Gpu2 Client Is Here!!!! This Is Not A Joke!!

[Mechromancer]

http://folding.typepad.com/

Take a look for your own damn selves!!!!! Finally!!!!

I'm too shocked to type intelligent comments right now so heres all I have: F'ing YAY!

[Mechromancer]

Please fold with your Crossfire and 3870X2 setups. I'm very interested to know if they work right out of the box with this new client. Unfortunately there is no fancy GUI with this one but FahMon is projecting 1330PPD which ain't bad for a GPU. Thats at 3D speeds by the way. You will most likely have to force 3D clock speeds with ATITool 0.27 Beta 4. ENJOY!

[sc00p]

Good one Mechromancer, I just looked couple of hours ago and there wasn't anything.

Thanx for the news... now fold on those who've ATIs

Would be interesting for those considering folding with gpus, that how much cpu cycles it's going to bind with gpu-client.

[Mechromancer]

Good one Mechromancer, I just looked couple of hours ago and there wasn't anything.

Thanx for the news... now fold on those who've ATIs

Would be interesting for those considering folding with gpus, that how much cpu cycles it's going to bind with gpu-client.

Actually folding with the SMP client and GPU client on a dual core IS NOT A GOOD IDEA! My PPD on both dropped dramatically. The GPU client needs an entire core to feed it information. You quad core owners shouldn't have any problems though. Luckily for me, the GPU clients puts out more PPD than my Opteron 185 with the SMP client . I'm now a full time GPU folder.

[[XC] riptide]

YAHOOOOOOOO! I've been waiitng for this for a whilw. Time to unlease hell!

Hope the PCIEX X1900 256mem pausing issue is fixed.

[Mechromancer]

Heres the GPU2 FAQ: http://folding.stanford.edu/English/FAQ-ATI2

Unfortunately you can only run 1 GPU2 client on the 3870X2 at the moment. Crossfire has to be DISABLED to run two clients on a multi-GPU system also. At the moment, two HD3870's is looking like an economical option. Until the 3870X2 is fully supported in a future release, having 2+ separate cards is the way to go.

[Mechromancer]

I think I'll overclock my PCI-Express bus to 125Mhz to see if I get more PPD due to more bandwidth between the CPU and GPU. I already have an overclocked Diamond HD 2900XT@800/900 from the factory. The onboard memory bandwidth should be more than enough so I'll just bump the core up to 850 or so later on today just to see what happens.

Please let us know how your HD 3870's perform as well. Also, faster CPUs should yield more GPU PPD. If you have an E8500 overclocked to hell then PLEASE let us know your PPD!

[Stijndp]

Heres the GPU2 FAQ: http://folding.stanford.edu/English/FAQ-ATI2

Unfortunately you can only run 1 GPU2 client on the 3870X2 at the moment. Crossfire has to be DISABLED to run two clients on a multi-GPU system also. At the moment, two HD3870's is looking like an economical option. Until the 3870X2 is fully supported in a future release, having 2+ separate cards is the way to go.

What about multi-gpu support and the -gpu flag?

Multi-gpu is supported through the -gpu flag provided on the command line. Similar to running multiple SysTray CPU clients, copy your \Application Data/Folding@home-gpu to Folding@home-gpu2/, and copy the shortcut and edit it to point to that new working directory. If you run the client from the command line, the current directory will be used for data and thus the .dll files must be there as well (see DLL issues below). The display must be active on the board you plan to use, and –gpu 0 will select the first board, –gpu 1 will select the second board, -gpu 2 the third board, and so-on. You will need to disable crossfire for multiple boards to be detected. Currently, only one client is supported on a 3870X2.

Now they only need to use 55% of the market share instead of the miserable 20%.
I see a quadcore running SMP + doule GPU2. SMP on idle and gpu's on low ^^

[[XC] gomeler]

Woah! Now that's what I'm talking about. In ~10 days I'll be finished benching my HD3850's and I can then turn them over to science as they ease into retirement. While I own them I'll get them running F@H, it'll be 2 weeks or so till I can do it though.

[Bobsama]

Woah! Now that's what I'm talking about. In ~10 days I'll be finished benching my HD3850's and I can then turn them over to science as they ease into retirement. While I own them I'll get them running F@H, it'll be 2 weeks or so till I can do it though.

That's quite the retirement... letting them slowly wilt and die under the constant pressure! As for my HD3850--I'll probably use this client.

[Mechromancer]

Apparently the GPU client can work on ANY R6xx GPU. If an HD 2400 series card can do it then a 780G integrated GPU should be able to add a few hundred extra PPD. I wish a mobo manufacturer would come out with a decent 780G motherboard that doesn't blow MOSFETS with 125watt TDP CPUs. If somebody does soon, I'll build a 9850BE+780G+Crossfire 3870 system for maximum folding capability.

BTW, I think the GPU2 Client allows for more PPD than the PS3.

