Cuda 3.0 beta RUNNING

Moderators: slegrand, Site Moderators, PandeGroup

Re: Cuda 3.0 beta RUNNING

Postby toTOW » Wed Nov 25, 2009 10:04 am

I can confirm what andromeda says : Fahcores have always used more CPU time on Wine than on native Windows ... that's probably an overhead introduced by translation of calls from Wine and the wrapper ...
Folding@Home beta tester since 2002. Folding Forum moderator since July 2008.

FAH-Addict : latest news, tests and reviews about Folding@Home project.

Image
User avatar
toTOW
Site Moderator
 
Posts: 8931
Joined: Sun Dec 02, 2007 10:38 am
Location: Bordeaux, France

Re: Cuda 3.0 beta RUNNING

Postby willmore » Wed Nov 25, 2009 5:54 pm

Andromeda,

Thanks for the sanity check. I didn't keep good stats on the box that used to host this board and do my linux/wine folding. Plus, that was a dual core and this is a single core proc. (AMD X2 64 5600+ vs Sempron 140) I'm running this headless, so there's no X for me to notice the smoothness of. I'm glad to hear that has improved. It was a huge pain under XP on my desktop. Win7 with Aero enabled is tolerable with the 190 drivers. I'm curious if things will improve there as well with the 195 series.

Cheers.
Image
willmore
 
Posts: 16
Joined: Wed Oct 15, 2008 7:26 pm

Re: Cuda 3.0 beta RUNNING

Postby sydney » Wed Nov 25, 2009 10:20 pm

Still doesn't solve the problem that when the CPU client is running also, the GPU client barely gets and CPU time and therefore it suffers great PPD loss.

Even with throttled CPU client the GPU client isn't getting nearly as much ppd as it did on <2.6.30 kernels.
Image
E8500 @ 4.0 GHz, 8800GTX @ 1500 shader, 8800GTS (G92) @ 1800 shader
Debian 5.0 Lenny with wine CUDA wrapper.
sydney
 
Posts: 17
Joined: Mon Mar 16, 2009 7:56 pm

Re: Cuda 3.0 beta RUNNING

Postby willmore » Thu Nov 26, 2009 3:51 am

Sydney,

I'm not sure I'm following you. I'm using F12 which has a .31 kernel. I'm getting no PPD loss for the GPU applicaiton, but I'm getting 12-15% loss on the CPU task. But, I'll live with that for now. What are you experiencing?

Cheers.
willmore
 
Posts: 16
Joined: Wed Oct 15, 2008 7:26 pm

Re: Cuda 3.0 beta RUNNING

Postby Viper666 » Fri Nov 27, 2009 10:30 am

toTOW wrote:I can confirm what andromeda says : Fahcores have always used more CPU time on Wine than on native Windows ... that's probably an overhead introduced by translation of calls from Wine and the wrapper ...

No way to cheat in linux, I can show a pic of the current windows GPU using 30% ofcourse for some reason the windows task manager doesnt see the usage, but a 3rd party tool does !The processor is abit underclocked also!
Last edited by Viper666 on Fri Nov 27, 2009 10:35 am, edited 1 time in total.
ImageImage
Powered by AMD and Open SUSe 11.2 ----The New servers have nothing to do with fixing the stats problem !
User avatar
Viper666
 
Posts: 229
Joined: Fri Dec 14, 2007 9:57 pm

Re: Cuda 3.0 beta RUNNING

Postby Viper666 » Fri Nov 27, 2009 10:34 am

Anyway i came to post this i have looked ages for something like it for MB that couldnt be over clocked in the bios

LINUX OVER CLOCK TOOL for CPU's it will work on server hardware 2

http://www.ztex.de/misc/index.e.html
User avatar
Viper666
 
Posts: 229
Joined: Fri Dec 14, 2007 9:57 pm

Re: Cuda 3.0 beta RUNNING

Postby cichlid » Sat Nov 28, 2009 11:57 pm

My experience with 64 bit UBUNTU 9.10.

Q6600
PNY 9800-GT
Full GUI desktop.
All current updates
Kernel 2.6.31-15
CUDA Driver 64 bit 195.17
Cuda Toolkit 32-bit 2.3. 3.0 would install, but , but gave error messages when starting the client about not finding files and libraries.

It does run, ...... sort of.

Frame time is good for core 11 project, 1 min. 5 sec. Client setting is nice - 19. Config setting is Idle.

Any other desktop activities are pointless however. If I click with the mouse to open a terminal, browser, or other application however, there is a 30 sec lag while the display updates. That is, it takes 30 sec to open a terminal or start a firefox session. Forget using firefox. Sstop the client and everything works fine.

According to TOP, there is about 12% CPU usage by the core 11. NO other Processes using CPU. I can see on the system monitor graph that opening a terminal or app, causes a prolonged increase in CPU usage.

This was just an experimental side by side install of Karmic, so I can always go back the 9.04.

Thoughts??
cichlid
 
Posts: 37
Joined: Wed Sep 03, 2008 3:12 am

Re: Cuda 3.0 beta

Postby Mindmatter » Tue Dec 01, 2009 9:32 am

toTOW wrote:With current version of the Wine wrapper, it will probably not work on Linux : the wrapper only work with 180.xx drivers.


