NVIDIA 6.12 beta 6 GPU client Released!

Come join the... uh... er... fold.

Moderators: just brew it!, farmpuma

Re: NVIDIA 6.12 beta 6 GPU client Released!

Postposted on Sat Jun 21, 2008 12:23 pm

What I'm really liking about the client is that just like with CPU folding the GPU client gives up cycles as needed. I can leave it running, playing TF2 for a couple hours, and when I'm done still have gotten a WU nearly completed with no slow down in TF2... Might have something to do with having vsync always on, that way I'l not wasting power rendering extra frames...?
TR BBQ ~XI~ has come and gone but you can relive it by checking out this gallery!
drfish
Maximum Gerbil
Gold subscriber
 
 
Posts: 4536
Joined: Wed Jan 22, 2003 7:53 pm
Location: Zeeland, MI

Re: NVIDIA 6.12 beta 6 GPU client Released!

Postposted on Sat Jun 21, 2008 1:03 pm

FYI, the new nVidia drivers now have fan speed control in the panels. You can override the default "quiet speeds yet hot" and make it "tolerable yet cooler". Man this thing crunches work units. On my AMD X2 box I have one core on a console client, and I am letting the other core do whatever with the GPU client. It leaves plenty of cycles for getting work done!

Now I have a T-storm coming so I have to power down :-(.

-LS
liquidsquid
Minister of Gerbil Affairs
 
Posts: 2452
Joined: Wed May 29, 2002 10:49 am
Location: New York

Re: NVIDIA 6.12 beta 6 GPU client Released!

Postposted on Sat Jun 21, 2008 2:32 pm

drfish wrote:What I'm really liking about the client is that just like with CPU folding the GPU client gives up cycles as needed. I can leave it running, playing TF2 for a couple hours, and when I'm done still have gotten a WU nearly completed with no slow down in TF2... Might have something to do with having vsync always on, that way I'l not wasting power rendering extra frames...?


I dunno, that hasn't been my experience, but I haven't tried playing games with it.

I fired up the GPU2 client last night, and I'm getting weird slowdowns in normal web browsing and just cycling through programs. I tried setting the core affinity, like one person said, but that doesn't seem to help much. And the viewer can really bring a system down!

Could have something to do with the fact that I'm running it on XP x64 (using a modified .ini file)...
mongoosesRawesome
Gerbil XP
Silver subscriber
 
 
Posts: 366
Joined: Mon May 10, 2004 5:36 pm
Location: Maryland, USA

Re: NVIDIA 6.12 beta 6 GPU client Released!

Postposted on Sun Jun 22, 2008 2:18 pm

CB5000 wrote:
ReAp3r-G wrote:edit: tried installing the 177.35 CUDA-drivers and it seems that my system setup (see sig) doesn't quite like it...just gives a message saying that no drivers are to be found and setup will now exit


you have to install the 174.55 version. The 177.35 only works with 8800GTX cards AFAIK.
Download here... http://www.nvidia.com/object/get_cuda2_beta.html


No, the 177.35 only support GT200 out of the box. All other cards can be enabled/used, but you have to mod the inf or force card selection.

FWIW, on 4850 and any old driver, viewer is rock solid. Nvidia has to get working on that.
Siglessness is boring.
Image - M4800-Eight1
Image - Vargr-Z97
Forge
Lord High Gerbil
Silver subscriber
 
 
Posts: 8061
Joined: Wed Dec 26, 2001 7:00 pm
Location: SouthEast PA

Re: NVIDIA 6.12 beta 6 GPU client Released!

Postposted on Sun Jun 22, 2008 10:02 pm

I modded the .inf file and and I'm using 177.35 instead of 174.55. Now I'm getting about 3300 PPD instead of 1800 PPD with 177.35 with my 9600GT. My CPU usage for FaH went down from 60% down to 3-9%. Looks like 177.35 have a lot of optimizations to take better advantage of the GPU cause now the GPU runs about 8 degrees hotter too.

