V10/11 of modified SETI MB CUDA + opt AP package for full multi-GPU+CPU use

[elec999]

3/25/2009 12:36:25|SETI@home|Message from server: (Project has no jobs available)
Seems dried up. I need some cuda work.
Thank you

[elec999]

Raistmer do you recommend moving from Boinc 6.4.5
Thank you

[Raistmer]

Raistmer do you recommend moving from Boinc 6.4.5
Thank you

To recommend smth I wanna try it first. I still use 6.4.5, but it doesn't mean that this is the best way to go. Simple I didn't try 6.6.17 still.

[br83taylor]

Sorry if this has been asked before.

I am running V10a with the latest FFTW update. All was running well but then as the MB work has dried up on my machine, it has started to run 3 astropulse units on my dual cpu. Is this normal? I guess this is a side effect of the cc file stating 3 cpus?

http://setiathome.berkeley.edu/show_host_detail.php?hostid=3366055

[Josef W. Segur]

Sorry if this has been asked before.

I am running V10a with the latest FFTW update. All was running well but then as the MB work has dried up on my machine, it has started to run 3 astropulse units on my dual cpu. Is this normal? I guess this is a side effect of the cc file stating 3 cpus?

http://setiathome.berkeley.edu/show_host_detail.php?hostid=3366055

Yes it is normal, in fact that's the official purpose of that setting; to emulate more CPUs than the host actually has. Running 3 APs on a dual CPU system is somewhat less efficient than two, so you could temporarily change the cc_config. However, the project does have one mb_splitter running now, so you may get at least some GPU work.
                                                                       Joe

[br83taylor]

ok thamnks for clarifying that.

didn't bother changing the cc again.

however did managed to get a few MB units, however bonic didn't start them, carried on doing the three AP jobs.

i tried suspening one of the AP units, thinking it would go onto the next MB unit, but it didn't, it just tried to start the next AP unit??

ended up having to suspend 10 AP units before it got to the MB work unit and started that, which it did pass to GPU for processing.

[frigens]

Just out of curiosity, sorry if it's a stupid question. I'm kinda new to this.

I'm using BOINC v6.6.18 and NVIDIA driver v182.08 on WIndows XP x64. I have QX6700 and 3 GPUs (GTX295+8600GTS). I've set the <ncpus> option to 7 (4+3). and so far it run it seems to run ok but rather on a slow side.

I'm crunching on SETI@Home Beta and none of these tasks run on CUDA. All my SETI WUs are setiathome_enchanced 6.08, aren't these WUs supposed to be run by optimized CUDA app according to the app_info.xml? If so, then I might have a problem? I concurrently run GPUGRID project and it used all 3 CUDA tasks while SETI Beta taking 3 running slots but none show it it's utilizing CUDA GPU.

[Claggy]

Just out of curiosity, sorry if it's a stupid question. I'm kinda new to this.

I'm using BOINC v6.6.18 and NVIDIA driver v182.08 on WIndows XP x64. I have QX6700 and 3 GPUs (GTX295+8600GTS). I've set the <ncpus> option to 7 (4+3). and so far it run it seems to run ok but rather on a slow side.

I'm crunching on SETI@Home Beta and none of these tasks run on CUDA. All my SETI WUs are setiathome_enchanced 6.08, aren't these WUs supposed to be run by optimized CUDA app according to the app_info.xml? If so, then I might have a problem? I concurrently run GPUGRID project and it used all 3 CUDA tasks while SETI Beta taking 3 running slots but none show it it's utilizing CUDA GPU.

With Boinc 6.6.x you don't need an <ncpus> entry in a cc_config file, just remove the entry, it was for Boinc 6.4.x to fool it it into thinking it had an extra CPU, Boinc 6.6.x knows about GPU's and doesn't need the work around.

setiathome _enhanced_6.08 is the Cuda app, and please don't run optimised apps on Beta unless they are for special testing of new unrealeased apps,
I 've recently been matched up with this user's computer on Beta, he's running optimised Astropulse 5.00 r103 and trying to Validate against astropulse 5.04 (ie astropulse_v5) , Which isn't very helpful there, and a PM hasn't helped.
Run the optimised apps on the Main project, and visit this thread to get the special app_info for Boinc 6.6.1x.

Claggy

[Slawek]

Hi,
i need ask.. what the best build now ?  ( for X2 Athlon and GTX 260 ) ? ( Cuda computing priorytet )

?

[Morten]

Hi,

What is the setting for using both cores in a GTX 295?

I have one GTX 295 and using the default setting of 1 in number_of_gpus does not seem to utilize more than one core, as the computations equals a GTX 285. A setting of 2 will spawn 2 GPU executables, but one of those will have a full CPU utilized continously, so it's not working..

I have seen references to various tweaks and also references to both cores running, but not become any wiser as to what is the correct combo of settings to utilize both cores on the GTX 295.

Regards
Morten Ross

[Raistmer]

Sum core numbers (each CPU core and each GPU core) and enter it as ncpus in cc_config.
Main rule - each core should run its own app instance.

@Slawek
V10/11 CUDA MB build from this thread fastest one for CUDA for the moment.

[Morten]

Hi,

That's also what I did, which results in the second cuda-executable with constant 100% on one CPU.

4 CPU cores + 2 GPU cores (one GTX 295 card) => ncpus=6 .

I would expect the two cuda-apps to utilize an equal share on no less than 4% CPU each.

To me there seems to be a problem in the second cuda app's ability of feeding the second core, as it instead is using CPU.

Nvidia driver 182.08, Windows 7 x64..

Morten

[Raistmer]

Hi,

That's also what I did, which results in the second cuda-executable with constant 100% on one CPU.

4 CPU cores + 2 GPU cores (one GTX 295 card) => ncpus=6 .

I would expect the two cuda-apps to utilize an equal share on no less than 4% CPU each.

To me there seems to be a problem in the second cuda app's ability of feeding the second core, as it instead is using CPU.

Nvidia driver 182.08, Windows 7 x64..

Morten

Check how many GPUs your BOINC sees.
What app do you use? If you use V10 pack - edit number_of_GPUs file and set 2 instead 1 there.
And make sure you don't use any kind of SLI mode.

[Morten]

Hi,

This is what Boinc sees:
"31.03.2009 21:18:16      CUDA device: GeForce GTX 295 (driver version 18208, CUDA version 1.3, 896MB, est. 106GFLOPS)", which is correct, as there is only one physical card in system - with two GPU cores.

I use Raistmer's_opt_pack_V10a_x64_SSSE3_MB_CPU_GPU_team, as as said earlier - I've set the value to 2 in "number_of_GPUs" file.

Morten

[Raistmer]

Hi,

This is what Boinc sees:
"31.03.2009 21:18:16      CUDA device: GeForce GTX 295 (driver version 18208, CUDA version 1.3, 896MB, est. 106GFLOPS)", which is correct, as there is only one physical card in system - with two GPU cores.

I use Raistmer's_opt_pack_V10a_x64_SSSE3_MB_CPU_GPU_team, as as said earlier - I've set the value to 2 in "number_of_GPUs" file.

Morten

Search SETI main forum when it will return from outage. There was thread about multicore GPUs and how to enable both cores.