2nd Generation GPU client getting closer

7im

Limp Gawd
Joined
Aug 23, 2006
Messages
169
F@h News: GPU2 update

We've internally dubbed the new GPU client/core "GPU2". Here's an update. The GPU2 closed beta testing has been moving along well. There are a few bugs which have shown up, but nothing we'd call "showstoppers" just yet. The beta test is outside of Stanford and ATI now. If all looks good, we plan to next have a completely open beta.
 
Thanks for the heads up 7im. Good news and personally can't wait. Is this client going to support newer architectures, and if so which ones specifically? TIA.

 
Might give me a good reason to get an ATI card for a (currently) non existent HTPC ;)

 
I'm going to keep my eye on this. If all goes well, I'm going to pop a used ATI card into my single-core, P4 CD duplicator rig at work. I never cared for the existing client.

 
Depending on the machine, can you run Dual SMP clients AND a GPU Client at the same time?

 
Depending on the machine, can you run Dual SMP clients AND a GPU Client at the same time?
I assume that is possible, but I wouldn't try it on anything other than a dual quad-core Xeon server with 8 cores. Even so, with the current GPU client, you are likely to see a drop in PPD because the client needs the full processing power of a single core.
 
Any word on nVidia and folding? I like their stuff. I have not had an ATI since the 9800Pro.
 
No official confirmation but I can say it would be there.

 
Yeah I get to put in my 3850 AGP and fold on it... when the time comes. Supposed to be nearly double the speed of the 1950... so /cross fingers!

 
From reading the project News posts, I can tell you the new GPU2 client is only ATI 2xxx and 3xxx at the current time. A 3870X2 is a 3xxx model. ;) However, nothing has been said if it runs only one client or two clients.

There was also mention of CAL in the News a few posts back, so the new client is different from the 1st Gen DX9 based GPU client. No specifics on what that means for performance and stability, but I take it to indicate an improvement, otherwise they wouldn't have changed it, IMO.

And an NV GPU client IS still in the works.
 
Yeah I get to put in my 3850 AGP and fold on it... when the time comes. Supposed to be nearly double the speed of the 1950... so /cross fingers!


Could you tell us how loud that thing is when folding once you get it running? That's the card I'd have to install on the duplicator computer since it only has an AGP slot.


 
Sure can as soon as they release the client I'll upgrade that machine and report the difference in Watts as well :) and be happy to provide PPD. Right now with the CPU client I get 582PPD from the GPU and about 134PPD from the CPU just over 700 PPD.

So I'll definitely chime in on PPD changes as well as wattage performance!

 
If what you said about performance is as true with folding as we know with gaming performance, I would be incredibly pleased with 1200 PPD out of a P4 2.4 rig at work. Hopefully I can find a used AGP 3850.

 
I assume that is possible, but I wouldn't try it on anything other than a dual quad-core Xeon server with 8 cores. Even so, with the current GPU client, you are likely to see a drop in PPD because the client needs the full processing power of a single core.

This is what I planned on doing, but the darn box gets way too hot with 2 2900XTs (even at idle) and 2 quads folding. Maybe with the new 45 nm quads and the 3800 series it'll stay decently cool on air.
 
This is what I planned on doing, but the darn box gets way too hot with 2 2900XTs (even at idle) and 2 quads folding. Maybe with the new 45 nm quads and the 3800 series it'll stay decently cool on air.
I haven't tried the 45nm quads yet, but might be receiving two ES Harpertowns sometime this month and will find out about temps as well as the performance. Honestly, unless production goes way up with the new card support, I would not install a GPU client on a Xeon machine. Anything else yes.

As far as the GPUs are concerned, I wouldn't bother with the 2900s in a dual box, especially with FB-DIMMs. I think you're asking for trouble unless you put everything on WC, which would cost megabucks. Your best bet is to do what I and others do, leave your box permanently open. You can have decent OCs that way for CPUs+GPUs and need not worry about heat buildup in the case.

I was thinking of getting a HD3870 X2 for my dual Opteron box. Two out of the 4 cores can be used for two GPU clients and the other two cores can stay with the SMP client. Since I need to upgrade my video card and can't use up too many expansion slots, this option seems good. However, I might just wait for a NVidia client. We'll see when the time comes and whether or not Stanford has something out by then.
 
I haven't tried the 45nm quads yet, but might be receiving two ES Harpertowns sometime this month and will find out about temps as well as the performance. Honestly, unless production goes way up with the new card support, I would not install a GPU client on a Xeon machine. Anything else yes.

As far as the GPUs are concerned, I wouldn't bother with the 2900s in a dual box, especially with FB-DIMMs. I think you're asking for trouble unless you put everything on WC, which would cost megabucks. Your best bet is to do what I and others do, leave your box permanently open. You can have decent OCs that way for CPUs+GPUs and need not worry about heat buildup in the case.

I was thinking of getting a HD3870 X2 for my dual Opteron box. Two out of the 4 cores can be used for two GPU clients and the other two cores can stay with the SMP client. Since I need to upgrade my video card and can't use up too many expansion slots, this option seems good. However, I might just wait for a NVidia client. We'll see when the time comes and whether or not Stanford has something out by then.

Thing is, I use it as my main workstation at home and it does play Crysis halfway decently with crossfire enabled. But, my office gets way to hot and my wife complains. So I actually just got an 8800GTS 512 in and will be using it for the time being instead of the 2900s. They will replace my 1950s in my Q6600 box and my 1950s will go into a PD 975x box. If the 2900 with the 512 ring bus turns out to be a monster folder, I'll stick them in the PD box and just have GPU folding on it.
 
Thing is, I use it as my main workstation at home and it does play Crysis halfway decently with crossfire enabled. But, my office gets way to hot and my wife complains. So I actually just got an 8800GTS 512 in and will be using it for the time being instead of the 2900s. They will replace my 1950s in my Q6600 box and my 1950s will go into a PD 975x box. If the 2900 with the 512 ring bus turns out to be a monster folder, I'll stick them in the PD box and just have GPU folding on it.
That sounds like a very good idea. I wish I had either a 2000 or 3000-series card to try it out. I expect a 50% increase in production at bare minimum compared to the 1900s, possibly much more. It might make a lot of people switch back.
 
Back
Top