[Marvin_The_Martian]

Thanks for the heads up!

[Sparky]

Even though I still have an X1950XT

[Marvin_The_Martian]

Lol yeah I dl and treid to run it on my ( not XS worthy ) 1600pro :P

[Mechromancer]

toTOW on the Folding@Home Forums posted some VERY important information. The GPU client is heavily dependent on CPU speed. PERFECT FOR OVERCLOCKERS! Unfortunately for me, my weak Opteron 185 is once again holding me back from making high PPD .

To answer many of the potential questions, I will give you the result of benchmarking I did ...

toTOW wrote:Number from HD3870.

Q6600 @ 2.4 GHz (266 * 9), GPU @ 780 MHz : 1310 PPD
Q6600 @ 3.4 GHz (485 * 7), GPU @ 780 MHz : 1860 PPD
Q6600 @ 3.4 GHz (485 * 7), GPU @ 850 MHz : 1860 PPD

More numbers with the 2900XT ...

Dual Opteron 2212 @ 2.7 GHz : 1350 PPD (with GPU @ 739 and 847 MHz)
Q6600 @ 3.4 GHz (485 * 7), GPU @ 739 and 847 MHz : 1860 PPD
Q6600 @ 3.8 GHz (475 * 8), GPU @ 739 MHz : 1860 PPD
Q6600 @ 3.8 GHz (475 * 8), GPU @ 847 MHz : 2100 PPD
Q6600 @ 3.8 GHz (475 * 8), GPU @ 860 MHz : 2100 PPD



Yes, most people are going to be CPU limited with the GPU2 client ...

toTow we really appreciate the info. Heres the link to the original post: http://foldingforum.org/viewtopic.ph...t=2020&start=0

[Sparky]

At any rate I'm going to be folding on my X1950XT again probably since I have to sell my phenom. I wonder if this version works any better than the previous one did with these cards? I'm guessing not...

[[XC] riptide]

I'm downloading now. I'll be firing this onto teh X2900XT. Now I could play games at 900/900. Will be interesting to see what clocks I can sustain this at.! I'll be back with results. WooHoo!

[[XC] DragonOrta]

I'll give it a try on my 3850 later tonight. Time to start OCing the card to see when core clocks stop helping with a 4ghz q6600.

[Marvin_The_Martian]

At any rate I'm going to be folding on my X1950XT again probably since I have to sell my phenom. I wonder if this version works any better than the previous one did with these cards? I'm guessing not...

This version only works with 6xx gpu's ( 2900 and up ). Uses a diffrent core so it makes sence, also no longer relying on DX but on Ati's CAL.

[[XC] riptide]

First Screenshot in....

I this is X2900XT at stock 3D clocks/voltages... or close.

It isn't heating much at all!!!

BTW. Thats also a E6850 @ 3.6Ghz.

Latest log file...

[18:24:33] Project: 2799 (Run 1, Clone 65, Gen 0)
[18:24:33]
[18:24:33] Assembly optimizations on if available.
[18:24:33] Entering M.D.
[18:24:47] Working on 582 p2799_N68H_AM03
[18:24:48] Starting GUI Server
[18:26:25] Completed 1%
[18:27:58] Completed 2%
[18:29:27] Completed 3%
[18:30:58] Completed 4%

[Marvin_The_Martian]

Project is wurth 97points

[18:26:25] Completed 1%
[18:27:58] Completed 2% 1m33s
[18:29:27] Completed 3% 1m29s
[18:30:58] Completed 4% 1m31s

Avg frame time 1m31s = 91s ( x 100 frames = 9100s )
86400 / 9100 = 9.4945 wu/day = 97*9.945 = 920.97PPD

I guess?

Edit: need more wu's though this one might stink.

[[XC] riptide]

Marvin!!!! The thing that worries me is ... I then set the card at 900/900. Same time needed roughly... 91s?

[[XC] DragonOrta]

3850 @ 700/855
Q6600 @ 4ghz

Project 2799
[18:44:08] Starting GUI Server
[18:44:55] Completed 1%
[18:45:41] Completed 2%
[18:46:26] Completed 3%
[18:47:12] Completed 4%
[18:47:57] Completed 5%
[18:48:42] Completed 6%
[18:49:27] Completed 7%
[18:50:13] Completed 8%
[18:50:58] Completed 9%
[18:51:43] Completed 10%
[18:52:29] Completed 11%
[18:53:14] Completed 12%
[18:54:01] Completed 13%

Little over 45 sec per frame.

[Sparky]

Remember it just got released. Sometimes they put out really tiny units to begin with to be sure things are going smoothly before they put out the real deal.

[[XC] DragonOrta]

Marvin!!!! The thing that worries me is ... I then set the card at 900/900. Same time needed roughly... 91s?

Looks like you need to stop :banana::banana::banana::banana::banana: footing around with that CPU overclock.