GPU3 beta available

[jfb9301]

double post.... rum again.....

 

[sirmonkey1985]

anyone elses HFM look like this?

cant even see my PPD.. but i know the TPF is wrong ive sat here with a stop watch and the TPF is around 58 seconds not 1:42..

 

[Mr. Pedantic]

Can you post a log?

 

[sirmonkey1985]

Can you post a log?

# Windows GPU Systray Edition #################################################
###############################################################################

                       Folding@Home Client Version 6.30r2

                          http://folding.stanford.edu

###############################################################################
###############################################################################

Launch directory: C:\Program Files\Folding@home\Folding@home-gpu


[07:09:44] - Ask before connecting: No
[07:09:44] - User name: HoosierDad (Team 33)
[07:09:44] - User ID: 5DA71F095CA616BB
[07:09:44] - Machine ID: 3
[07:09:44] 
[07:09:44] Gpu type=2 species=30.
[07:09:44] Loaded queue successfully.
[07:09:44] Initialization complete
[07:09:44] 
[07:09:44] + Processing work unit
[07:09:44] Core required: FahCore_15.exe
[07:09:44] Core found.
[07:09:44] Working on queue slot 01 [May 25 07:09:44 UTC]
[07:09:44] + Working ...
[07:09:44] 
[07:09:44] *------------------------------*
[07:09:44] Folding@Home GPU Core -- Beta
[07:09:44] Version 2.09 (Thu May 20 11:51:02 PDT 2010)
[07:09:44] 
[07:09:44] Compiler  : Microsoft (R) 32-bit C/C++ Optimizing Compiler Version 14.00.50727.42 for 80x86 
[07:09:44] Build host: amoeba
[07:09:44] Board Type: Nvidia
[07:09:44] Core      : 
[07:09:44] Preparing to commence simulation
[07:09:44] - Looking at optimizations...
[07:09:44] DeleteFrameFiles: successfully deleted file=work/wudata_01.ckp
[07:09:44] - Created dyn
[07:09:44] - Files status OK
[07:09:44] sizeof(CORE_PACKET_HDR) = 512 file=<>
[07:09:45] - Expanded 43748 -> 163067 (decompressed 372.7 percent)
[07:09:45] Called DecompressByteArray: compressed_data_size=43748 data_size=163067, decompressed_data_size=163067 diff=0
[07:09:45] - Digital signature verified
[07:09:45] 
[07:09:45] Project: 10626 (Run 99, Clone 2, Gen 1)
[07:09:45] 
[07:09:45] Assembly optimizations on if available.
[07:09:45] Entering M.D.
[07:09:51] Tpr hash work/wudata_01.tpr:  1182461422 3045222755 3256658692 2151507014 2392407070
[07:09:51] Working on 582 p2750_N68H_AM03
[07:09:51] Client config found, loading data.
[07:09:51] Starting GUI Server
[07:11:34] Completed 1%
[07:13:14] Completed 2%
[07:14:56] Completed 3%
[07:16:38] Completed 4%
[07:18:23] Completed 5%
[07:20:03] Completed 6%
[07:21:44] Completed 7%
[07:23:25] Completed 8%
[07:25:07] Completed 9%
[07:26:49] Completed 10%
[07:28:33] Completed 11%
[07:30:14] Completed 12%
[07:31:55] Completed 13%
[07:33:37] Completed 14%
[07:35:17] Completed 15%
[07:36:58] Completed 16%

thats what i got so far but i want to see my friggin ppd..

 

[Mr. Pedantic]

58 seconds gets you nearly 9100 ppd.
1:42 (which is roughly what the log says) gets you around 5150 ppd.

But that's interesting. Are you sure you timed it right? Because I can't think of a way an automatically generated log could get the times so wrong.

 

[sirmonkey1985]

58 seconds gets you nearly 9100 ppd.
1:42 (which is roughly what the log says) gets you around 5150 ppd.

But that's interesting. Are you sure you timed it right? Because I can't think of a way an automatically generated log could get the times so wrong.

then i have no clue ill blame my system before either of the clients.. but hopefully an update for HFM comes out soon..