So far 26 WU's finished for Team Tech Report in about a Day!
CB5000
Gerbil XP
 
Posts: 365
Joined: Wed Mar 26, 2008 4:46 pm
Location: NW region

Re: NVIDIA 6.12 beta 6 GPU client Released!

Postposted on Sun Jun 22, 2008 10:25 pm

CB5000 wrote:My CPU usage for FaH went down from 60% down to 3-9%.


That typically shouldn't happen. F@H expects to burn a single CPU core just monitoring the GPU and talking to it on occasion. If CPU usage is low, then either the GPU is super-saturated or something is broken.
Siglessness is boring.
Image - M4800-Eight1
Image - Vargr-Z97
Forge
Lord High Gerbil
Silver subscriber
 
 
Posts: 8061
Joined: Wed Dec 26, 2001 7:00 pm
Location: SouthEast PA

Re: NVIDIA 6.12 beta 6 GPU client Released!

Postposted on Sun Jun 22, 2008 10:34 pm

Forge wrote:
CB5000 wrote:My CPU usage for FaH went down from 60% down to 3-9%.


That typically shouldn't happen. F@H expects to burn a single CPU core just monitoring the GPU and talking to it on occasion. If CPU usage is low, then either the GPU is super-saturated or something is broken.


I've seen other people using 177.35 drivers getting similar results. Besides I'm cranking out Work units almost twice as fast now... :-/ The main problem before was the huge CPU overhead and the kernel time or that red bar on the task manager. Now that red bar is tiny. I think what might have happened was that the entire workunit may have been pushed into the GPU memory instead of the system memory... as it eats a 10th of the system RAM as it did before.
CB5000
Gerbil XP
 
Posts: 365
Joined: Wed Mar 26, 2008 4:46 pm
Location: NW region

Re: NVIDIA 6.12 beta 6 GPU client Released!

Postposted on Mon Jun 23, 2008 9:52 am

Yikes, just take a look at this!

http://folding.extremeoverclocking.com/ ... s=&u=25591

Two 9600's one quad core (SMP client) and one single instance on a dual core.

Where did all of my threats go!?!?

BTW I tried the 177.35 on the quad core, and it didn't go very well. System got unstable, graphics choppy, and other weird crap. I would await the next nVidia release that actually supports 9600 properly. Running 175 right now.

Anyone have a step-by-step on how they got 177.35 to install and the GPU client to work properly? I only hacked the driver install .inf file (got it from an nVidia forum). I would love to double the computation on the GPU and free a core.

-LS
liquidsquid
Minister of Gerbil Affairs
 
Posts: 2452
Joined: Wed May 29, 2002 10:49 am
Location: New York

Re: NVIDIA 6.12 beta 6 GPU client Released!

Postposted on Mon Jun 23, 2008 10:16 am

I'm running the 9600GT with the standard, recommended driver (174.55) on Windows XP-32. I'm getting 3600ppd (if I didn't game and then watch movies). Even so, I'm picking up 2k ppd with it. I'm not sure you are going to be able to use the SLI. From what I've read around it isn't looking good for that scenario.
Sony a7
Sony Zeiss 55/1.8 SSM, 24-70/4 SSM
Minolta 100/2, 100-300 APO
TheEmrys
Minister of Gerbil Affairs
Silver subscriber
 
 
Posts: 2198
Joined: Wed May 29, 2002 8:22 pm
Location: Northern Colorado

Re: NVIDIA 6.12 beta 6 GPU client Released!

Postposted on Mon Jun 23, 2008 10:55 am

The two 9600GTs are on different machines:

Machine 1:
9600 GT, AMD X2 4200+
Machine 2:
9600GT, Intel Quad-core Q6600 @ 2.4G, this is the one that didn't like the beta drivers.
liquidsquid
Minister of Gerbil Affairs
 
Posts: 2452
Joined: Wed May 29, 2002 10:49 am
Location: New York

Re: NVIDIA 6.12 beta 6 GPU client Released!

Postposted on Mon Jun 23, 2008 12:25 pm

liquidsquid wrote:The two 9600GTs are on different machines:

Machine 1:
9600 GT, AMD X2 4200+
Machine 2:
9600GT, Intel Quad-core Q6600 @ 2.4G, this is the one that didn't like the beta drivers.


