GPU Client - News From Stanford

Associate
Joined
3 Jul 2004
Posts
866
Location
Helena, Montana
Is this old news or "New - News"

The primary issue now appears to be the CPU use of the GPU core. Due to how graphics drivers work in Windows, the CPU must poll to see if the GPU has completed. This polling is very CPU intensive (as the GPU does complete its work fairly quickly). We are working on a fix to this, but it is also likely that future GPU cores may use CPU power for scientific calcualtions which cannot be run on the GPU.

Thus, we are asking donors who run the GPU core to leave some CPU power (~1 core) available for GPU's to use. We need to compensate donors for this additional use of resources, so the points have been increased. The very idea of a GPU core and GPU software is new, so we are still working out what's the best way to handle these issues, but in general, we will of course award points based on the hardware used -- more hardware used, more points. As we develop the GPU core, the points may need to be changed (possibly up, especially if more CPU is used, possibly down if essentially no CPU is needed).
Looks like all the GPU WU's are now at 330 Points. {2711, 21, 23 & 25}
.
 
Excellent stuff - well that works out to 660 PPD as the benchmark so it should go some way to making up for the loss of a core, it's about double the max PPD I've seen on my X2 for any WU so far so seems fair enough really :D
 
rich99million said:
... ... ... it's about double the max PPD I've seen on my X2 for any WU so far so seems fair enough really :D
That makes it a No-Brainer Uncle Rich .. Sell the X2 rig for a C2D & X1900 :D

Dated as today, and since Stanford are what, 8 hours behind kinda suggests its new
I should have clarified my quesiton a bit more obviously :D
.
 
SiriusB said:
you get 660 PPD!? :O

Why does my graphics card take 13 hours a WU :(

SiriusB


Because...

a) thats not what he "is" getting, he is working out the new PPD based on the article

and

b) because you are still crunching on the old PPD for the GPU client

hope that helps dude ;)
 
Regardless of the amount of PPD I am getting, a points increase doesn't mean my WUs take any less time. So I wont be getting the full 660PPD.

Not sure where this benchmark comes from either... not seen anything "official" that suggests a WU takes 12 hours. So I can only assume the benchmark is rich's own.

:p

SiriusB
 
SiriusB said:
Regardless of the amount of PPD I am getting, a points increase doesn't mean my WUs take any less time. So I wont be getting the full 660PPD.

Not sure where this benchmark comes from either... not seen anything "official" that suggests a WU takes 12 hours. So I can only assume the benchmark is rich's own.

:p

SiriusB

Well I get exactly the same as you unclocked, so I would say an X1900 xt is around 12 to 13 hours. If I clock mine a little it gets closer to 12 hours around the same as a xt-x but I can't be bothered as it makes no difference in games. If the increase to a normal WU PPD score is 50% you will get around that figure, give or take.
 
The 660 PPD figure is based on the 440 PPD figure which was given as the benchmark result when Stanford first decided what points they would give for the GPU work. I guess that means their benchmark machine with a GPU gets a WU done in about 12hrs.
I'm not sure exactly what model of card it is but it will be on a dedicated machine which is doing nothing but folding, I'm not surprised if your machine which you're trying to use for other stuff won't reach the 12 hr/wu target.

I don't have my X1900XT yet - but it shouldn't be long now, just waiting for payment to clear :cool:
 
This polling is very CPU intensive

I had been meaning to ask about this, from the other thread people have been saying the GPU has been taking up 50% of a core on a core2duo. Now in CPU terms thats about the same as my AMD64 3000 at home, does that mean if i run GPU on it the CPU would be 100% utilised? and what happens if people try to run it on less powerfull CPUs (i'm thinking early p4's/celerons are probably the slowest CPUs that could be run due to GPU folding needing a PCI-E slot), would the GPU folding be limited by the CPU power.