 

[Vaulter98c]

Jfb;

Just wondering, I seem to have it in my head you are Army, or someone here is, but for somereason I'm thinking you were (my memory is shit)
Anyways if you are, and you are refering to 6 month field ops, may I suggest getting in touch with your Commo/Supply guys

The VSAT falls under supply, but atleast in my case I keep it under my control (acting BN comm's chief). anyways, if you want access to your rigs in the field, I may suggest going that route... I've not tried it 100% but I can get access to my remote app thru there, and as long as the VSAT isn't on any secure network, you could take that route

I'm not gonna go into anymore detail here than you could google, but try that option, it seems to work for me
Posted via [H] Mobile Device

 

[sirmonkey1985]

ok so far the clients been running for some where around an hour now..

list of important things ive noticed with my system..

1. the client no longer uses an insane amount of video memory.. my current video memory load is at 84mb(54mb when client is off) which is a good thing.. means i know longer have to close the client after pausing it to play my games which means my log files will never be reset every time i have to close the client..
2. it does not put the gpu under full load.. highest ive seen it is 97%..
3. my gpu temps are insanely low.. 55C for standard gpu2 WU's.. 46C for the gpu3 WU.. (ambient temp right now is 11C)
4. the client no longer makes your system lag when running applications that are 3d accelerated or at least doesnt for me.. i can no move dxva movies back and forth between screens when the gpu 2 client would cause the secondary screen to lag like hell..

 

[APOLLO]

list of important things ive noticed..

1. the client no longer uses an insane amount of video memory.. my current video memory load is at 84mb(54mb when client is off)
2. it does not put the gpu under full load.. highest ive seen it is 97%..
3. my gpu temps are insanely low.. 60C for standard gpu2 WU's.. 46C for the gpu3 WU..
4. the client no longer makes your system lag when running applications that are 3d accelerated or at least doesnt for me.. i can no move dxva movies back and forth between screens when the gpu 2 client would cause the secondary screen to lag like hell..

Wow, all I could say is improvement all around. Hope the PPD will be comparable to GPU2 after all the kinks are ironed out.

 

[sirmonkey1985]

Wow, all I could say is improvement all around. Hope the PPD will be comparable to GPU2 after all the kinks are ironed out.

sorry did some editing on my original post to make sure the information was correct..

my guess is the PPD will be low for everyone that doesnt have a GTX 480 card.. but it all comes down to how many shader cores these WU's actually need..

 

[Vaulter98c]

I've seen 99% on my 470, memory used in te high 80's, temps nice for Fermi's
Posted via [H] Mobile Device

 

[Elledan]

I'll keep GPU2 running on my 9800GX2 and GTX275 until there's a good reason to switch

 

[coldstorm]

ok so like another 3 months for ati gpu3 client to release at this rate.

 

[T4rd]

Jfb;

Just wondering, I seem to have it in my head you are Army, or someone here is, but for somereason I'm thinking you were (my memory is shit)
Anyways if you are, and you are refering to 6 month field ops, may I suggest getting in touch with your Commo/Supply guys

The VSAT falls under supply, but atleast in my case I keep it under my control (acting BN comm's chief). anyways, if you want access to your rigs in the field, I may suggest going that route... I've not tried it 100% but I can get access to my remote app thru there, and as long as the VSAT isn't on any secure network, you could take that route

I'm not gonna go into anymore detail here than you could google, but try that option, it seems to work for me
Posted via [H] Mobile Device

I think you're thinking of me, Vault . I'm the 25B SGT that posted in the "show us your pic" (or whatever the title was) thread and we chatted about being fellow 25Bs, heh.

Now that the GPU3 client is out, it seems that the GPU2 client will still be best for my 3x 285s. But now I have a good (for me, at least ) reason for me to sell them toward a 480 or 2, hah. I was hoping this wouldn't happen until the 512 SPU version of the 480 (485 I would guess) came out so it would be better on heat/power usage (assuming it had a die shrink) .

