Really back on track now

i7GTX550Ti8Gb · **Really back on track now -** 11-23-2011, 20:23 | posts: 88 | Location: Belgium

ok, afetr a while of messing around with comp issues due to need for processor cycles

I am back on track, installed me an antec H20 920 cooler, so now i can fold with full smp slots instead of only 2 or 4, and the gpu is cooled better now aswell, due to imporved vent in system case.

Hope to catch up all that i've missed

iancook221188 · 11-23-2011, 23:27 | posts: 1,246 | Location: uk

good good welcome back

k1net1cs · 11-24-2011, 02:35 | posts: 3,327

Nice; we need more Nvidia owners on GPU folding.
GPU folding using AMD cards wastes CPU cores.

Kaktus · 11-24-2011, 06:00 | posts: 303 | Location: South Africa

Welcome back...

PantherX · 11-25-2011, 18:22 | posts: 1,007 | Location: F@H Forums

Welcome back pal

i7GTX550Ti8Gb · 11-25-2011, 19:12 | posts: 88 | Location: Belgium

ty ty

i do have a question to you all though, i noticed i can set smp slot to 1/3/5/7 now instead of 2/4/6/8 what do u guys consider being the best? set smp at -1 so it takes all "8" cores? or manually put at 7 so i got some spare cpu power for gpu?

Just a q

PantherX · 11-25-2011, 19:27 | posts: 1,007 | Location: F@H Forums

I would avoid 7 since there are issues with it being prime and I know that some projects don't like 7. My suggestion is to use 6 as it is the best case and that is currently what I am using as my system isn't dedicated and it works like a charm

i7GTX550Ti8Gb · 11-25-2011, 20:37 | posts: 88 | Location: Belgium

Quote:

Originally Posted by PantherX

I would avoid 7 since there are issues with it being prime and I know that some projects don't like 7. My suggestion is to use 6 as it is the best case and that is currently what I am using as my system isn't dedicated and it works like a charm

ok, using 6 for now, does it make a lot of difference between 8 or 6?

just another q

Agent-A01 · 11-25-2011, 20:49 | posts: 5,627 | Location: USA

Nice name lol

k1net1cs · 11-26-2011, 06:49 | posts: 3,327

Quote:

Originally Posted by i7GTX550Ti8Gb

ok, using 6 for now, does it make a lot of difference between 8 or 6?

If you're using Nvidia GPU, you don't need to set a spare CPU core for GPU folding.
Sparing a core only applies to folding with AMD cards.
And setting to prime or odd numbers would usually just make a core using the nearest lower even number of cores.

PantherX · 11-26-2011, 07:38 | posts: 1,007 | Location: F@H Forums

On my system which isn't dedicated, I don't find much difference. The reason is that the folding process will wait for the slowest thread to finish and since majority of my applications use 1/2 CPUs, this is the best case for my system.

Do note that if you set the GPU (for Nvidia only) to higher and CPU to idle, then you don't really need "spare CPU Cycles" as the OS will sort it out.

sykozis · 11-26-2011, 16:49 | posts: 13,493 | Location: US East Coast

Don't actually need to spare a CPU core with AMD/ATI cards. I used to run SMP+GPU just fine with my 4850. Due to thermal constraints, I can only run CPU or GPU now, not both.

i7GTX550Ti8Gb · 11-26-2011, 20:58 | posts: 88 | Location: Belgium

ok, ty for the good decent replies, will fold 100% on systemfor now, i'll c what happens, atm, seems i don't "loose" too much "power" of my system, i can still use it for the day to day tasks, if i use my main software applications, i'll just pause system, just to habe all the power i need.

again, TY to all

k1net1cs · 11-27-2011, 07:19 | posts: 3,327

Quote:

Originally Posted by sykozis

Don't actually need to spare a CPU core with AMD/ATI cards.

Because when I ran v7 with just GPU folding with 5650 Mobility, it uses a CPU core 100% (a.k.a. 25% on a 2+2 cores like an i5-520M).
It's either inherent to GPU2 clients or OpenCL implementation in folding clients in general.

