Does the driver version have to be quite that high? I'm running cuda60 tasks - successfully, so far as I can tell - with driver 335.28: driver 343 or above would only be needed to support cuda65.

CUDA60 will stay around for the time being, but won't get any of performance updates I am anticipating for CUDA65.

Matt

We make the scheduling decision on the major version of the driver only. Later minor releases of earlier major versions have back-ported CUDA 6.5 support but we ignore that as it complicates the scheduling code too much.

Matt

Your 340.58 is obviously too old for CUDA 6.5, but will be fine for 6.0 for a while. Maybe by the point 6.5 becomes appreciably better than 6.0 your Ubuntu folks may provide a recent driver. Otherwise you may want to manually install the driver directly from nVidia.

MrS

Your 340.58 is obviously too old for CUDA 6.5, but will be fine for 6.0 for a while. Maybe by the point 6.5 becomes appreciably better than 6.0 your Ubuntu folks may provide a recent driver. Otherwise you may want to manually install the driver directly from nVidia.

MrS

Hello: I think it does not have good information, Nvidia 340.58 driver for Linux is the latest released on November 5, 2014 and evidently supports Cuda 6.5 ... do not understand the problem GPUGRID with Linux and CUDA

NVIDIA driver 343.22 is the most recently supported non-beta Linux driver. I have been using this driver and I have no exceptions. A search on the NVIDIA web site for the most recent driver for the non-900 series card will result in 340.58 being the most recent supported driver for 700 series card and lower, however this web page for 343.22 driver (http://www.geforce.com/drivers/results/77844) states this driver supports NVIDIA cards going back to the 400 series, among others.

It would appear NVIDIA web site has not been properly updated to add the 343.22 driver for supported NVIDIA cards, other than the 900 series.

This is because the app isn't refined enough to adequately distinguish between sub 343 drivers that support 6.5 and those that do not. The reason is that they don't all support cuda 6.5 equally, and Matt cannot code, test and debug for every minor driver release that pops up (for every OS).

So, I have this card:

23/11/2014 07:34:14 | | CUDA: NVIDIA GPU 0: GeForce GT 640 (driver version 314.07, CUDA version 5.0, compute capability 3.0, 2048MB, 1966MB available, 692 GFLOPS peak)

I understand it will no longer crunch for GPUGrid now, that's right? If I update the driver, will it work?

Many thanks.

So, I have this card:

23/11/2014 07:34:14 | | CUDA: NVIDIA GPU 0: GeForce GT 640 (driver version 314.07, CUDA version 5.0, compute capability 3.0, 2048MB, 1966MB available, 692 GFLOPS peak)

I understand it will no longer crunch for GPUGrid now, that's right? If I update the driver, will it work?

Many thanks.

Hi, I had to install the driver for Linux, Nvidia 346.16 Beta to accomplish tasks 8.46-Cuda 6.5.

By the way this version allows control of voltage and OC of the cores as well as the manual fan control. Greetings.


NOTE:The voltage I still can not make it work ... we'll see.

Hi, I had to install the driver for Linux, Nvidia 346.16 Beta to accomplish tasks 8.46-Cuda 6.5.

We had to discontinue cuda 4.2 apps because of some time limit in the executable.
As they were going to be discontinued anyway, we will only reintroduce it if really necessary.

If you were using cuda4.2, update your driver


* Discontinued CUDA 4.2
We need CUDA 6 to properly support the new Maxwell, so I'll be retiring the old CUDA 4.2 build. If you are still receiving version 4.2 WUs, please consider updating your driver to version 343 or above.

Will Quadro driver v341.05 be enough with CUDA drivers?
http://www.nvidia.co.uk/download/driverResults.aspx/79245/en-uk


EDIT: for Quadro tesla, all v340+ drivers should work! Just downloaded new tasks... ;)

Will Quadro driver v341.05 be enough with CUDA drivers?

Yes 341.05 will suffice. As stated by the webpage to which you link, the quadro 341.05 driver features CUDA 6.5. My Quadro K4000 is now getting tasks again after upgrading to 341.05.

From what I understand, the GPUs that recently were excluded from the latest GeForce driver sets... are still supported with R340 driver releases until April 2016.
http://nvidia.custhelp.com/app/answers/detail/a_id/3473

I even have an open NVIDIA Bug with them right now, where R340 (up to and including current 341.05 for my FX3800 M)... broke some of their OpenCL calculations that were working just fine in R337. They are supposedly working to fix it.
http://setiathome.berkeley.edu/forum_thread.php?id=75633&postid=1597385

Will Quadro driver v341.05 be enough with CUDA drivers?

