Folding PPD: CPU vs GPU

[verbal]

I'm sure this must have been discussed but I can't find a definitive answer.

As things stand at the moment do you get more points from the best GPU WU's or more points from 2 of the best CPU WU's.

In other words would a highly clocked core 2 duo get more PPD (running 2 instances) than a highly clocked X1900?

I've got no idea how close it is between the two. I've not been around on the forum much lately.

I know it's not normally a choice as people are running one c2d core and the gpu client. This tells me that the gpu client gives more PPD than a single c2d core... i guess.

 

[MGP]

The problem is that you can't guarantee what work you are going to get sent.

The GPU WUs are only one size, at 330 points, My [email protected] with X1950X-TX unclocked is doing a WU every 13.5 hours or so, so thats worth around 580 PPD give or take, but needs a CPU core too.

The remaining CPU core really varies dependent on the WU you get. The 600 pointer 2414 Gromacs have been worth up to 480 ppd. Other Gromacs and Amber Wus have been down as low as 180ppd.

On my reckoning yes you get more all the time doing the GPU, but it's probably not that significant if you could guarantee the 600 pointers.

 

[SiriusB]

I wouldn't be too sure about that!

If you could guarantee the 2414s all the time you would be getting ~480PPD per core, which comes to ~960PPD.

My X1900XT at stock 2D does a frame in about 8 minutes. That's 13.3 hours per WU which is 594 PPD. So, not quite close to the two C2Ds running guaranteed 600 pointers.

Now, at the stock 3D settings my card does a frame in almost exactly 6.5 minutes. That's about 10.8 hours per WU which is 731 PPD. Again not quite as high as the C2Ds running the 600 pointers.

Okay, lets move on to overclocked settings... I haven't overclocked my card yet but I have read people have managed 5 minutes per frame. 8.3 hours per WU which gives 950 PPD. Which is about the same as the C2Ds.

So yes, at stock speeds an X1900 or X1950 may not have the same output as an overclocked E6600 with guaranteed WUs, but guaranteed 600 pointers isn't ever going to happen. The GPU clients WUs at the moment are pretty much standard and in the long term will give higher PPD. There is also talk of further increasing the points for GPU WUs, so the output could get significantly higher.

So, if you have an X19xx series card, I suggest you use it. By itself at stock speeds it can come very close to stock speed C2Ds, E6300/6400 maybe even a stock E6600, running both cores on the "best" WUs. Also, not only are you getting roughly a whole C2D in one GPU, you still have a spare C2D core to crunch away. Meaning with just a C2D rig with a X1900 you could be getting in excess of 1000 PPD.

The last few days I have increased production to about 1K PPD - and thats with around 8 clients!

SiriusB

 

[MGP]

hmm, so your figures would imply my GPU is not working to it's potential, and only at 2D speeds. How do you switch it to 3D?

 

[SiriusB]

If you have ATi Tray Tools, you can setup a profile with 3D clock speeds, and choose to run that profile all the time.

Go to the graphics card forum and check out Concorde's sticky - all the info should be there.

I should warn you the card gets a little toasty, more so than when running games!

SiriusB

 

[verbal]

It looks to me then that a generalised answer to my question would be. That a c2d running two instances is comparable to a X1900. It wouldn't suprise me if Stanford had simply engineered the points that way so that neither the people who want to try the beta nor the people who don't lose out points wise.

 

[SiriusB]

I wouldn't say they were comparable because unless you get some great WUs on both cores of the C2D you can expect output to only be around 500-600 PPD.

GPU currently can give you 600PPD at stock 2D speeds. At 3D settings you get over 700 PPD.

In my mind there is no competition between the two.

SiriusB