Next gen GPU client in open beta

jaric · 13 Apr 2008 at 00:47

I managed to get 1900ppd with my 3850 and Q6600 @ 3.7. One CPU core @ 100% and GPU @ around 90%. Not bad but not enough to switch from running LinuxSMP in a VM ... yet.

NickK · 13 Apr 2008 at 12:05

ghost101 said:
Is a sign of cpu limitation that the gpu core doesnt get fully loaded? I only get around 74% load on the gpu while one of my cpu cores is fully loaded.

edit: overclocking the cpu did the trick. gpu usage now is ~95%. I get 2095ppd with the 2799 wu doing each frame in 40s. hd3850(256mb) clocked at 864/999.

more like memory bus speed too.

Pilgrim57 · 14 Apr 2008 at 12:42

jaric said:
I managed to get 1900ppd with my 3850 and Q6600 @ 3.7. One CPU core @ 100% and GPU @ around 90%. Not bad but not enough to switch from running LinuxSMP in a VM ... yet.

Interesting..........Linux SMP for the win.

Yeggstry · 14 Apr 2008 at 13:57

How on earth are you getting 1900 ppd on one core?

I'm running at 3.4 and getting about 1300 on the diskless install

shadowscotland · 14 Apr 2008 at 14:13

two vm's with smp's on each at a guess so all four cores are used

Yeggstry · 14 Apr 2008 at 14:35

But jaric was saying that its not worth moving to use the GPU client if under linux. This implies that one core is getting 1900PPD, otherwise it would be worth moving to this. Replacing one smp client with the GPU client would probably give about 200-400 PPD extra, but would significantly increase the power used per point earnt

jaric · 14 Apr 2008 at 18:29

To run the GPU client I had to loose 1 VM, otherwise the second VM would be running on a single core.

4 Linux SMP clients in 2 VMs = 5200ppd

2 Linux SMP clients in 1 VM + GPU client + Standard client = 4800ppd

OK, not a massive difference but as you say it's still a 400ppd drop for an increase in power consumption.

Once the larger GPU WUs appear it might be worth switching over, but we'll have to wait and see.