Gotcha. I have read some posts where running the SMP client (only with a quad-core) will alleviate some of those problems. Its worth a try. Also, I'd try everything on the rec'd drivers until it can be shown to be solid, then play with some of the mod'd solutions.
Sony a7
Sony Zeiss 55/1.8 SSM, 24-70/4 SSM
Minolta 100/2, 100-300 APO
TheEmrys
Minister of Gerbil Affairs
Silver subscriber
 
 
Posts: 2198
Joined: Wed May 29, 2002 8:22 pm
Location: Northern Colorado

Re: NVIDIA 6.12 beta 6 GPU client Released!

Postposted on Mon Jun 23, 2008 12:56 pm

That's what I did. Right now on the quad I have the SMP client running and the GPU client with older drivers for the GPU, and it is stable. The 177.35 caused me problems, so I had to back up.

-LS
liquidsquid
Minister of Gerbil Affairs
 
Posts: 2452
Joined: Wed May 29, 2002 10:49 am
Location: New York

Re: NVIDIA 6.12 beta 6 GPU client Released!

Postposted on Mon Jun 23, 2008 10:49 pm

I'm running a MSI 8800GTS 512MB with 177.35 drivers on Windows XP (32bit). The card is OCed a bit from the factory. I did not modify the inf file to install the drivers. Had been getting a lot of 98 pointers but now I seem to be getting WUs worth 479 points. FahMon has been reading any where from 4300 to 5000+ PPD while working on it's current WU. One core dedicated to the GPU client (core 3). The FahCore_11.exe in task manager is taking up 25% CPU usage.

I have a Q9300 with 2 VMware server instances running the SMP linux client. ATM one is set to core 0 and 1. The other set to 1 and 2 (core 1 is overlap). Seems to work well. I figured overlapping the SMP clients would give more points than running just 1 SMP and a single core client. Could be wrong but we shall see. Just didn't want anything to take away from the GPU2 client.
i3-530 | HR-01 Plus (passive) | DH55TC | 4GB Kingston DDR3 | Toshiba 250GB 2.5" HDD | Mini P180 | picoPSU 150-xt w/102w brick | 21w idle
Pegasus
Gerbil First Class
 
Posts: 161
Joined: Fri Jun 30, 2006 1:13 am

Re: NVIDIA 6.12 beta 6 GPU client Released!

Postposted on Mon Jun 23, 2008 11:16 pm

Hahaha! Just for fun I tried it out on my 8400GS:) It worked fine and didn't crash but only turns in about 450 PPD. It is in a box running an AMD X23800 (stock at 2.0) that itself gets about 515 PPD With the SMP Client!

If I had a single core PCIe box sitting around I would put it in there but for now I'm just going back to the SMP Client. Nice to know that it worked, though!
Nothing clever.
Lockheed_Tvr
Gerbil
 
Posts: 87
Joined: Thu Dec 23, 2004 1:30 pm
Location: San Diego

Re: NVIDIA 6.12 beta 6 GPU client Released!

Postposted on Mon Jun 23, 2008 11:37 pm

Lockheed_Tvr wrote:Hahaha! Just for fun I tried it out on my 8400GS:) It worked fine and didn't crash but only turns in about 450 PPD. It is in a box running an AMD X23800 (stock at 2.0) that itself gets about 515 PPD With the SMP Client!

If I had a single core PCIe box sitting around I would put it in there but for now I'm just going back to the SMP Client. Nice to know that it worked, though!


with which version of the CUDA drivers?
mongoosesRawesome
Gerbil XP
Silver subscriber
 
 
Posts: 366
Joined: Mon May 10, 2004 5:36 pm
Location: Maryland, USA

Re: NVIDIA 6.12 beta 6 GPU client Released!

Postposted on Mon Jun 23, 2008 11:44 pm

mongoosesRawesome wrote:
Lockheed_Tvr wrote:Hahaha! Just for fun I tried it out on my 8400GS:) It worked fine and didn't crash but only turns in about 450 PPD. It is in a box running an AMD X23800 (stock at 2.0) that itself gets about 515 PPD With the SMP Client!

