What to use to only do MB on GPU and AP on CPU

[Helmchen]

I want to use the CPU for only AP, while my GPU should do MB tasks.

Is this possible? What files do i have to use, i can't find anything about it. Actually i use Raistmers V10 but i can't always babysit the PC so that it does what i want.

greetings
Holger

[KarVi]

Its possible to set this up.

The best way, would be to edit the app_info.xml, so that it only contains refrences to MB (setiathome_enhanced) rev 608 (603 could work to) pointing to the GPU/CUDA executable, and AP (astropulse_v5) rev 503, pointing to the AP_v5  executable.

This would make your client only do AP on the CPU and MB on the GPU, provided off course theres any work available.

[M_M]

I have Q9650 (quad-core) and GTX295 (2xGPU). WinXP 32bit. Most of the time, everything works OK, as expected.

However, I find Boinc still not to intelegent... Quite often lately, in cache there are dozens of setiathome_enhanced and astropulse_v5 tasks, and I find boinc crunching 6x astropulse (on 4 logical CPUs!!!), while GPUs being idle... Again, there are setiathome_enhanced tasks waiting to be running...

So I have manually to suspend appropriate numbers of AP units, and resume them later to get everything back in order!

[KarVi]

M_M:

The only way to do this and be sure its handled correctly, is to get the latest BOINC developement versions (6.6.17 or 6.6.18). These can handle CPU and GPU tasks individually, and makes sure such things don't happen. There is no need to change numbers of CPU's and so on, since BOINC then handles the allocation to CPU's and GPU's.

Since they are developement versions, they have other problems, but they work well enough on my CUDA system.

[Helmchen]

I think i got things sorted, but sadly couldn't actually test it as i don't get any MB WUs

BOINC 6.6.17 did the trick, while 6.6.18 did nothing but killing WUs

thank you

[Brutus]

I'm experiencing the same issue with the latest 6.6.20, 6.6.21 & now 6.6.23.

It works perfectly when you do NOT have Astropulse units.
I set Boinc to use 100% of the CPUs and it runs 4 instances of the Seti Enhanced application.  3 of them run on the CPU (Q6600).  I of them uses 0% CPU resources but it appears that CPU core feeds the GPU work unit (GTX260).
Works flawlessly.  4 units at a time, no issues.

However, as soon as I downloaded AP units, Boinc starts running 4 instances of AP and no GPU.

I found that by editing the cc_config.xml to 5 ncpus, I could get Boinc to run 4 AP units and 1 GPU unit. It sort of solves the problem but it slows the GPU units and makes the computer sluggish.  Only solution I have found is to suspend all but 3 AP units and to control it manually.  (don't use a cc_config.xml for this to work)  I get 3 AP and 1 GPU this way

What would be perfect would be 3 AP and 1 GPU units just like it does with the SETI Enhanced Units with out manual intervention.

The other alternative that I have considered but not tried yet is to set Boinc to 100% of the CPUs 75% of the time and then run 5 ncpus in the cc_config.xml.  That should run 4APs and 1 GPU unit and give some CPU resources to still run the PC but it will increase the CPU turn around time for a unit.  Overall should be the same.

Anyone have a better solution?

[Claggy]

See this thread Which version?
I've been running a similar app_info (to the one I've posted) since about Boinc 6.6.15, and always have 2 CPU WU's and one Cuda WU's processing.

Claggy

[Brutus]

Thanks,  Followed the guide and it seems to work now.

[The Borg]

Hi there,

Now to this forum, usually post over at anandtech - DC.

I would like to know how to get all 4 cores of my 2 x GTX 295's crunching data. So far I have tried SLI on and off, GPUGRID and SETI, but the most I can get is 2 WU's, and from waht I can tell, only one per card.

Any comments or suggestions please?

[Brutus]

I have 2 single GPU video cards so I can't tell you what to do on the GTX295s.  In various posts I've seen, I believe turning off SLI is a must.

As I stated, I followed the guide, pulled in the Astropulse executables etc then I modified the app_info.xml to use Raistmaister's modified cuda client rather than the std client.

I was originally using the team mod executable when things didn't work for me.  Along the way, I upgraded to Boinc 6.6.23, changed to files recommended by Claggy and finally modified the .xml file.  Somewhere along the way it started working.


By the way, ( this is just my opinion of course )
- those 2 cards have a TDP of 295 watts each, right??
- where I am, power is 11.4c/KWatt hr, thats 2 x .295 x 24 x .114 = $1.61 US$/day.  That's a nasty electric bill.

That's too much for me but I can't know that the math is right.  Let us know once you get it working..