GPU2 client for nVidia cards

[PorPorMe]

Originally Posted by gedon

1. the 8800GT has 112 shaders (streamprocessors) the 9600GT only 64
2. I have a factory overclocked card (700/1700/1000 - core/shader/mem)
3. both of us overclocked the shaders, I reached the maximum at 1850MHz, I think ThunderRds 9600GT has only a few MHz more - 50 or 100MHz

OK, lets see if I've got this right. [If I do, It's just as confusing] When you talk about shaders, you mean steamprocessors. Your card is OC'd core-700, shader-1700, mem-1000. but then the 8800gt has 112 shaders and the 9600gt has 64 shaders. Then to top it off, there is no steamprocessors catagory in the spec sheet and the only catagory that coralates to the numbers 112-8800gt and 64-9600gt is config clock-and it's not exact.
I feel like putting my thumb in my mouth and walking in circles.

[gedon]

stream processors = unified shaders

[gedon]

allright, I have to correct me. The 177.89 is as fast as the .35
but I clocked the shaders down from 1849MHZ to 1836MHz to reduce the noise coming from the graphiccard or psu after I installed the .89
I should write down things like that
Today I increased gpu-memclock from 1066MHz to 1095MHz. It compensates the lower shaderclocks.

[dabaerman]

8800GT's are becoming more and more affordable. the real question is, if
using two cards, single or dual slot cooling?
I can guess that getting the heat out of the box is most important. that
makes the dual slot card choice #1. it also
means that more air to the cards is necessary.
I use the heat to keep my room warm in winter. gets too toasty in
Summer. working on a better plan for the room.

Ron

[ThunderRd]

Originally Posted by PorPorMe

OK, lets see if I've got this right. [If I do, It's just as confusing] When you talk about shaders, you mean steamprocessors. Your card is OC'd core-700, shader-1700, mem-1000. but then the 8800gt has 112 shaders and the 9600gt has 64 shaders. Then to top it off, there is no steamprocessors catagory in the spec sheet and the only catagory that coralates to the numbers 112-8800gt and 64-9600gt is config clock-and it's not exact.
I feel like putting my thumb in my mouth and walking in circles.

The 9600GT has fewer stream processors but I have mine running at a much higher speed than stock.

Because the number of stream processors is lower but the speed is higher, my card is more efficient [read: (much)cooler] and can still be competitive graphics-wise (and folding-wise) with many of the 8800s out there. A pair of them in SLI could [I'm fairly sure] easily outperform an 8800GTX-and they are only $160 or so. I strongly recommend this card for the under-$200 market. It's proven to be a most worthy card IMO.

[ThunderRd]

Stanford has finally addressed a long standing need for some properly written setup guides for the different F@H clients. They have created a guides page and the URL is here:

Folding@home - Guide

I've read through them; there isn't any new information there but all of the various pitfalls of installation have now been collected into one place, instead of having to search through tons of posts on the community forum. I would say that they are directed at the new donor and are quite straightforward. I am familiar with the writer, he's a regular contributor at the forum and a generally helpful guy.

Everything is there, from the release clients to the Linux client to the GPU and all of the Windows betas in all of their flavors. Check it out.

[cross-posted in the SMP thread]

[PorPorMe]

Originally Posted by ThunderRd

The 9600GT has fewer stream processors but I have mine running at a much higher speed than stock.

Because the number of stream processors is lower but the speed is higher, my card is more efficient [read: (much)cooler] and can still be competitive graphics-wise (and folding-wise) with many of the 8800s out there. A pair of them in SLI could [I'm fairly sure] easily outperform an 8800GTX-and they are only $160 or so. I strongly recommend this card for the under-$200 market. It's proven to be a most worthy card IMO.

I don't understand most of that and will go with what I do know. I will probly do some basic OC'ing on whatever card I get. I don't need all the info you know about it. I've decided on either the 8800gt or the 9600gt.
Both cards are beyond my video needs but I figure they are worth the xtra money for the increase in folding speed. I figure the GTX would be a little over kill. Both in $ and in video needs, and saveing a hundred is easier than 160! I'm into imediate gratification.

[dabaerman]

Originally Posted by PorPorMe

I've decided on either the 8800gt or the 9600gt.
Both cards are beyond my video needs but I figure they are worth the xtra money for the increase in folding speed. I'm into immediate gratification.

I know this is about nVidia cards and about time as well. however, there are the competitions cards.
I have 2 AMD/ATI mainboards that may or may not like nVidia cards. they both have ATI cards (X300) and these aren't able to fold. without looking for stats to compare both brands, I'd say that price will determine where I go for GPU Folding. it's the part about immediate gratification.