 

[jfb9301]

Jfb;

Just wondering, I seem to have it in my head you are Army, or someone here is, but for somereason I'm thinking you were (my memory is shit)
Anyways if you are, and you are refering to 6 month field ops, may I suggest getting in touch with your Commo/Supply guys

The VSAT falls under supply, but atleast in my case I keep it under my control (acting BN comm's chief). anyways, if you want access to your rigs in the field, I may suggest going that route... I've not tried it 100% but I can get access to my remote app thru there, and as long as the VSAT isn't on any secure network, you could take that route

I'm not gonna go into anymore detail here than you could google, but try that option, it seems to work for me
Posted via [H] Mobile Device

Navy, Submarine......

so transmitting is a b!tch......

 

[DooKey]

Ok, both my 480s are running the new client. My tpf is 36 seconds. Anybody know the ppd for that?

 

[T4rd]

Ok, both my 480s are running the new client. My tpf is 36 seconds. Anybody know the ppd for that?

According to the graph here, that should be about 15k even PPD if you compare it to the guy getting 35 sec tfp getting 15,083 PPD .

Having seen that, I might be selling that 285 OC (along with one or two of my other ones) you're about so send me toward a 480 or two , hah.

 

[The_Heretic]

Navy, Submarine......

so transmitting is a b!tch......

I'm sure the Old Man won't mind you going into radio and using the VLF.

 

[zero2dash]

I installed 197.45's last night on both, no problems (so far). PPD (gpu2) seems the same.

re: ATi v nV f@h
Not sure about GFLOPS, dumb number estimation in my head it looks like f@h favors green because nV uses a larger qty of stream processors with a lower gpu core clock vs. red (ATi) using a smaller qty SPs with a higher gpu core clock. (edit on the old cards at least) Folding craves/needs/wants more SPs than more mhz in the gpu area (as far as I can tell). Again though just my dumb math, just wanted to throw my $0.02 cents in the hat on that subject.

Let me rephrase/elaborate: (sorry )
nV - lower gpu core clock, higher # of shaders, higher shader clock
ATi - higher gpu core clock, lower # of shaders, lower shader clock
f@h eats the shaders, depends on # of shaders and shader clock. ATi cards have less of both that's why they fold like $#!t.

 

[Agent_N]

I installed 197.45's last night on both, no problems (so far). PPD (gpu2) seems the same.

re: ATi v nV f@h
Not sure about GFLOPS, dumb number estimation in my head it looks like f@h favors green because nV uses a larger qty of lower clocked stream processors vs. red (ATi) using a smaller qty of higher clocked SP's. Folding craves/needs/wants more SPs than more mhz in the gpu area (as far as I can tell). Again though just my dumb math, just wanted to throw my $0.02 cents in the hat on that subject.

Um, I thought AMD had higher stream processor numbers and clocks 5870 has 1600 SP's @ 850Mhz versus GTX480 with 480 stream processors @ 700Mhz. I saw a review of the 5970 and it is supposedly capable of almost 4.5 teraFLOPS single precision and 980+ gigaFLOPS double precision.

Shader clock makes a huge difference outside of streaming processor quantity.

 

[Parja]

Um, I thought AMD had higher stream processor numbers and clocks 5870 has 1600 SP's @ 850Mhz versus GTX480 with 480 stream processors @ 700Mhz. I saw a review of the 5970 and it is supposedly capable of almost 4.5 teraFLOPS single precision and 980+ gigaFLOPS double precision.

Shader clock makes a huge difference outside of streaming processor quantity.

Shaders run faster than the core on NVIDIA cards. Stock 480 shaders run at 1401MHz.

 

[zero2dash]

Sorry, I revised my post after seeing yours...(Agent_N)
I haven't kept up with the newer cards ('cause I'm not a shopper)...the old cards though it seemed that way (nV = more SPs with lower gpu core clock vs ATi = less SPs with a higher gpu core clock). At least I thought that was the deal when I looked at it while deciding between the 260s and a 48xx setup.