The wrapper is specific for F@H not for CUDA. If Stanford starts using a CUDA call not in the wrapper then it will break the wrapper. If Nvidia breaks backward compatibility in CUDA then it will break support for F@H on the wrapper and native on Windows. What happened here is that any driver over 180.xx had a bug that broke something on the Linux side. I'm not sure why this never carried over to Windows as CUDA and the Nvidia video driver share over 97% of the code across platforms. But the wrapper has nothing specific about it that ties it to any one driver unless Nvidia stops backward compatibility. The reason there is a version 2.0 wrapper was due to the beta team contacting us and letting us know that core 14 has some new functions in it that were not supported in the old wrapper. This may happen again with GPU3 but I'm hoping for a native client by then.
Mindmatter
 
Posts: 48
Joined: Tue May 27, 2008 1:53 pm

Re: Cuda 3.0 beta

Postby bruce » Tue Dec 01, 2009 8:18 pm

Mindmatter wrote:
toTOW wrote:This may happen again with GPU3 but I'm hoping for a native client by then.


GPU3 will use OpenMM, which will probably produce an effective native client. (Your shell will probably become unnecessary when GPU3 is released.) From what I understand, OpenMM was first developed as a shell between the FahCore and CUDA/CAL, but the last post I saw said that OpenMM is being modified so that it talks directly to the drivers that support OpenCL. Thus it is expected to support most current generation GPUs on most OSs (any which support OpenCL -- not just those that use CAL or CUDA).

I'm not implying anything that minimizes what you've accomplished or the usefulness of the shell in its current form. It is certainly a useful tool for the current GPU2 code.
bruce
 
Posts: 21413
Joined: Thu Nov 29, 2007 10:13 pm
Location: So. Cal.

Re: Cuda 3.0 beta RUNNING

Postby Skiesare » Wed Dec 02, 2009 4:29 pm

cichlid

I had exactly the same thing using a similar set-up, but strangely it seems to be OK if I also run the CPU SMP client. The desktop is available for a second or two at a time with about 30 seconds of freezing in-between but once I have the CPU client running all is quite normal.

It isn't a proper solution but it works for now.
Skiesare
 
Posts: 17
Joined: Mon Aug 25, 2008 6:46 am

Re: Cuda 3.0 beta RUNNING

Postby cichlid » Thu Dec 03, 2009 4:25 am

Skiesare wrote:cichlid

I had exactly the same thing using a similar set-up, but strangely it seems to be OK if I also run the CPU SMP client. The desktop is available for a second or two at a time with about 30 seconds of freezing in-between but once I have the CPU client running all is quite normal.

It isn't a proper solution but it works for now.


Thanks for the suggestion, I might try that.
cichlid
 
Posts: 37
Joined: Wed Sep 03, 2008 3:12 am

Re: Cuda 3.0 beta RUNNING

Postby shatteredsilicon » Thu Dec 03, 2009 1:24 pm

sydney wrote:Still doesn't solve the problem that when the CPU client is running also, the GPU client barely gets and CPU time and therefore it suffers great PPD loss.

Even with throttled CPU client the GPU client isn't getting nearly as much ppd as it did on <2.6.30 kernels.


I never experienced this problem. With a high-res timer and tickless kernel with a server (high latency high throughput) scheduler, I used to run 6x GPU clients and an SMP client, and never lost any PPD. I was running the SMP client at niceness 19 and the GPU clients at niceness 15. Having said that, I was running 2.6.24.x kernels at the time, so there could be a bug on the newer kernels.
Image
1x Q6600 @ 3.2GHz, 4GB DDR3-1333
1x Phenom X4 9950 @ 2.6GHz, 4GB DDR2-1066
3x GeForce 9800GX2
1x GeForce 8800GT
CentOS 5 x86-64, WINE 1.x with CUDA wrappers
shatteredsilicon
 
Posts: 699
Joined: Tue Jul 08, 2008 2:27 pm

Re: Cuda 3.0 beta RUNNING

Postby rjwilmsi » Thu Dec 03, 2009 11:42 pm

Just to confirm that on opensuse 11.2 64-bit GPU folding doesn't work with the 190.x drivers in the NVIDIA repository, but the 195.22 betas (64 bit) with 32-bit CUDA 2.x work fine.
rjwilmsi
 
Posts: 7
Joined: Wed Nov 12, 2008 12:22 am

Re: Cuda 3.0 beta RUNNING

Postby Skiesare » Mon Dec 07, 2009 7:36 am

Having been running this for a little while now, I can say that the display freezing problem only seems to happen sometimes and I can't see a pattern yet but I don't think that it is dependent on the unit it's folding.

As for PPD dropping when using the CPU SMP client, I have seen a very great drop. For instance, folding a 5913 unit was taking about 26 minutes per 1% when running both clients but with the CPU client stopped the time per 1% fell to just over 5 minutes. Clearly with that sort of difference it makes no sense to run the CPU client as well.

I have also noticed that when running the GPU client the display is slowed down. Window effects are slow and video playback is very poor, so it's not something you want running all the time.

I'm using CUDA 2.1 because when I tried to use CUDA 3 I got a missing file error when setting up the wrapper. At least two people in the tread say they are using 3, so how did they do it? That could be the solution.
Skiesare
 
Posts: 17
Joined: Mon Aug 25, 2008 6:46 am

Re: Cuda 3.0 beta RUNNING

Postby shatteredsilicon » Mon Dec 07, 2009 3:17 pm

Have you niced the processes to suitable priorities (GPU clients to 10-15 and CPU/SMP clients to 19)? Are you running a kernel with high-res timer support?
shatteredsilicon
 
Posts: 699
Joined: Tue Jul 08, 2008 2:27 pm

PreviousNext

Return to unOfficial Linux GPU (WINE wrapper) (3rd party support)

Who is online

Users browsing this forum: No registered users and 1 guest

cron