Ron

[PorPorMe]

We now have 100% success!
the 9600GT with 177.89 drivers @ E4300 with the GPU client 3427 PPD! Awesome! That comes with a surprise to. It frees up one CPU as it's running of the GPU. Who woulda thunk!
The other CPU is running the Fah6 service. Great stuff! I love it when a plan comes together!

[Samuknow]

Got 2 GPU cores running. Dang it is screaming.

I think I will throw some water blocks on these things and see what they can do. Running around 60C while folding.

[PorPorMe]

Originally Posted by Samuknow

Got 2 GPU cores running. Dang it is screaming.

I think I will throw some water blocks on these things and see what they can do. Running around 60C while folding.

My CPU temps dropped to 60c and the Gpu is at 52c. I've read, i think it was you and Thunder that you can increase shaders w/o increasing temps. I'll read that over mañana.
I'm going to crank out some WU's! So now that the hardware and software are done, [not completely, I have to set up .bat files to shut down and start the clients when I want but that's just mechanics now that I've done it once.
This is what i set out to do--Damn, I like it! Next week, I won't even be thinking about it.
Then I'll start working on getting a couple of folding rigs that aren't Dogs.

The problem is i think that during my manic phase I fried 3 CPU's. I haven't wanted to admit it but it's the only answer that fits. I have 4 socket A boards and 3 AMD CPU's and the CPU's won't get a bios beep on any of the boards. I must have cooked them some where along the line.
One of them was my Xp 2400+ so it's got to be that I fried them. So I'm running 750's. the shame of it all.

[Samuknow]

I think I have a couple of cpus left around here.

[ThunderRd]

@Sam: What kind of production are you getting? Just curious.

@PPM: Pushing the shaders will raise the temps some, but not nearly as much as pushing the core. I run the shaders 22% over and the core only 10%, and the machine has been stable so far. The shaders have the most pronounced effect on the speed. I was not able to do anything with the memory clock. On my machine it crashed without fail if I went even 5% over on the memory. I have left it alone as a result. I am doing 1% in 1:25-1:35 at these settings, about 4100-4500 ppd from each card. Since I have a quad I also have SMP running on 3 cores, with the GPU affinity of both cards assigned to the 4th core.

[ThunderRd]

Wow. Big news for GPU folders here regarding new projects now in the wild:

http://foldingforum.org/viewtopic.ph...=t&sd=a#p54321

These projects will radically affect your ppd, so your machine isn't malfunctioning if you get them. I got my first yesterday and the client went from 4200ppd to 2700. I'm happy to do the science so I don't care, but there will be complaints for sure...

Long thread but a good read.

[Samuknow]

Good thing I have two clients running then...LOL

Need to find a cheap 8600 card for my wife's machine....

[ThunderRd]

Don't laugh... ') There's a guy in the forum that says he will lose 25k ppd because of this :0

All in all it's OK. The NVidia cards were getting a bonus for a while on the smaller WUs, because they ran them faster than ATI cards could. Now, the WUs get bigger, and the ppd decreases. ATI cards will be able to get more of an even playing field now, so it's good for them. The ATI guys are still trying to get some weird technicality worked out in the next version of the core and at that point will fly. The ATI architecture has significantly more potential on the bigger WUs than NVidia.

It's all good. Fold whatever they give you-that's what it's all about.

[PorPorMe]

I just took a look and my PPD has dropped 1k. I don't know if it's related or not, I've just started to trouble shoot, but my rig is dogged down. both cores are at 100% and there is a new one in process, FahCore_aO.exe. Nothing in taskmaneger indicates that kind of work being put on the CPU, but something is going on.

Edit: I shut down both clients and my rig is running smooth agian. Any ideas? Maybe my OC without Physx?
2nd Edit: Went to the forum to see if i might find some clue, didn't find anything. Like most, the numbers I can deal without any problems-the folding is what's important-but I'm going to have serious problems with my rig running like it's a 650Mhz. And it is the GPU client that is doing it. I have to figure something out.

[ThunderRd]

Originally Posted by PorPorMe

I just took a look and my PPD has dropped 1k. I don't know if it's related or not, I've just started to trouble shoot, but my rig is dogged down. both cores are at 100% and there is a new one in process, FahCore_aO.exe. Nothing in taskmaneger indicates that kind of work being put on the CPU, but something is going on.

Edit: I shut down both clients and my rig is running smooth agian. Any ideas? Maybe my OC without Physx?
2nd Edit: Went to the forum to see if i might find some clue, didn't find anything. Like most, the numbers I can deal without any problems-the folding is what's important-but I'm going to have serious problems with my rig running like it's a 650Mhz. And it is the GPU client that is doing it. I have to figure something out.

