• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

FAH GPU2 client scaling with GTX 280

C

Cutthroat

Golden Member
I was just thinking...

I don't think the GPU2 client is using 100% of all the cores of my GTX 280. I think it should be capable of doing more than 2x the ppd as a 8800GTX.

This is very familiar, I just went through this when I changed from a dual to a quad CPU. With a E6400 and the SMP client I was capable of about 2000ppd, but when I switched to a quad it was only capable of 3000ppd. That is until I started running two SMP clients and then I was capable of more than 4000ppd. So we know the SMP client does not use 100% of all 4 cores even though monitoring programs show it does.

So a 8800GTX has 128 cores and a GTX 280 has 240, the core and shader speed is similar. But the GTX 280 only does 50% more ppd at default clocks. Seems like the same issue as the SMP client to me.

I have no idea how to test this theory though because I cannot run 2 GPU2 clients at the same time.
 
Maybe i'm not understanding you..but why are you comparing CPU and GPU? The 2 are very different in PPD output. The speed change on a GPU vs a CPU should be about %50 more vs the %10 (roughly) on the CPU you showed.
 
I'm not trying to compare CPU & GPU, just using CPU and the SMP client as an example.

What I'm saying is with twice as many cores on the GTX 280, it should be twice as fast as a 8800GTX.
 
I can't address your GPU2 scaling question but I can confirm that windows task manager does not accurately show cpu usage for 1xSMP on a Q6600.

In linux, 1xSMP on a Q6600 shows 60-70% cpu usage. 2xSMP pushes cpu usage over 90%.
 
Interesting new utility: Cuda-Z

It doesn't show GPU usage (yet?) but it does show memory bandwith which is part of the equation.
 
Yes and no - there's a reason for it. Right now most of the Work Units are simply too small to feed a GT200. The same goes for ATI 4000 series cards. The devs would have also focused on optimizing (and testing on) the well-established 8000 products first.

Right now everything on the GPU2 project list is 544 or 576 atoms. Large WUs with 1200+ atoms and improved cores are in the works.
 
You must log in or register to reply here.

TRENDING THREADS

Back
Top