I am processing GPU tasks on multiple GPU's and I also wanted to process CPU tasks. However, I received Separation tasks for the CPU which are better processed with the GPU. I there any way to limit the CPU processing to only tasks that specifically require processing this way, such as N-Body?

Create two different venues. One with cpu only and one with gpu only. Then run two BOINC clients with each one crunching either the cpu or gpu venue.

---

Thank you

How did you decide that NBody are better for CPU than Separation?

Pointswise Separation tasks grants you more credits: 227 credits for about 1,5 hrs of 1 CPU
While Nbody 8 CPUs for 15 min only 36 credits,

That's actually quite frustrating 8*15=120 min CPU time for only 36 credits =( very very conservative

---

How did you decide that NBody are better for CPU than Separation?

Pointswise Separation tasks grants you more credits: 227 credits for about 1,5 hrs of 1 CPU
While Nbody 8 CPUs for 15 min only 36 credits,

That's actually quite frustrating 8*15=120 min CPU time for only 36 credits =( very very conservative

But if you run 6 tasks in the same 1 1/2 hours you will get 216 credits

But if you run 6 tasks in the same 1 1/2 hours you will get 216 credits

Don't forget core count!

NBody: 6 tasks Ñ… 1 set of 8 CPUs yield 216 in ~1,5 hrs
Separation: 1 task * 8 CPUs yield 1816 in a same ~1,5 hrs

---

How did you decide that NBody are better for CPU than Separation?

Pointswise Separation tasks grants you more credits: 227 credits for about 1,5 hrs of 1 CPU
While Nbody 8 CPUs for 15 min only 36 credits,

That's actually quite frustrating 8*15=120 min CPU time for only 36 credits =( very very conservative

I think he compared the time of the separation tasks when processed on GPU vs CPU.

Separation on GPU: 12 - 300 seconds (depending on the gpu)
Separation on CPU: 5400 seconds

less than 300 seconds < 5400 seconds --> so it is not efficient to process separation on CPU

As NBody can only be processed on the CPU he doesn't want to block his CPU with Separation units.

If you use a GPU the you don't care about the small amount of credits you can make with a CPU.
(Separation: 1 task * 8 CPUs yield 1816 in a same ~1,5 hrs --> separation on Radeon VII: 12 Secs/task --> 300 Tasks per hour --> 450 tasks per 1.5 hours = 68,100 credits)

So if you want to maximize scientific output (not credits) for the project you should do separation on GPU and NBody on CPU. I think this was the intention of Bill Eckard. If yes, I agree with him :-)