If I had a single core PCIe box sitting around I would put it in there but for now I'm just going back to the SMP Client. Nice to know that it worked, though!


with which version of the CUDA drivers?


Version 175.19 running on XP SP2
Nothing clever.
Lockheed_Tvr
Gerbil
 
Posts: 87
Joined: Thu Dec 23, 2004 1:30 pm
Location: San Diego

Re: NVIDIA 6.12 beta 6 GPU client Released!

Postposted on Tue Jun 24, 2008 7:05 am

Lockheed_Tvr wrote:
mongoosesRawesome wrote:
Lockheed_Tvr wrote:Hahaha! Just for fun I tried it out on my 8400GS:) It worked fine and didn't crash but only turns in about 450 PPD. It is in a box running an AMD X23800 (stock at 2.0) that itself gets about 515 PPD With the SMP Client!

If I had a single core PCIe box sitting around I would put it in there but for now I'm just going back to the SMP Client. Nice to know that it worked, though!


with which version of the CUDA drivers?


Version 175.19 running on XP SP2


You may have a higher PPD with the newer 177 drivers.
mongoosesRawesome
Gerbil XP
Silver subscriber
 
 
Posts: 366
Joined: Mon May 10, 2004 5:36 pm
Location: Maryland, USA

Re: NVIDIA 6.12 beta 6 GPU client Released!

Postposted on Tue Jun 24, 2008 1:23 pm

Just figured out how to install 177 drivers on a 8400. Restated and will have point numbers pretty soon!
Nothing clever.
Lockheed_Tvr
Gerbil
 
Posts: 87
Joined: Thu Dec 23, 2004 1:30 pm
Location: San Diego

Re: NVIDIA 6.12 beta 6 GPU client Released!

Postposted on Tue Jun 24, 2008 1:43 pm

Ok! 177 drivers on an 8400GS:

Big improvement! Jumped from around 450PPD to about 560PPD both on the 98 point WUs. Good enough to scrap SMP, add a console client on one core and get a nice bump in points!
Nothing clever.
Lockheed_Tvr
Gerbil
 
Posts: 87
Joined: Thu Dec 23, 2004 1:30 pm
Location: San Diego

Re: NVIDIA 6.12 beta 6 GPU client Released!

Postposted on Tue Jun 24, 2008 1:56 pm

Lockheed_Tvr wrote:Ok! 177 drivers on an 8400GS:

Big improvement! Jumped from around 450PPD to about 560PPD both on the 98 point WUs. Good enough to scrap SMP, add a console client on one core and get a nice bump in points!

You mean your dual core CPU is doing less than 500ppd? My "lowly" S939 3800+ @2.4GHz is doing 800. So to me it's not quite good enough?
Image
The Model M is not for the faint of heart. You either like them or hate them.

Gerbils unite! Fold for UnitedGerbilNation, team 2630.
Flying Fox
Gerbil God
 
Posts: 24584
Joined: Mon May 24, 2004 2:19 am

Re: NVIDIA 6.12 beta 6 GPU client Released!

Postposted on Tue Jun 24, 2008 2:12 pm

Flying Fox wrote:
Lockheed_Tvr wrote:Ok! 177 drivers on an 8400GS:

Big improvement! Jumped from around 450PPD to about 560PPD both on the 98 point WUs. Good enough to scrap SMP, add a console client on one core and get a nice bump in points!

You mean your dual core CPU is doing less than 500ppd? My "lowly" S939 3800+ @2.4GHz is doing 800. So to me it's not quite good enough?


My S939 3800+ is lowlier(sp?) than yours :D Mine has never been able to overclock. I am using a number of overclock techniques just to get it stable at 2.0. So... with the memory at very loose timings and the chip only at 2.0, I just never got very good PPD.

I actually have a Pentium D805 that turns in significantly more points!

The 3800 was my first dual core machine and my main rig for a long time. But I was always disappointed in it :(
Nothing clever.
Lockheed_Tvr
Gerbil
 
Posts: 87
Joined: Thu Dec 23, 2004 1:30 pm
Location: San Diego

Re: NVIDIA 6.12 beta 6 GPU client Released!

