Reducing 1 CPU with one GPU

maeax
maeax
Joined: 6 Aug 12
Posts: 21
Credit: 1423623213
RAC: 2509345
Topic 219495

When the CPU is less than 1 (for example 0.998) a other Project is possible to run in Boinc (7.14.2) parallel.

Splitting the GPU with CPU as the parameter in the Preferences shown is difficult therefore.

 

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 5872
Credit: 117589673152
RAC: 35203889

If you need the ability to

If you need the ability to control both fractional use of CPU and fractional use of GPU separately, it already exists.  Please see the instructions for Project-level Configuration on the BOINC website.

Please be aware that you could actually make the crunching efficiency worse by using these parameters incorrectly.  Also be aware that if you use the mechanism described on the BOINC website, it is not a simple matter to return to using the Project supplied GPU utilization factor if you decide you don't like the change.  Essentially, you have to reset the project in order to return to the former method of control.

If you explain the details of what you want to achieve, and the hardware you wish to use, somebody will be able to give you proper advice on the best way to do it.

Cheers,
Gary.

TiddlesHeisenberg
TiddlesHeisenberg
Joined: 10 Mar 15
Posts: 2
Credit: 17531824
RAC: 0

My experience with the latest

My experience with the latest Intel GPU tasks is that the notional CPU/GPU job description doesn't necessarily entirely describe the way that resources will be used.

Like:  hsgamma_FGRPB1G_1.22_windows_x86_64__FGRPopencl-intel_gpu.exe

For example, many GPU jobs effectively require no real CPU input - often running at 0.4% of a core while the Intel GPU runs flat out 100%.

This means, if you are running multiple Einstein tasks or other background processes (in my case often some Folding@Home tasks) - on a 4-core Intel i5-4430 CPU with built-in GPU , if you set 2 cores for Folding@Home & 50% of CPU (ie. 2 cores) for Einstein, the resources used will be:  2 cores for Folding@Home + 1 core running a non-GPU Einstein task + the built-in GPU running an Einstein GPU task & its owning task only using ~ 0.4% of a core (0.1% of the CPU) = total ~ 75% of CPU.

Stepping Folding@Home up to 3 cores then works fine:  3 cores for Folding@Home + 1 core running a non-GPU Einstein task + the built-in GPU running an Einstein GPU task & its owning task only using ~ 0.4% of a core (0.1% of the CPU) = total ~ 100% of CPU.

The reason I say the actual usage doesn't match the task descriptions is that the Einstein GPU tasks I have received tend to say 0.5 CPU + 1 GPU or 1 CPU + 1 GPU while in practice either tends to require almost no CPU core time to run.

The only difference on an i5-4430 between 0.5 CPU + 1 GPU vs. 1 CPU + 1 GPU is that a 1 GPU task makes screen update slow & rather impractical for word-processing or spreadsheet input whereas a 0.5 GPU task seems to have enough screen-update leeway to be bearable.

LOW-POWER

On a low-power Intel processor with built-in GPU the results are rather different (eg. m3-7y30 - a 2-core, 2-hyperthread core CPU [effectively appearing like a 4-core CPU] with built-in GPU).

A 0.5 GPU task seems to run pretty well & you can run 4 other cores to use ~ 100% CPU while GPU may run ~ 25%.

Running a 1 GPU task does not seem to work at all well - total CPU usage tends to be limited down to ~50% while GPU use is ~100%.  I presume the CPU-limiting is the usual low-power Intel CPU approach of reducing the CPU processing to keep the CPU at a sensible temperature.

My conclusion is that it would be handy to be able to specify 0.5 GPU tasks are ok but 1 GPU tasks are not practical but at the moment I have both types download - ideally I only want the viable ones with 0.5 GPU.

Is it possible to specify 0.5 GPU tasks only?

Thanks

AG

Holmis
Joined: 4 Jan 05
Posts: 1118
Credit: 1055935564
RAC: 0

TiddlesHeisenberg skrev:Is it

TiddlesHeisenberg wrote:
Is it possible to specify 0.5 GPU tasks only?

Sure there is, but first we need to establish what applications or searches correspond to each of your descriptions.
There are 2 different searches that can run on a Intel GPU, "Binary Radio Pulsar Search (Arecibo)" and "Gamma-ray pulsar binary search #1 (GPU)". So which of them corresponds to your 0.5 GPU and 1 GPU description?

When this is cleared up then go to your project preferences and deselect the search you don't want. To use different settings for different computers then use the "Preference set:" at top and choose one of the other sets, then go to your computer list and select the computer to change and at the bottom change it's location to match the new set.

maeax
maeax
Joined: 6 Aug 12
Posts: 21
Credit: 1423623213
RAC: 2509345

Binary Radio Pulsar Search

Binary Radio Pulsar Search (Arecibo) v1.34 () windows_x86_64

have now a Intel-GPU Beta with 0.5 CPU and 1 GPU.

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.