Unfortunatly i can't join in the GPU folding fun yet, not until they release it for my X1800xt.
 
br83taylor said:
I had been meaning to ask about this, from the other thread people have been saying the GPU has been taking up 50% of a core on a core2duo. Now in CPU terms thats about the same as my AMD64 3000 at home, does that mean if i run GPU on it the CPU would be 100% utilised? and what happens if people try to run it on less powerfull CPUs (i'm thinking early p4's/celerons are probably the slowest CPUs that could be run due to GPU folding needing a PCI-E slot), would the GPU folding be limited by the CPU power.

Unfortunatly i can't join in the GPU folding fun yet, not until they release it for my X1800xt.
It's nothing to do with the power of the CPU - it requires one core (in the case of a core2duo that's 50% of the CPU) almost continuously to poll the GPU to see if it's completed its calculation so it can get some more for it.
But since your AMD64 3000 is a single-core CPU then yes there won't be much left for a standard folding client to be run on the CPU - that's the reasoning behind increasing the GPU points an extra 50% to make up for losing a normal folding client from your machine.

From the sounds of it they will make a client which uses both the GPU and CPU for actual work and minimise any waste :)
 
rich99million said:
It's nothing to do with the power of the CPU - it requires one core (in the case of a core2duo that's 50% of the CPU) almost continuously to poll the GPU to see if it's completed its calculation so it can get some more for it.
But since your AMD64 3000 is a single-core CPU then yes there won't be much left for a standard folding client to be run on the CPU - that's the reasoning behind increasing the GPU points an extra 50% to make up for losing a normal folding client from your machine.

From the sounds of it they will make a client which uses both the GPU and CPU for actual work and minimise any waste :)

What he said!

I tried to run both my CPU clients as normal, but the way Windows works it basically assigned half of each core to the GPU client. Not really a good idea as then both CPU clients only get 25% of your CPU and doesn't get much done.

It is far better to run only 1 CPU client on a dual core machine and none at all on a single core machine. That is why the GPU WUs get more points, in recognition of the fact anyone running a GPU client can't run a CPU client [or multiple clients with a dual core system].

If you read the news announcement again you will see they are also working on a client that uses both the CPU and GPU. The idea is that anything the GPU can't do, for whatever reason, the CPU can do it. This I imagine would pave the way for some very tasty WUs :D

SiriusB
 
SiriusB said:
What he said!

I tried to run both my CPU clients as normal, but the way Windows works it basically assigned half of each core to the GPU client. Not really a good idea as then both CPU clients only get 25% of your CPU and doesn't get much done.

It is far better to run only 1 CPU client on a dual core machine and none at all on a single core machine. That is why the GPU WUs get more points, in recognition of the fact anyone running a GPU client can't run a CPU client [or multiple clients with a dual core system].

If you read the news announcement again you will see they are also working on a client that uses both the CPU and GPU. The idea is that anything the GPU can't do, for whatever reason, the CPU can do it. This I imagine would pave the way for some very tasty WUs :D

SiriusB

Nah! Running 2 CPU clients = better because it scales nicely, plus it gets whatever is left over from Core 1 ;)
 
Concorde Rules said:
Nah! Running 2 CPU clients = better because it scales nicely, plus it gets whatever is left over from Core 1 ;)

I don't agree. I had a 153 pointer with an ETA of several days on one core and a 600 pointer taking a day longer than usual on the other. I would rather have one core going full steam ahead than two taking ages.

SiriusB
 
SiriusB said:
I would rather have one core going full steam ahead than two taking ages.
..and so would the science

I'm not sure what I'll do yet - I should get my X1900XT tomorrow if RM do their Special Delivery thang - I'll possibly try a few different ways as I'll have a WU or two to finish off anyway :)
 
Excellent purchase, sah!

Question: Did you buy an X1900 for gaming or just for folding? :D

I wonder if ATi will use Folding@Home as a marketing angle once it takes off properly?

"Our range of graphics cards have the power to cure diseases"

SiriusB
 
Back
Top Bottom