GPU Clients eating CPU cores

[Deathlizard]

System I'm testing this on:

MB: ASUS Crosshair IV Formula 890FX Chipset
CPU: AMD Phenom X6 1090T
Vid: 2 x XFX 5970 Black Edition using Catalyst 10.11 Accelerated Parallel Processing Edition
Nothing currently OC'd right now. Ventilation and temps look good.

I'm using the FAH GPU Tracker to handle all the clients and I'm new to F@H so I probably got something set wrong.

When I enable the 4 GPU clients, I notice that CPU usage maxes out 1 core per GPU client. Looking through the Catalyst Control Center, I also notice that the GPU's do not push to 100%, but hover around 70-80%. I noticed that if I drop two of the GPU clients, GPU Usage will climb to 96%, but of course that cuts PPD in half so that's not a good option and does nothing for the CPU Usage.

Setting the processor affinity with Task Manager to one core per GPU client gets the GPU usage to 85-90% but still not maxing out the GPU's and maxes out 4 CPU cores (5 if you count the system overhead) that I could be using on the SMP client.

Any suggestions?

 

[sirmonkey1985]

you have 2 options. not run the gpu clients or try this..

http://hardforum.com/showthread.php?t=1419852
(ignore the 9.4 drivers being needed)

 

[Tobit]

Deathlizard, to add to what sirmonkey said, the current ATI client performs "terribad" on current, and previous, generation GPUs. It hasn't been updated in a long time. A new client is coming, currently in Alpha, but just when is hard to say. There are optimazitions one can do to lessen the impact on CPU clients but current ATI GPUs will never come close to the same performance as NV GPUs with the current clients. The tweaks in the linked guide are the best you can do for now.

 

[lehmann]

Just run the smp client

 

[Deathlizard]

you have 2 options. not run the gpu clients or try this..

http://hardforum.com/showthread.php?t=1419852
(ignore the 9.4 drivers being needed)

Tried those settings with the same CPU results, although the PPD did rise a bit.

If it's a client issue, then I can deal with it until the next client comes out. the PPD from the GPU clients is still far better than from the SMP client alone (getting roughly 10000-12000PPD from them). So I'll just run the 4 GPU clients.