 

[Mr. Pedantic]

The difference in numbers of shaders and clocks is only partly the difference; theoretically, AMD cards vastly outperform NVidia cards in double precision floating point. I think it's mostly a coding thing, more than anything else.

 

[zero2dash]

I really wish ATi or Stanford (or both) would get their butts in gear and put up some respectable PPD; I'm itching for a new card or two and I'm also wanting something with less heat....but since my systems spend any idle time f@h and my game time is low these days, I'm stuck with nV. (not that that is necessarily a bad thing; I just want ATi to amp up the competition in another area [f@h])

GT4xx are out of the question...I won't deal with the heat. 200s already had enough heat and I'm already kinda sick of that heat. No way I'm buying a louder, hotter, more power sucking card from the green side.

 

[sirmonkey1985]

I installed 197.45's last night on both, no problems (so far). PPD (gpu2) seems the same.

re: ATi v nV f@h
Not sure about GFLOPS, dumb number estimation in my head it looks like f@h favors green because nV uses a larger qty of stream processors with a lower gpu core clock vs. red (ATi) using a smaller qty SPs with a higher gpu core clock. (edit on the old cards at least) Folding craves/needs/wants more SPs than more mhz in the gpu area (as far as I can tell). Again though just my dumb math, just wanted to throw my $0.02 cents in the hat on that subject.

Let me rephrase/elaborate: (sorry )
nV - lower gpu core clock, higher # of shaders, higher shader clock
ATi - higher gpu core clock, lower # of shaders, lower shader clock
f@h eats the shaders, depends on # of shaders and shader clock. ATi cards have less of both that's why they fold like $#!t.

completely different shader tech too.. Nvidia uses 2 way shaders at a much higher clock speed then ATI's shaders where ATI uses 5 way shaders at a much lower clock speed then Nvidia's shaders but can do 5 times the amount of work at the same exact time which is why the ATI cards do so badly in F@H since the WU's are not taking advantage of the 1600 shader's with the HD5870 for example.. so say the WU uses 200 shaders, a GTX 260 can easily handle 200 shaders at twice the speed where as the HD5870's shaders are basically sitting idle.. if anything the ATI card should be able to do SMP WU's due to the shader count but i doubt that'll ever happen..

 

[Elledan]

The difference in numbers of shaders and clocks is only partly the difference; theoretically, AMD cards vastly outperform NVidia cards in double precision floating point. I think it's mostly a coding thing, more than anything else.

From what I understand the ATI GPU architecture at the moment isn't aimed at GPGPU at all, only graphics processing, which thus introduces many bottlenecks if one wants to run more generic code on it than a graphics shader.

 

[sirmonkey1985]

From what I understand the ATI GPU architecture at the moment isn't aimed at GPGPU at all, only graphics processing, which thus introduces many bottlenecks if one wants to run more generic code on it than a graphics shader.

its more geared toward openGL and direct compute when it comes to gpgpu which no one really uses right now.. were still stuck with F@H using old tech coding and i dont see them moving to direct compute or openGL any time soon.. those 2 areas are where ATI kills Nvidia..

 

[Nessus85]

its more geared toward openGL and direct compute when it comes to gpgpu which no one really uses right now.. were still stuck with F@H using old tech coding and i dont see them moving to direct compute or openGL any time soon.. those 2 areas are where ATI kills Nvidia..

Too bad this is equivalent to me saying you may be better at English, but my Latin blows yours away!

 

[coldstorm]

well with opencl both card as suppose to be equally at do gpgpu. the 480 just has less Gflops in both single and doubler floating when compared to 5870. now if gpu3 would take advantage of those extra gflop then maybe we would see ati performance in F@H coming out on top until then nvidia is king of the hill

 

[tjmagneto]

HFM.NET finally was able to update the projects and the numbers I got from the benchmarks for project 10627 was Avg. Time / Frame : 00:01:21 - 6,517.33 PPD.

This is for a single GTX 275 with no other folding apps running. Clocked at 648/1502/1152.