Postposted on Tue Jun 24, 2008 2:15 pm

Is there an option where you allow large work units on the SMP client? Maybe Lockheed_Tvr has that disabled, and that's why he only gets 500 PPD with the SMP client.

The lower clock speed should only account for 20% difference. The difference that FF shows is MUCH higher.

How are you measuring the PPD? Using FAHmon?
mongoosesRawesome
Gerbil XP
Silver subscriber
 
 
Posts: 366
Joined: Mon May 10, 2004 5:36 pm
Location: Maryland, USA

Re: NVIDIA 6.12 beta 6 GPU client Released!

Postposted on Tue Jun 24, 2008 2:22 pm

A few things we need to sort out first:
  1. I have to admit, I'm using the X2's "magical" 64-bit VM on 32-bit host support and is running the Linux SMP client based on notfred's diskless folding stuff. The Linux SMP client is known to do more ppd than the Windows version
  2. I don't play a lot of games so the cycles are mostly used for Folding, except when some bad Flash ad decides to eat up a whole core
  3. My ppd number was obtained on arguably the most generous WU, the 2605's. Strange is that that Linux VM never gets 2665's while my Pentium E2160 gets 2665's now all the time.

We are not exactly comparing the same thing (GHz difference aside), unless we can get some common 2605 numbers.

Aside: the X2 3800+ is supposed to be 2GHz at stock, if it was not stable why didn't you RMA it before?
Image
The Model M is not for the faint of heart. You either like them or hate them.

Gerbils unite! Fold for UnitedGerbilNation, team 2630.
Flying Fox
Gerbil God
 
Posts: 24584
Joined: Mon May 24, 2004 2:19 am

Re: NVIDIA 6.12 beta 6 GPU client Released!

Postposted on Tue Jun 24, 2008 2:25 pm

mongoosesRawesome wrote:Is there an option where you allow large work units on the SMP client? Maybe Lockheed_Tvr has that disabled, and that's why he only gets 500 PPD with the SMP client.

The lower clock speed should only account for 20% difference. The difference that FF shows is MUCH higher.

How are you measuring the PPD? Using FAHmon?


I am using FAHmon for the numbers.

I do have it set up to allow large work units and I get the same WUs that I get on my quads and other dual core machines.
Nothing clever.
Lockheed_Tvr
Gerbil
 
Posts: 87
Joined: Thu Dec 23, 2004 1:30 pm
Location: San Diego

Re: NVIDIA 6.12 beta 6 GPU client Released!

Postposted on Tue Jun 24, 2008 2:31 pm

Flying Fox wrote:A few things we need to sort out first:
  1. I have to admit, I'm using the X2's "magical" 64-bit VM on 32-bit host support and is running the Linux SMP client based on notfred's diskless folding stuff. The Linux SMP client is known to do more ppd than the Windows version
  2. I don't play a lot of games so the cycles are mostly used for Folding, except when some bad Flash ad decides to eat up a whole core
  3. My ppd number was obtained on arguably the most generous WU, the 2605's. Strange is that that Linux VM never gets 2665's while my Pentium E2160 gets 2665's now all the time.

We are not exactly comparing the same thing (GHz difference aside), unless we can get some common 2605 numbers.

Aside: the X2 3800+ is supposed to be 2GHz at stock, if it was not stable why didn't you RMA it before?


Ahhhh. This explains a lot of the difference. I am running the windows smp client and it is my work machine so while I don't exactly do big number cruching on it, it is in use for regular day to day stuff.

I tried Notfred's diskless folding suite on the aformentioned Pentium D805 and it jumped from ~700PPD to ~900PPD but it was not stable and would crash hard after 1 or 2 WUs so I went back to windows. I keep meaning to try it again but I haven't gotten around to it.

I think the performance difference between the linux and windows clients accounts for most of the point difference. If I get some free time maybe I'll play around with VMware.