@ PPM: fahcore_a0.exe is not a GPU core. Are you running another client? What WU are you running on all of your GPU and/or CPU clients? Is it possible to see your log[s]? Have you read my previous post about the new WUs now released?

PhysX will have no impact on f@h, so don't fret about that. Why do you think that it's the GPU client at fault here? [I'm not being cheeky-I'm not as familiar with your rig as you are, and you must have some reason to suspect this that you haven't said].

@Sam: It might be a good idea to wait and see if an 8600 can cut these WUs in time-the deadlines are going to get tighter and if we struggle on 88-9600s, well...

@ Everyone: I am finding these new WUs very tough to run[I'm not alone; read the thread I posted earlier]. I was hit by a half-dozen EUEs this morning in the logs; the WUs had been downloaded during the night. I have played with the OC settings-I had to back off from 22% to about 12% to get units to complete; the production is nearly halved from the past WUs, partly because of the reduced overclock, but mainly because they are larger proteins with many more calculations to perform. There are also a lot of reports from users that they are running much hotter as well, so much so that some have been afraid to continue 24/7. [My temps are up a few degrees but not to the level some are reporting]. They use a boatload more resources, too. [About 25% more memory usage in my case]. Some are reporting sluggish response, laggy browsers, etc etc. So be prepared for a bumpy ride for a while as things smooth out.

That being said, the Nvidia developers are actively watching the feedback, and have already responded that they are currently looking at the heat problems as well as trying to tune the optimizations in the code to get back some of the lost production. Remember that these are still new WUs; they normally do get better and more stable as time goes by.

EDIT: Looks like the settings I've backed off have successfully completed 2 WUs on each card, so they look OK for now. Down to 2:40 from 1:25.

[PorPorMe]

My apologies. It was me jumping to a conclusion i shouldn't have. The graphics had slowed to the point that it was worse than onboard video and when I shut down the GPU client all was well, but I jumped to a conclusion.
I haven't made any changes, I've just been hunting for a phantom program as both cores were running at 100%.
I haven't found it but I rebooted twice and things are better. I think I'll go on the attack and do a full bug sweep.
This kind of acts like one.

[ThunderRd]

Logs are normal except for this:
You finished a couple of 5506 WUs[the older,smaller ones] then you got one of the new, slower monsters-5019, and this happened:

[06:29:04] Completed 98%

[06:31:49] Completed 99%

[06:34:34] Completed 100%

[06:35:34] 

[06:35:34] Finished Work Unit:

[06:35:34] - Reading up to 30216 from "work/wudata_06.trr": Read 30216

[06:35:34] trr file hash check passed.

[06:35:34] - Reading up to 199240 from "work/wudata_06.xtc": Read 199240

[06:35:34] xtc file hash check passed.

[06:35:34] edr file hash check passed.

[06:35:34] logfile size: 35998

[06:35:34] Gromacs cannot continue further.

[06:35:34] Going to send back what have done.

[06:35:35] logfile size: 35998 info=35998 bed=0 hdr=23

[06:35:35] - Writing 36534 bytes of core data to disk...

[06:35:35] Done: 36022 -> 7214 (compressed to 20.0 percent)

[06:35:35]   ... Done.

[06:35:35] 

[06:35:35] Folding@home Core Shutdown: UNSTABLE_MACHINE

[06:35:38] CoreStatus = 7A (122)

[06:35:38] Sending work to server

[06:35:38] Project: 5019 (Run 5, Clone 15, Gen 3)





[06:35:38] + Attempting to send results [September 10 06:35:38 UTC]

[06:35:38] - Reading file work/wuresults_06.dat from core

[06:35:38]   (Read 7726 bytes from disk)

[06:35:38] Connecting to http://171.64.65.20:8080/

[06:35:39] Posted data.

[06:35:39] Initial: 0000; - Uploaded at ~8 kB/s

[06:35:39] - Averaged speed for that direction ~48 kB/s

[06:35:39] + Results successfully sent

[06:35:39] Thank you for your contribution to Folding@Home.

Now, that UNSTABLE MACHINE thingy after completion isn't normal, and shouldn't happen. However, it's a known problem and it appears it's a WU problem, or a driver problem, or a combination of the two.

The NVidia boys are aware of it and are working on a solution in the drivers, and the Pande guys have been informed of the same problem on a different WU in the new series, so I'd just forget it and go forth. You got at least some partial credit for it anyhow, and downloaded a 5506 again, so if it completes there's no hardware problem to worry about.

I'm going to report it for you in the forum when I get a minute.


EDIT: Done. Here's the link if you want to follow it:

http://foldingforum.org/viewtopic.ph...&t=5513#p54897