 

[Zero82z]

HFM.NET finally was able to update the projects and the numbers I got from the benchmarks for project 10627 was Avg. Time / Frame : 00:01:21 - 6,517.33 PPD.

This is for a single GTX 275 with no other folding apps running. Clocked at 648/1502/1152.

That's in line with what I was getting from my GTX260. Hence why I switched back to GPU2.

 

[ppp353]

I get nothing but "shut down unstable machine" with my 8800GTS. Never had any trouble with GPU2 client. I'm running the 197.45's.

Maybe I'll just wait till the next release.

 

[sirmonkey1985]

HFM.NET finally was able to update the projects and the numbers I got from the benchmarks for project 10627 was Avg. Time / Frame : 00:01:21 - 6,517.33 PPD.

This is for a single GTX 275 with no other folding apps running. Clocked at 648/1502/1152.

not as bad as i thought.. ive been stuck doing the slow ass WU's on the gpu2 client a lot so im only losing about 200-300 PPD.. was doing 6800ish PPD using the gpu2 client on my gtx 260.. but i dont feel like switching back to the gpu2 client so ill just let it keep running..

my TPF is 1:27 with my GTX 260 @ 600/1512/1161..

I get nothing but "shut down unstable machine" with my 8800GTS. Never had any trouble with GPU2 client. I'm running the 197.45's.

Maybe I'll just wait till the next release.

probably not worth it for you to run the gpu3 client anyways since the WU's are much larger and i think your PPD would suffer a lot more then what the GTX 200 series are suffering..

 

[sirmonkey1985]

lol i was looking through my benchmark thing on HFM and i saw this.. just had to post it..

Project ID: 10626
 Core: OPENMMGPU
 Credit: 610
 Frames: 100


 Name: GTX260
 Path: C:\Program Files\Folding@home\Folding@home-gpu\
 Number of Frames Observed: 300

 Min. Time / Frame : 00:00:23 - 22,914.8 PPD
 Avg. Time / Frame : 00:01:25 - 6,200.5 PPD

look at the PPD on the minimum time / frame lol

 

[Agent_N]

My PPD went from 5,000PPD - 6,000PPD depending on WU to 4888PPD or less with GPU 3. I'm going to use GPU2 until it expires and I have to switch. seeing 50-52 sec TPF to 109+ sec TPF is sad! It looks like Fermi is the architectrue to have for GPU3, at least until Stanford and AMD get things worked out for OpenCL...

 

[Kendrak]

It looks like Fermi is the architectrue to have for GPU3, at least until Stanford and AMD get things worked out for OpenCL...

When genetically engineered pigs fly.

 

[Agent_N]

In terms of PPD, they seem to do better since we can't compare them on GPU2 client. I get less PPD with GPU3 with G92 hardware, thus, Fermi is better for GPU3 for PPD if one wished to use GPU's for folding, however, nothing can beat SMP!

 

[sirmonkey1985]

In terms of PPD, they seem to do better since we can't compare them on GPU2 client. I get less PPD with GPU3 with G92 hardware, thus, Fermi is better for GPU3 for PPD if one wished to use GPU's for folding, however, nothing can beat SMP!

um yeah the gpu 2 with any GTX 200 series card kills the smp client when you only have a quad core lol

but yeah for me the benefits of the client out weigh the loss in PPD so ill continue to run it..

 

[Linden]

I like the 9400 ppd I'm getting from GPU2 (project 6606 for example) versus the 6500 ppd from GPU3.

Exactly. I tested GPU3 last overnight on a 4XGPU machine and your is similar to my experience. The machine I tested with is configured with a Q6600, GTX295, and 9800GX2. OS is Win7/64 and Nvidia drivers 197.15. All of the GPU cores' production were down by 30-40% as compared to the same configuration with GPU2 clients running.

I think I'll wait until GPU2 is obsolete or until big improvements in the GPU3 client are made before I permanently switch. For now, it's back to GPU2.

 

[sirmonkey1985]

and on that note the gpu3 server is now out of work.. arg!!!

edit looks like theres more WU's again..