | 14:54:24 UTC 1 Feb 2015 | 15:12:24 UTC Terminé et validé 1,061.62 7,840.83 8,721.16 OpenCL for AMD GPUs v9.15 (mt)
13765548 10604021 197813 1 Feb 2015 | 16:07:56 UTC 1 Feb 2015 | 18:13:03 UTC Terminé et validé 1,076.36 7,901.26 237.89 OpenCL for AMD GPUs v9.15 (mt)
13765546 10604019 197813 1 Feb 2015 | 15:50:46 UTC 1 Feb 2015 | 17:54:40 UTC Terminé et validé 1,133.30 8,401.45 232.80 OpenCL for AMD GPUs v9.15 (mt)
13765530 10604003 197813 1 Feb 2015 | 15:12:24 UTC 1 Feb 2015 | 17:19:47 UTC Terminé et validé 1,456.66 10,339.33 277.93 OpenCL for AMD GPUs v9.15 (mt)
13765525 10603998 197813 1 Feb 2015 | 14:56:11 UTC 1 Feb 2015 | 16:07:56 UTC Terminé et validé 1,029.07 7,702.50 152.70 OpenCL for AMD GPUs v9.15 (mt)
13765523 10603996 197813 1 Feb 2015 | 14:57:24 UTC 1 Feb 2015 | 16:38:51 UTC Terminé et validé 921.62 6,911.97 150.75 OpenCL for AMD GPUs v9.15 (mt)
13765522 10603995 197813 1 Feb 2015 | 14:56:47 UTC 1 Feb 2015 | 16:23:26 UTC Terminé et validé 930.95 6,987.43 146.14 OpenCL for AMD GPUs v9.15 (mt)
13765498 10603971 197813 1 Feb 2015 | 15:03:01 UTC 1 Feb 2015 | 16:55:29 UTC Terminé et validé 995.72 7,425.57 174.81 OpenCL for AMD GPUs v9.15 (mt)
13765494 10603967 197813 1 Feb 2015 | 13:32:03 UTC 1 Feb 2015 | 14:52:31 UTC Terminé et validé 998.30 7,420.52 8,201.05 OpenCL for AMD GPUs v9.15 (mt)
13765493 10603966 197813 1 Feb 2015 | 13:32:03 UTC 1 Feb 2015 | 14:52:31 UTC Terminé et validé 989.75 7,359.32 8,130.77 OpenCL for AMD GPUs v9.15 (mt)
13765492 10603965 197813 1 Feb 2015 | 13:32:03 UTC 1 Feb 2015 | 14:52:31 UTC Terminé et validé 996.33 7,397.32 8,184.82 OpenCL for AMD GPUs v9.15 (mt)

I could not resist and rearranged 3 of my systems to have only amd gpu's available.

http://www.gpugrid.net/results.php?hostid=161299 is a i3 with a HD7770 gpu-load ~57% cpu-load ~57%

http://www.gpugrid.net/results.php?hostid=196771 is a AMD A10 7700 APU gpu-load up to 100%, cpu-load ~ 60%

http://www.gpugrid.net/results.php?hostid=197814 is a AMD Phaenom x4 with a HD6950 gpu, gpu load ~74%, cpu-load ~67%

All run win7 x64
all finish and validate
all systems run gpugrid only
cpu load as of win task manager, gpu load as gpu-z reports it. All systems set to use 2 cpu's. Giving more cores did not increase gpu load.

If any further info may helpful pls request it here.

Alexander

I'm seeing some unexplainable discrepencies in points attributions ...

Here are some examples :

For everyone, who wants to share their data which is actually a spreadsheet, I'd like to give some tips on how to format the data the way others can read it, and see the data in a similar way as it appeared at its origin.
1. Use the "pre" tag, as it marks the enclosed text as pre-formatted text, so the forum's web engine won't do its own formatting, like filtering multiple spaces, and the enclosed text will be displayed using a non-proportional font, so every letter will have exactly the same width.
2. Delete the unnecessary white spaces and tabulators, and add white spaces to align numbers to the right.
After that this spreadsheet will look like this:

