Message boards :
Graphics cards (GPUs) :
Cuda32 app?
Message board moderation
| Author | Message |
|---|---|
|
Send message Joined: 24 Dec 08 Posts: 738 Credit: 200,909,904 RAC: 0 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
A question for the devs: Is there a cuda 3.2 app in the works? The 260.89 drivers are now whql'ed. I'm not sure if the cuda 3.2 DLL's are released or still under NDA. Assuming they are released apart from fixing some bugs with cuda 3.1 they have supposedly optimised the DLL's for use on the Fermi chips. BOINC blog |
|
Send message Joined: 11 Jul 09 Posts: 1639 Credit: 10,159,968,649 RAC: 261 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]()
|
A question for the devs: Is there a cuda 3.2 app in the works? The CUDA 3.2 toolkit is still at 'release candidate' stage, according to http://developer.nvidia.com/object/cuda_archive.html. |
|
Send message Joined: 15 Feb 07 Posts: 134 Credit: 1,349,535,983 RAC: 0 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]()
|
Hi, I've started testing with 3.2RC, but don't expect anything until after the full release is made. You shouldn't have any trouble moving to post 3.1-release 260-series drivers, however. MJH |
|
Send message Joined: 24 Dec 08 Posts: 738 Credit: 200,909,904 RAC: 0 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
I've updated drivers to 260.89 at present, just waiting on the DLL's and apps to become available to use it. Older cuda apps seem to be fine on it. Presumably there will be some speed improvement for Fermi-based cards when using cuda 3.2. BOINC blog |
|
Send message Joined: 11 Jul 09 Posts: 1639 Credit: 10,159,968,649 RAC: 261 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]()
|
Similarly, I've been running 260.89 drivers for a while, in three combinations: WinXP/Fermi, WinXP/9800GT, Win7/9800GTX+. No problems, and if anything GPUGrid has been running more reliably on the 9800GT-series since the upgrade - though that may just the the variability of the tasks issued (I still can't get the HIVPR_n1 series to run on the older cards). |
skgivenSend message Joined: 23 Apr 09 Posts: 3968 Credit: 1,995,359,260 RAC: 0 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
I have been using it since it was in Beta, without too much trouble. By itself it brought no speed increase over the previous final release on a GTX470 XP x86. The dll files in 3.2 might help, but only for some tasks; for those already at 93% to 98% GPU utilization there is little improvement to be had. While the GTX460 could do with a speed bump for Windows, hopefully MJH is also trying to get a working app together for Linux users of the GTX460 and GTS450. These have similar architectures to the still unreleased GTX475 and presently don't work under Linux. |
BeyondSend message Joined: 23 Nov 08 Posts: 1112 Credit: 6,162,416,256 RAC: 0 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]()
|
Similarly, I've been running 260.89 drivers for a while, Has anyone noticed a slowdown (lower GPU usage) with v260.89? Is there a way to force the driver to use max power instead of adaptive mode? The max power option seems to have disappeared from the preferences. |
skgivenSend message Joined: 23 Apr 09 Posts: 3968 Credit: 1,995,359,260 RAC: 0 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
Yes, but I think it is the lightweight IBUCH tasks rather than the driver. I was running the latest driver when I noticed the almost halving of the GPU Utilization, so I tried the previous full driver and it did not improve. When I get the odd non-IBUCH task the utilization is much higher. Not sure about the adaptive setting making any difference for Fermis but sometimes did for CC1.3 cards. I did notice a small performance drop from the 260.89 driver to the previous beta version, but it was around 1% less so prone to inaccuracy. |
BeyondSend message Joined: 23 Nov 08 Posts: 1112 Credit: 6,162,416,256 RAC: 0 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]()
|
Yes, but I think it is the lightweight IBUCH tasks rather than the driver. I did not say less GPU utilization, I said SLOWER. After more fiddling it turns out that with 260.89 the GPU runs at the correct speed after rebooting for about 1/2 hour, then drops to about 55% shader and core speed. The only way I can get it to go back to full speed is to reboot WinXP-64. Anyone have a solution or just go back to the 258.xx drivers? I can't find a way to turn off the adaptive power setting... |
|
Send message Joined: 17 Aug 08 Posts: 2705 Credit: 1,311,122,549 RAC: 0 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
Beyond: set the windows power settings to maximum performance? SK wrote: The dll files in 3.2 might help, but only for some tasks; for those already at 93% to 98% GPU utilization there is little improvement to be had. They could still optimize routines and avoid some unneccessary calculations here and there. Not saying this is possible with the current code, just that generally even at 100% load there may still be room for avoiding calculations ;) MrS Scanning for our furry friends since Jan 2002 |
skgivenSend message Joined: 23 Apr 09 Posts: 3968 Credit: 1,995,359,260 RAC: 0 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
Thought you would pick up on that ;) In the latest full release windows version of NVidia Control Panel there is no way to set the Power management mode to “prefer maximum performance”. This is installed with the latest driver Full Release NVidia driver (260.89). I can’t presently check to see what comes with the latest Linux driver (system rebuild in progress). Power management preferences are normally found within Manage 3d settings, but were absent with the latest control panel version when I checked (Win x86, GTX470). It was an unfortunate coincidence that these frequency reductions occurred during the release of many lightweight tasks. I came across cards reducing their frequencies before (as did many) and found that by using EVGA Precision and leaving it open on the desktop my cards did not slow down, but that workaround may not work on this occasion. Going back to the previous driver might be the best option, as 260.89 brought no performance improvements, though you could try to just uninstall the NVidia Control Panel from Add/Remove Programs, and then install the Control Panel that came with the 258 driver. As it appears the 260.89 drivers are BAD for GTX260’s and/or GT240’s under Windows operating systems, I would suggest people stay away from 260.89 if they have a CC1.3 card or below. There is a 260.99 beta out now, that might bring a fix, but there are few reasons to upgrade NVidia 200 series cards for here. For anyone with this problem and wanting to go back to 258, be careful the way you uninstall the present driver and reinstall 258, or you will end up keeping the latest NVidia Control Panel. The usual safe method of going back is running a full uninstall, rebooting into safe mode and installing the older driver, though some of the latest drivers offer up more options to uninstall/install. |