SMP+GPU would of course run fine because in those setup the GPU client is usually being set with a priority higher than the SMP client, and the SMP client doesn't really lose much from 'forfeiting' a single core to the GPU client.
You need to run the GPU client alone to see its CPU usage.

In any case, it's what's been explained over at F@H forum.
AMD cards will use one CPU core along when folding, while Nvidia cards won't.
Maybe it's OpenCL vs CUDA, or sloppy OpenCL implementation; dunno.

sykozis · 11-27-2011, 18:16 | posts: 13,493 | Location: US East Coast

My GTX460 uses a noticable amount of CPU time when folding.....but, it doesn't prohibit the use of all cores when running SMP+GPU. OpenCL makes use of CPU cores regardless of what GPU you're using, as does CUDA....

k1net1cs · 11-27-2011, 19:49 | posts: 3,327

Quote:

Originally Posted by sykozis

My GTX460 uses a noticable amount of CPU time when folding.....

And that is exactly what I was talking about; GPU folding also requires some CPU cycles.
It's just that with AMD cards the impact is more severe.

Quote:

Originally Posted by sykozis

... but, it doesn't prohibit the use of all cores when running SMP+GPU.

And no one said it does.
I simply stated that on SMP+GPU setup, you'd likely to forfeit a core to the GPU client, especially with AMD cards.
Here's something from F@H forum regarding GPU folding CPU usage.
The more GPU clients you use (for systems with multiple cards), the more you lose CPU cycles to GPU folding.

This is also why with SMP+GPU it is recommended to put the GPU client's priority a level above the SMP client's, otherwise the SMP client could starve the GPU client out of CPU cycles.

sykozis · *Last edited by sykozis; 11-27-2011 at 23:20.*

I get 13-15k PPD from my i7 870 without the GPU client running (keep in mind I have HT enabled). I get 10k PPD with the GPU client running....that's up to a 30% drop in PPD from running the GPU client. That's a bit more than "1/2 a core"....

k1net1cs · 11-28-2011, 02:44 | posts: 3,327

So, what exactly are you trying to point out here?

You said...

Quote:

Originally Posted by sykozis

Don't actually need to spare a CPU core with AMD/ATI cards.

But then, you said...

Quote:

Originally Posted by sykozis

OpenCL makes use of CPU cores regardless of what GPU you're using, as does CUDA....

...contrasting your previous statement.

Moreover, when you said...

Quote:

Originally Posted by sykozis

My GTX460 uses a noticable amount of CPU time when folding.....

...and I gave a link that had an approximate CPU utilization on a Fermi GPU folding...

Quote:

Originally Posted by sykozis

That's a bit more than "1/2 a core"....

...you said it like, "No, it's higher than that.".

So, what is it?
That GPU doesn't affect SMP on SMP+GPU, which you initially insisted, or that GPU does affect SMP on SMP+GPU, which I was always talking about?

sykozis · 11-28-2011, 02:56 | posts: 13,493 | Location: US East Coast

Never said that GPU folding didn't effect SMP folding. I simply said you don't need to "spare a core".

Also, some of the Fermi cards use more CPU time than others for some reason....at least that seems to be the general consensus... Unless you set process affinity to force OpenCL/CUDA and the drivers onto the same core, they'll run on whatever core Windows assigns them to. Either way, you'll lose some PPD to OpenCL/CUDA and the drivers. How much depends on how the drivers are written to function. But, in either case, there's no actual need to "spare a core" just to run the SMP and GPU clients together. Either way your PPD is going to decrease as a result, so might as well let the SMP and GPU apps use every bit they can get.

PantherX · 11-28-2011, 14:57 | posts: 1,007 | Location: F@H Forums

Since every setup is different, I would suggest that you try out the following:
SMP only = A PPD
SMP + GPU = B PPD

If B PPD > A PPD, then it is an overall gain. You can experiment with smp values 6 and 8 to see which gives the best balance between F@H and your needs. Best of luck and do report your findings