GPU folding won't run when CPU folding is running

[ho_k]

I am just wondering, does SMP option run 1 instances across multiple cores, or 1 instance per core?

If I upgrade to a current multi core cpu, say 1090T, would I still have the same problem where I have to choose either GPU or SMP, but not both?

[RETARD]

the SMP client utilises itself across multiple cores.
Because the processor has to lend some of its cycles to the GPU client when it runs .. it can upset the cpu and not leave enough processing power do do normal things.
As someone said before - on faster machine you can run both and it isnt as noticable because there are alot more resources to draw from. On a slower machine just use one or the other (which ever will produce better ppd) Its just a matter of making the best of what you got with out crippling the machine for when you want to use it.

So to answer your question ... upgrading to a 1090T you could run both the GPU & SMP with out the problems you face now

hope this helps

Originally Posted by ho_k

I am just wondering, does SMP option run 1 instances across multiple cores, or 1 instance per core?

If I upgrade to a current multi core cpu, say 1090T, would I still have the same problem where I have to choose either GPU or SMP, but not both?

[Belial669]

Originally Posted by ho_k

I am just wondering, does SMP option run 1 instances across multiple cores, or 1 instance per core?

If I upgrade to a current multi core cpu, say 1090T, would I still have the same problem where I have to choose either GPU or SMP, but not both?

Technically you one run SMP client, and when it starts it creates 1 processes per core available.
Eg with a 1090T if you were just running SMP it would create 6 processes.

Previously the way we freed up a core was to set core affinity so for example with a quad core, we'd set affinity so that when the SMP client created the processes, we'd assign the 4 processes created to only have affintiy (process on) on 3 cores only, hence leaving 1 core free so you could then run a GPU client as well as it would automatically use the free core for its processing requirements.

However with the current clients we can specifiy how many processes it starts so you can limit it to leave one core free, however in practice using the latest V7 client the new control centre pretty much takes care of the scheduling side but it does take a slight performance hit compared to a properly setup set of v6 clients

[EvilGenius]

Originally Posted by Fatmantan

As mentioned, you will not be able to run both CPU and GPU on a dual core cpu with any great success. Run GPU only.

Originally Posted by Fatmantan

/exits thread.

If you wont listen to our advice why keep asking?

MACHINE 1: Stick to GPU ONLY
MACHINE 2: Stick to CPU ONLY

Both machines arent specced high enough to run both at the same time.

IF you decided to stick the 460 with the Q6600 machine then you would be able to fold both at the same time.

WTF are you lot on about???

You CAN run GPU and CPU at the same time. Did anyone think to ask the OP to check on % of cpu used for the GPU process? How about a little trouble shooting before making ridiculous crap up?


If the OP is using the single core client, one instance on each core and a gpu client it should work. He perhaps needs to fiddle with core priorities, but and outright stop like this shouldn't happen.

If the OP has 2 instances of the SMP client on there, that's a problem. Unlikely, but..



ho_k, can you post up what versions of the client you're using and how you have them configured?

[leroyxx]

Yeah, I do believe that is what I posted Evil... im not sure but of late my posts seem to be skimmed over.. so this will probably not get noticed either

[ho_k]

Originally Posted by EvilGenius

WTF are you lot on about???

If the OP has 2 instances of the SMP client on there, that's a problem. Unlikely, but..

ho_k, can you post up what versions of the client you're using and how you have them configured?

I originally downloaded the GPU and CPU client x2 from FAH.
I set 2 instances of CPU (1 for each core) ID 1, 2. - Folding@home-Win32-x86-623
I set 1 instance of GPU ID 3. Folding@home-Win32-GPU_XP-641
They are all console version, default settings except I set them to my username, team 24, passkey, use big units, -advmethods.
I notice that when any 1 or 2 instances of CPU running, GPU stopped.

I then use the FAH tracker as suggested... setting up SMP and GPU / CPU and GPU / SMP 1 core and GPU combinations, same deal.

Just now, I played around with priorities... manually set the GPU to "below normal" and CPU to "low" in task manager. Still the same.

Also tried running 1 CPU instances on FAH tracker, and run the GPU on the console Folding@home-Win32-GPU_XP-641. Set the GPU to "low" instead of idle (hopefully have a higher priority). Still the same.

I am using Windows 7 Ultimate. A64 x2 4200+ @ 2.33ghz, with 2 gb ram and gtx460 1gb @ 865/1730/1025.

I can understand what others are saying about the need of a CPU to handle GPU load. But that can't explain why running on 1 of the 2 cores on the CPU and reserving 1 core for the GPU still stops the GPU running.

If you have good suggestions, please let me know what to do. The CPU isn't terribly fast. I suppose I can live without it running.

[EvilGenius]

You've quoted the 623 cpu client, that one has been superseded by the 6.30 client, dl link here:

http://www.stanford.edu/group/pandeg...-win32-SMP.zip

I know it says SMP client, but if you don't use the -smp flag it will just run standard units. Also I'd stick with normal work units. Quite possibly will make no difference to what units you actually receive, but I'd still save the big setting for a cpu with a little more stick.


Next thing, make sure you're putting this directly on C:\, ie C:\CPU1, C:\CPU2, C:\GPU etc.


As for core priorities, don't bother setting these in task manager, do it in the client config. Leave the cpu's on idle and set the GPU to Low. If this still isn't enough, grab winAFC and set a rule for the gpu core process to run at normal and see what happens.


If all this weirdness keeps up, run the GPU client by itself and see what sort of cpu usage you're getting.