mersenneforum.org  

Go Back   mersenneforum.org > Great Internet Mersenne Prime Search > Hardware > GPU Computing

Reply
 
Thread Tools
Old 2012-06-25, 21:57   #1
Chuck
 
Chuck's Avatar
 
May 2011
Orange Park, FL

2·5·83 Posts
Default Nvidia GPU driver level

Does it really matter whether or not I keep up with the "latest and greatest" Nvidia driver level? I noticed this weekend that Nvidia is into the 300 series (I hadn't checked for quite a while).

I went ahead and upgraded to 301.42 (have a GTX 580). I then also had to update my EVGA monitor and fan control utility since the old version didn't work anymore.

Now everything is up to date, but it doesn't seem to make any difference in the performance of mfaktc.

Chuck
Chuck is offline   Reply With Quote
Old 2012-06-25, 22:04   #2
Dubslow
Basketry That Evening!
 
Dubslow's Avatar
 
"Bunslow the Bold"
Jun 2011
40<A<43 -89<O<-88

3×29×83 Posts
Default

Not really, the only difference it will generally make is if you get an mfaktc version that was compiled for a higher CUDA version, then you'll need new drivers. There's a small chance some revision might give a minor performance increase, but I'd say the chance is small enough that updating every six months or so is fine. (You really shouldn't leave it longer than that anyways.)
Dubslow is offline   Reply With Quote
Old 2012-06-25, 23:09   #3
Batalov
 
Batalov's Avatar
 
"Serge"
Mar 2008
Phi(3,3^1118781+1)/3

22·5·11·41 Posts
Default

Judging from primeGrid forums, there were known problems with ver.295 and similar 29x versions. You want to jump from 285 to 300's (they don't seem to have a bad rep). 300's contain CUDA 4.2.
Batalov is offline   Reply With Quote
Old 2012-06-26, 03:55   #4
Xyzzy
 
Xyzzy's Avatar
 
"Mike"
Aug 2002

750610 Posts
Default

We set up our GPUs and drivers a year ago and haven't changed them. It is too risky and what we have works. It would take a huge improvement in throughput for us to consider changing anything.
Xyzzy is offline   Reply With Quote
Old 2012-06-26, 06:36   #5
LaurV
Romulan Interpreter
 
LaurV's Avatar
 
Jun 2011
Thailand

205268 Posts
Default

Quote:
Originally Posted by Xyzzy View Post
We set up our GPUs and drivers a year ago and haven't changed them. It is too risky and what we have works. It would take a huge improvement in throughput for us to consider changing anything.
Good thinking. If it works, don't fix it!
LaurV is offline   Reply With Quote
Old 2012-06-26, 19:24   #6
Batalov
 
Batalov's Avatar
 
"Serge"
Mar 2008
Phi(3,3^1118781+1)/3

233C16 Posts
Default

For 6xx cards, that evil must be necessary though.
But for 5xx's, absolutely.
Batalov is offline   Reply With Quote
Old 2012-06-26, 20:22   #7
Xyzzy
 
Xyzzy's Avatar
 
"Mike"
Aug 2002

2·33·139 Posts
Default

FWIW, we are using the 270.61 driver.
Xyzzy is offline   Reply With Quote
Old 2012-06-26, 21:41   #8
Batalov
 
Batalov's Avatar
 
"Serge"
Mar 2008
Phi(3,3^1118781+1)/3

22·5·11·41 Posts
Default

A few years ago, I had huge conflicts between some NVidia drivers (I believe, 180-ish to early 200's) and a Java-based game client to a (board-type) game server. The black and white checkers/chess pieces would flicker between black and white state at every mouse event and they would stay in a random state. It was a disaster! ...Downgraded the drivers back (to something like 169) then and all went to normal.


Non-sequitur: Gaaah! The [xy] editing button is added! Yeah! Yeah! Oh, yeaaaah.
Batalov is offline   Reply With Quote
Old 2012-06-26, 21:53   #9
Dubslow
Basketry That Evening!
 