BeyondSend message Joined: 23 Nov 08 Posts: 1112 Credit: 6,162,416,256 RAC: 0 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]()
|
Beyond: set the windows power settings to maximum performance? Yep, always. Last night installed the new v260.99 on 2 machines with GT 240 cards. One seems to be staying at high power while the other one is dropping to low power even faster. Once they drop to low power it seems that NOTHING will bring them back short of rebooting. |
skgivenSend message Joined: 23 Apr 09 Posts: 3968 Credit: 1,995,359,260 RAC: 0 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
For driver 260.99 the Power Management Preferences are now under the Program Settings Tab, however I was not able to set Power management preferences to Maximum; there is only the one setting, Adaptive. So avoid this Beta as well (NVidia Control Panel version 3.4.772.04). |
BeyondSend message Joined: 23 Nov 08 Posts: 1112 Credit: 6,162,416,256 RAC: 0 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]()
|
Thought you would pick up on that ;) It was an unfortunate coincidence that these frequency reductions occurred during the release of many lightweight tasks. I came across cards reducing their frequencies before (as did many) and found that by using EVGA Precision and leaving it open on the desktop my cards did not slow down, but that workaround may not work on this occasion. Good suggestion. I tried installing the NVidia Control Panel and NView from v258.96 (over v260.99). Still no power settings and still went to low power about 30 minutes after rebooting. Tried leaving MSI Afterburner and a host of other programs open to see if they would stop the down-clocking but no go. Actually v260.99 is also listed as WHQL and seems to have the same problem. I reverted to v258.96 and the problem is gone. Kind of looks like the NVidia driver team is trying to match the success of their chip designers... ;) |
skgivenSend message Joined: 23 Apr 09 Posts: 3968 Credit: 1,995,359,260 RAC: 0 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
At least the fans didn’t stop this time :) Finding good cards for this project is hard enough without trying to work out which app works best with which broken driver. GPU crunching is not easy. While this could just be bad design I’m half expecting NVidia to release their GTX475 after their driver success; It would mirror their first Fermi release – crippled cards with knobbled drivers. To see two broken WHQL released only a week apart tells me something is wrong there. Although the last Linux driver is 260.1912 released October 13th and the CUDA Toolkit is still at the 3.2 Release Candidate stage, I would not put it past them to release another unsupported card. GTX460 and GTS450 don’t work on Linux and the GTX460 is about 30% too slow on Windows, 3 months after release. |
|
Send message Joined: 17 Aug 08 Posts: 2705 Credit: 1,311,122,549 RAC: 0 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
To see two broken WHQL released only a week apart tells me something is wrong there. Could be by design. Thinking "so many people are wasting electricity by choosing high performance for no reason, so let's just make adaptive the standard" and setting the load detection wrong, so that compute doesn't wake the GPU up properly. This would be the real bug - for a company pushing GP-GPU so hard one would expect a little more understanding for the needs of the people actually buying their GPUs for crunching. And give us the Tesla drivers for desktop GPUs please, so we can manage our headless crunchers via remote desktop! MrS Scanning for our furry friends since Jan 2002 |
BeyondSend message Joined: 23 Nov 08 Posts: 1112 Credit: 6,162,416,256 RAC: 0 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]()
|
Just an update. Eventually all 3 machines on which I installed v260.89 & later v260.99 reverted to GPU low power mode when running GPUGRID with no way to return to normal speed except rebooting. After rebooting low power mode would return after ~ 30 minutes. All machines were XP-64 and all returned to preferred normal (remaining in max power mode) behavior by reverting to NVidia driver v258.96. |
|
Send message Joined: 17 Aug 08 Posts: 2705 Credit: 1,311,122,549 RAC: 0 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
low power mode when running GPUGRID with no way to return to normal speed except rebooting. What if you launch a game? Does it speed up and stay there for another 30 mins? Sorry if this has been answered before, I haven't paid much attention lately. MrS Scanning for our furry friends since Jan 2002 |
BeyondSend message Joined: 23 Nov 08 Posts: 1112 Credit: 6,162,416,256 RAC: 0 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]()
|
low power mode when running GPUGRID with no way to return to normal speed except rebooting. I don't game so can't tell you. I did try loading MANY misc apps and GPU monitoring programs to see if he GPU could be coaxed back to a higher speed, with no success at all. Only a reboot seemed to do it and then only for a short time. A REALLY bad bug IMO. |
|
Send message Joined: 11 Jul 09 Posts: 1639 Credit: 10,159,968,649 RAC: 261 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]()
|
A REALLY bad bug IMO. What has the feedback on the NVidia forums been like? |
©2025 Universitat Pompeu Fabra