As for RMAing it, I wasn't folding or pushing it very hard when I first got it and it took almost a year before I started to blame the random crashes on anything but Microsoft :D Even then, It took a lot of swapping ram and power supplies, etc. before I started to narrow it to the chip. By then I just sort of thought, oh well.
Nothing clever.
Lockheed_Tvr
Gerbil
 
Posts: 87
Joined: Thu Dec 23, 2004 1:30 pm
Location: San Diego

Re: NVIDIA 6.12 beta 6 GPU client Released!

Postposted on Tue Jun 24, 2008 11:29 pm

mongoosesRawesome
Gerbil XP
Silver subscriber
 
 
Posts: 366
Joined: Mon May 10, 2004 5:36 pm
Location: Maryland, USA

Re: NVIDIA 6.12 beta 6 GPU client Released!

Postposted on Wed Jun 25, 2008 6:40 am

So when are we going to talk about building folding nodes on nothing but graphics cards?

Yea lets see, 50$ for an el cheapo athlon 64 at 2.4ghz, 50 bucks for 2gb of ram, say 100 for a motherboard, give or take, 50$ for an efficient PSU to supply the folding beast, mount it all on a plank of wood and spend the rest on your graphics card. Lets say... 9800gtx. Oh yea, and the windows operating system volume license from a guy you know down the street in redmond WA.
round 550, 600 including tax for a 4000ppd machine?
Mr.Lif
Gerbil
 
Posts: 81
Joined: Sun May 06, 2007 8:27 pm

Re: NVIDIA 6.12 beta 6 GPU client Released!

Postposted on Wed Jun 25, 2008 7:58 am

Mr.Lif wrote:Yea lets see, 50$ for an el cheapo athlon 64 at 2.4ghz
Not quite sure yet, but how about a $30 Sempr0n? May not be fast enough though.

Mr.Lif wrote:50 bucks for 2gb of ram
Do we know if we really need 2 gigs of RAM?

Mr.Lif wrote:say 100 for a motherboard
How about $50 ($40 AR) for a motherboard?

Mr.Lif wrote:50$ for an efficient PSU to supply the folding beast, mount it all on a plank of wood and spend the rest on your graphics card. Lets say... 9800gtx. Oh yea, and the windows operating system volume license from a guy you know down the street in redmond WA.
round 550, 600 including tax for a 4000ppd machine?

;)

With the ppd and WUs still in flux we won't know the proper numbers. Also we need more data from the down models to hit that price/performance sweet spot.
Image
The Model M is not for the faint of heart. You either like them or hate them.

Gerbils unite! Fold for UnitedGerbilNation, team 2630.
Flying Fox
Gerbil God
 
Posts: 24584
Joined: Mon May 24, 2004 2:19 am

Re: NVIDIA 6.12 beta 6 GPU client Released!

Postposted on Wed Jun 25, 2008 11:29 pm

As soon as AMD submits the updated SP scheduler to Pande (right now RV770 PPD are in the toilet due to the SP scheduler which is designed for R6X0 not scaling to 800SPs) I'll be back into it.

GPU folding doesn't upset my VMware the way CPU folding does.
Siglessness is boring.
Image - M4800-Eight1
Image - Vargr-Z97
Forge
Lord High Gerbil
Silver subscriber
 
 
Posts: 8061
Joined: Wed Dec 26, 2001 7:00 pm
Location: SouthEast PA

Re: NVIDIA 6.12 beta 6 GPU client Released!

Postposted on Thu Jun 26, 2008 7:36 am

Has there been any news on a version for those of us who run 64-bit operating systems?
|Intel Q9450 @3.6 1.425v|Lapped TRUE|Gigabyte EP45-UD3P|Mushkin & Patriot 2x2gb DDR2-900|EVGA 8800 GTS 512 750/1873/1000|Antec 900|Corsair 520HX|Asus Xonar DX|
13895 3DMarks

Image
themattman
Gerbil Elite
Silver subscriber
 
 
Posts: 792
Joined: Fri Jun 29, 2007 8:34 pm
Location: PA

PreviousNext

Return to TR Distributed Computing Effort

Who is online

Users browsing this forum: Exabot [Bot] and 3 guests