Bill's Process Manager issue with GPU clients

[APOLLO]

I'm trying out different affinity tools and it seems that Bill's Process Manager is easiest and works well with most of my systems. I've been using it on one of my Ati GPU systems since summer. Now, I'm attempting to install it on the rest of my farm since I'm most familiar with it and it works fine except for one system.

It seems that no matter how high I prioritize the GPU processes on that particular system, the GPU clients stop working if I don't isolate them. I see barely no CPU utilization (under 5%) by the GPU clients. The only other CPU-intensive app I have running is a single VM, which has been configured to run as 'idle' in the .vmx file, and is reported as such by BPM.

So out of 5 systems, 4 are working fine with BPM and one isn't. If I don't isolate the GPU clients to a core, they stop working altogether on that system. Anyone familiar with BPM know what may be causing this or can suggest a possible solution? It appears as if the GPU clients are obstructed from access to the cores if I don't isolate them. Really weird why this system is behaving very differently from the others.

 

[Zero82z]

Have you tried using the option to manually set affinity and selecting all of the cores? That might do it.

 

[APOLLO]

I have the affinity already set to access all cores. Originally it was set to one core and then I changed it when I noticed that the priority level didn't make a difference. I restarted the system and I still see anomalies in the BPM process window where a GPU core will display as 'idle,' but the performance is full speed without core isolation. I think it is working OK for now and hope it will stay that way. I want to stick with one utility on all my systems.

 

[Zero82z]

That certainly is strange behaviour. Have you tried uninstalling and reinstalling Bill2's? If you haven't and it happens again, give that a shot.

 

[APOLLO]

That certainly is strange behaviour. Have you tried uninstalling and reinstalling Bill2's? If you haven't and it happens again, give that a shot.

I was going to reinstall and decided to restart the system one last time. After the restart it seems to be working OK for over an hour. I still find the reporting of one GPU core as 'idle' a little weird, so I might reinstall it anyway. We'll see how it goes.

Thanks for the suggestions. To date, I prefer BPM over the other affinity tools I tried. I wanted to get it installed on the rest of my systems and have been putting it off for months. Finally got around to it, LOL.

 

[nitrobass24]

Ok tell me everything that is in the system that is having trouble.

PSU
Mobo
CPU
GPUs Brand, model including onboard and non-folding gpus if applicable.

 

[APOLLO]

Ok tell me everything that is in the system that is having trouble.

PSU
Mobo
CPU
GPUs Brand, model including onboard and non-folding gpus if applicable.

Hey nitro,

The app seems to be working fine now and it has been over 12 hours. It is even reporting the GPU processes correctly unlike last night. So, if it continues working fine I'll leave it be.

This is an old dual Opteron system with four cores total. There are two BFG 8800GT cards and the motherboard is a Supermicro.

 

[nitrobass24]

Does it have onboard video?
Newer NV drivers were messing up GPGPU for me if it saw an ATI device in the system.

 

[APOLLO]

Does it have onboard video?
Newer NV drivers were messing up GPGPU for me if it saw an ATI device in the system.

No, this motherboard doesn't have any onboard video. I have some motherboards in my farm with onboard video (Ati and others) but these systems never had a problem. It's working fine now. If it acts up again, I'll try reinstalling the app.

 

[nitrobass24]

ok well glad you got it working.

 

[APOLLO]

ok well glad you got it working.

Actually I posted too soon.

I had to restart the system yesterday for some maintenance and it doesn't work anymore. I reinstalled the app and it still doesn't work. I'm at a loss to explain why this system is proving such a difficult case when I have 4 other systems flawlessly working with BPM...

Now, I'm back to isolating the GPU cores. I really don't want to do this since I'm losing a good part of a core for nothing more than running the GPU clients. Especially with the 1888-pt WUs that hardly use any CPU power.