Cpu hindering or helping w/ gpu

[Green]

If I start folding on my cpu, is it worth any drop in points I may incure on the gpu? I assume gpu folding does tax the cpu a little. I only have a 4600 in this machine. Anyways, on my mind as I drop 12 gigs of ram next week, because while f@h might not need it, a computer left on does need it.

[Mstenholm]

As long as you leave one core/thread for the GPU folding you will be good. I know for sure that using all threads on WCG and folding on a NVidia GPU will lose me more than 20 % PPD. If you want to have a more qualified answer we need to know what kind of hardware you plan to use.

[7im]

The Fah client automatically reserves one CPU thread to feed the gpu client. Folding on the remaining threads of the CPU has no impact on gpu folding.

[JonasTheMovie]

As long as you use "-1" 7im, right?

[7im]

Yes, and that is the default setting.

[DemonfangArun]

I'm going to assume that's an i5 chip, in which case you are correct in that -1 on the CPU slot will reserve 1 thread for the GPU, and use the remaining 3 threads for the CPU slot. More hardware info would be needed to determine how to best optimize your system, but I will leave some suggestions below.

AMD GPU: you should be fine with -1, as AMD GPU's don't need much CPU, and you should be fine with this
NVIDIA GPU: I would recommend reserving a CPU thread for the GPU, and another thread for daily tasks (web browsing, etc...). If this is a dedicated folding rig, then you should be fine running -1 in this scenario too.

[Green]

my point is I see a point loss of about 30000 when I run on cpu. Should I still run a cpu?

[DemonfangArun]

my point is I see a point loss of about 30000 when I run on cpu. Should I still run a cpu?

with only 2-3 cores available, it's probably not worth it. by my estimates that CPU slot, even optimized right, wouldn't do much. that said, if you want to fold on it, it probably won't hinder your GPU if you set the number of threads to use in the slot settings to whatever your system has minus 2

[Green]

30000 with 2 cpus

[DemonfangArun]

30000 with 2 cpus

Didn't know that was with that applied. In that case, I would say you are better off with no CPU folding. I can say this is the first instance I've come across where any CPU folding at all causes major GPU PPD reduction.

[JonasTheMovie]

I can see a massive drop on my GPU if I start NaCl on the CPU, was a bit surprised there. Thought that the GPU client would have higher priority to the CPU than the NaCl.

[Joe_H]

Running NaCl at which setting? If you are running it at 100%, then it is going to take all CPU cores. The NaCl client is totally separate from the FAH 7.4.4 client and does not have any interconnection to reserve a core for GPU folding. The NaCl process runs at the priority of the user process using Chrome.

The FAH client will by default reserve a core for the GPU slot unless overridden. It manages CPU and GPU folding slots setup within it, but can not do the same for CPU folding for another client, whether that is the NaCl one or the WCG client mentioned in Mstenholm's post.

[Sn1ken]

It is worth noting the big difference when folding Nvidia vs AMD.

Nvidia is requiring a lot from the CPU. And new assignments seems to push the limit even further, and this becomes especially clear when using 980 TI.

In some cases the drop in PPD when folding CPU and GPU is absolutely not worth it when PPD is a factor. And somehow PPD is the measure that shows the scientific value of the contribution. What is the purpose of using more power to accomplish less science?

On the other hand, folding AMD will give another picture. But still the PPD pr watt is far greater on GPU vs CPU.

[7im]

When you are so far up the points curve on the exponential QRB scale with very fast hardware like the 980, even the very smallest event has an extremely significant impact on the bonus points. Very much disproportional to the actual difference in the science getting done, IMO.

[bruce]

When Stanford designed NaCl, they assumed it would be the only FAH program running. That's not to say that you can't run both, but their plan didn't include accommodating any possible interactions.

The same is fundamentally true for the GPU core and for the CPU core .. except for the one accommodation for the fact that using all your CPUs will over-commit the CPU, causing unplanned delays. DemonfangArun alludes to that fact with his comments about sometimes reserving a second CPU for a computer that's frequently used for other things. When managing multiple clients, things get pretty complicated pretty rapidly.