13765549 10604022 197813 | 1 Feb 2015 | 14:54:24 UTC | 1 Feb 2015 | 15:12:24 UTC Terminé et validé 1,061.62  7,840.83 8,721.16 OpenCL for AMD GPUs v9.15 (mt)
13765548 10604021 197813 | 1 Feb 2015 | 16:07:56 UTC | 1 Feb 2015 | 18:13:03 UTC Terminé et validé 1,076.36  7,901.26   237.89 OpenCL for AMD GPUs v9.15 (mt)
13765546 10604019 197813 | 1 Feb 2015 | 15:50:46 UTC | 1 Feb 2015 | 17:54:40 UTC Terminé et validé 1,133.30  8,401.45   232.80 OpenCL for AMD GPUs v9.15 (mt)
13765530 10604003 197813 | 1 Feb 2015 | 15:12:24 UTC | 1 Feb 2015 | 17:19:47 UTC Terminé et validé 1,456.66 10,339.33   277.93 OpenCL for AMD GPUs v9.15 (mt)
13765525 10603998 197813 | 1 Feb 2015 | 14:56:11 UTC | 1 Feb 2015 | 16:07:56 UTC Terminé et validé 1,029.07  7,702.50   152.70 OpenCL for AMD GPUs v9.15 (mt)
13765523 10603996 197813 | 1 Feb 2015 | 14:57:24 UTC | 1 Feb 2015 | 16:38:51 UTC Terminé et validé   921.62  6,911.97   150.75 OpenCL for AMD GPUs v9.15 (mt)
13765522 10603995 197813 | 1 Feb 2015 | 14:56:47 UTC | 1 Feb 2015 | 16:23:26 UTC Terminé et validé   930.95  6,987.43   146.14 OpenCL for AMD GPUs v9.15 (mt)
13765498 10603971 197813 | 1 Feb 2015 | 15:03:01 UTC | 1 Feb 2015 | 16:55:29 UTC Terminé et validé   995.72  7,425.57   174.81 OpenCL for AMD GPUs v9.15 (mt)
13765494 10603967 197813 | 1 Feb 2015 | 13:32:03 UTC | 1 Feb 2015 | 14:52:31 UTC Terminé et validé   998.30  7,420.52 8,201.05 OpenCL for AMD GPUs v9.15 (mt)
13765493 10603966 197813 | 1 Feb 2015 | 13:32:03 UTC | 1 Feb 2015 | 14:52:31 UTC Terminé et validé   989.75  7,359.32 8,130.77 OpenCL for AMD GPUs v9.15 (mt)
13765492 10603965 197813 | 1 Feb 2015 | 13:32:03 UTC | 1 Feb 2015 | 14:52:31 UTC Terminé et validé   996.33  7,397.32 8,184.82 OpenCL for AMD GPUs v9.15 (mt)

This is still too wide, also we don't know the type of data contained in each column, but there are constant columns, and columns containing irrelevant data, let's delete them, and let's name the columns:

  TASK   |   WORK   | Run Time |  CPU Time | Credits
13765549 | 10604022 | 1,061.62 |  7,840.83 | 8,721.16
13765548 | 10604021 | 1,076.36 |  7,901.26 |   237.89
13765546 | 10604019 | 1,133.30 |  8,401.45 |   232.80
13765530 | 10604003 | 1,456.66 | 10,339.33 |   277.93
13765525 | 10603998 | 1,029.07 |  7,702.50 |   152.70
13765523 | 10603996 |   921.62 |  6,911.97 |   150.75
13765522 | 10603995 |   930.95 |  6,987.43 |   146.14
13765498 | 10603971 |   995.72 |  7,425.57 |   174.81
13765494 | 10603967 |   998.30 |  7,420.52 | 8,201.05
13765493 | 10603966 |   989.75 |  7,359.32 | 8,130.77
13765492 | 10603965 |   996.33 |  7,397.32 | 8,184.82

Well, that's a self-explanatory spreadsheet.
If the constant data is relevant, you can include them before or after the spreadsheet.

Do you expect that in the future ATI app will work exactly like app for NV? I mean low CPU usage, high GPU usage, multi GPUs?

No, it's a very different code - it's always going to need both.

Matt

I'm seeing some unexplainable discrepencies in points attributions ...

The credit allocation still isn't addressed yet.

Richard,

There is the website (or local) preference "On multiprocessors, use at most xxx% of the processors", but this controls how many CPU cores BOINC can use as a whole, across all projects

I was under the impression that that's per project. (Otherwise, ¿how does that setting work when a client's attached to multiple projects?). In fact, I'm pretty sure changing that setting that will do the job, as I'm seeing jobs go out with odd numbers of cores assigned.

Matt

No, it's a very different code - it's always going to need both.

Matt

Well... That's bad. My ATI's will be supporting folding@home :(

Richard,

There is the website (or local) preference "On multiprocessors, use at most xxx% of the processors", but this controls how many CPU cores BOINC can use as a whole, across all projects

I was under the impression that that's per project. (Otherwise, ¿how does that setting work when a client's attached to multiple projects?). In fact, I'm pretty sure changing that setting that will do the job, as I'm seeing jobs go out with odd numbers of cores assigned.

Matt

Sorry MJH, but you are wrong on how that setting works.

Sorry MJH, but you are wrong on how that setting works.

So how does the client resolve conflicting settings from different projects?