Dubslow's Avatar
 
"Bunslow the Bold"
Jun 2011
40<A<43 -89<O<-88

3×29×83 Posts
Default

Quote:
Originally Posted by Batalov View Post
Non-sequitur: Gaaah! The [xy] editing button is added! Yeah! Yeah! Oh, yeaaaah.
Wooooooooooooooooooo!!!!!!!!!
Dubslow is offline   Reply With Quote
Old 2012-06-26, 22:55   #10
Dubslow
Basketry That Evening!
 
Dubslow's Avatar
 
"Bunslow the Bold"
Jun 2011
40<A<43 -89<O<-88

3·29·83 Posts
Default Gah!

I just went from 295 to 302 on Linux, and my on-screen usability with CUDALucas has taken a fairly major hit.
Dubslow is offline   Reply With Quote
Old 2012-08-05, 18:26   #11
kladner
 
kladner's Avatar
 
"Kieren"
Jul 2011
In My Own Galaxy!

22×33×7×13 Posts
Default Driver follies

I am putting this account here because driver versions seemed to play a big role in it. I owe part of the problems' resolution to Batalov's remark above about avoiding 29x series drivers. Thanks Serge!

I just went through several days of trying to clear up what I've come to see as either a mixed-version driver conflict, or a combination of that with different MB BIOS versions.

I described the beginnings of this situation here http://mersenneforum.org/showpost.ph...8&postcount=38

The symptom I described as rising Sieve Primes actually was part of substantial loss of GPU throughput. A few times since then some combination of BIOS version and multiple Uninstall/reinstalls of the drivers restored the usual level of performance. I kept trying, though, in hopes of having the latest BIOS, and possibly a later driver. This combination never worked out.

Meanwhile, I had been trying to dig out where the lost performance was going. My experience with mfaktc has been that when it slows down, something else is stealing CPU time from sieving. Some searching and reading of various help sites put me onto a utility called Process Explorer.
http://technet.microsoft.com/en-us/s.../bb896653.aspx
It showed that when the performance hit was going on, "Hardware Interrupts and DPCs" was consuming 5-12% of CPU time. By this time I had also realized that P95 was affected as LL iteration went from .045-.047 to .068-.070.

On the occasions when things were working right this Interrupts load was only a fraction of a percent. When things were working badly, the high Interrupt load came and went depending on whether mfaktc was running.

I have now given up on the most recent BIOS, though I wanted the fix to HDMI sound it contains. I rolled it back to one a couple of years old. I finally got 285.62 driver running correctly after several attempts which included uninstalls and registry sweeps before installations. Interestingly, the current install is from the Express option. I had been doing Clean Install as a matter of course, but Express got me a good install.

I should mention that some of the above was precipitated by the addition of a GTX 570 to the box, and trying to get it all playing nice with the existing 460. It was marginally stable until I ran a 12 gauge extension cord from the back of the apartment to the front where my computer is. Getting off an overloaded circuit got me from barely 90v on really hot days when the power company was barely keeping up, up to 109-112v.

That helped, but stability was still poor. The Kill a Watt meter told me that full load was drawing just under 650w--the rating of the PSU. This seemed like a likely source of trouble, so I bit the bullet and replaced the (OCZ) PC Power and Cooling 650 with a well-reviewed OCZ 750w. This has settled everything down nicely, though I could do without the red LEDs glaring out the rear of my case.

Speaking of the case, I got temps under control by cutting a 120mm hole in the side panel right beside the GPUs. A filtered 2000 rpm fan blowing in on their mid-sections made a world of difference!
kladner is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
NVIDIA TITAN 320.59 driver with cuda 5.5 Manpowre GPU Computing 43 2013-08-22 12:28
Possible problems with nVidia 320.18 driver kladner GPU Computing 0 2013-06-15 15:33
New NVIDIA driver 310.70 - slower mfaktc? Chuck GPU Computing 3 2013-01-10 21:21
nVidia Driver causes error message? Nave Hardware 8 2004-08-11 18:15
Nvidia driver problem Sideshow Bob Software 4 2004-02-13 13:39

All times are UTC. The time now is 21:20.

Sun May 24 21:20:05 UTC 2020 up 60 days, 18:53, 0 users, load averages: 1.15, 1.13, 1.18

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2020, Jelsoft Enterprises Ltd.

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.