Yes 341.05 will suffice. As stated by the webpage to which you link, the quadro 341.05 driver features CUDA 6.5. My Quadro K4000 is now getting tasks again after upgrading to 341.05.

Here v342.05 will only get CUDA 6.0 tasks due to application restrictions. You'll need v343.22 or above for CUDA 6.5 in Linux. For Windows v344.65 is the latest WHQL driver and it runs fine. Might as well use it.

The drivers come from the same Release families (R337, R340, R343, etc.). And my point was, for the GPUs that have been deprecated from the R343 drivers.... The R340 drivers will have bugfix support for them until April 2016, for both Quadro and GeForce series of those R340 drivers.

341.05 is the latest R340 driver release for my Quadro FX3800 M, and as I said, I have an open bug with NVIDIA - they completely broke how the GPU calculates certain OpenCL functions, and they should hopefully have it fixed it in the upcoming R340 release. So far as I know, CUDA is calculating fine, but I have not personally tested it.

Regards,
Jacob

Many thanks.

I had trobule with the new driver. It locked the videos I tried to watch. I returned to the older drive and the problem ended, and this is because I still use an old driver.

I am not sure if will be able to use the newer driver. Has anyone had the same problem here? If so, how was it solved?

Many thanks,
Valter.

Many thanks.

I had trobule with the new driver. It locked the videos I tried to watch. I returned to the older drive and the problem ended, and this is because I still use an old driver.

I am not sure if will be able to use the newer driver. Has anyone had the same problem here? If so, how was it solved?

Many thanks,
Valter.

My computer with this graphics board used to get only CUDA 4.2 workunits.

12/5/2014 9:12:47 AM | | CUDA: NVIDIA GPU 0: GeForce GTX 560 (driver version 344.75, CUDA version 6.5, compute capability 2.1, 1024MB, 852MB available, 1116 GFLOPS peak)
12/5/2014 9:12:47 AM | | OpenCL: NVIDIA GPU 0: GeForce GTX 560 (driver version 344.75, device version OpenCL 1.1 CUDA, 1024MB, 852MB available, 1116 GFLOPS peak)

After 4.2 workunits were discontinued, it got one 6.0 workunit, which failed, and it now gets no GPU workunits from GPUGRID at all, only the CPU test workunits. What do I need to change in order the get GPU workunits again?

Hi Robert,

I tried to put together a comprehensive list of what is necessary to receive GPU workunits. You can either do step 1 and post back the results or take a look at your settings for steps 2-5 and see if anything needs to be changed.

- Happy crunching,
Steve

---------------------------------------------------------
1. Turn on the <work_fetch_debug> flag to see what BOINC
is doing during work fetch.
http://boinc.berkeley.edu/wiki/Client_configuration

-stop / restart boinc and post back with what the event
log says after it tries to get new work from GPUGrid.

- this will help identify if your issue is one of the
topics below or if we need to look deeper.
---------------------------------------------------------
2. Computing Preferences web
http://www.gpugrid.net/prefs.php?subset=global
Suspend GPU work while computer in use = no
- if yes then used in conjunction with ...
'In use' means mouse/keyboard activy in last "x" minutes
Suspend GPU work while computer in use = no
----------------------------------------------------------
3. Computing Preferences local (takes precendece over web)
Boinc Manager - Advanced View (ctrl+shift+a)
Tools menu --> Computing preferences
same as above but if you have set local preferences
then you will be always running local preferences and
nothing you chnage on the website for preferences will
matter until after you press the 'Clear' button. Keep in
mind this is "BOINC wide" setting so the last site you
chnaged this on across all of the projects you are
connected to takes precedence.
---------------------------------------------------------
4. Project preferences
http://www.gpugrid.net/prefs.php?subset=project
- Use NVIDIA GPU = on
- ACEMD short runs
----------------------------------------------------------
5. Activity Status
Boinc Manager - Advanced View (ctrl+shift+a)
Activity menu
- Suspend = off
- If 'Run based on preferences' = on then see Computing Preferences above
- Suspend GPU = off
- If 'Use GPU based on preferences' = on then see Computing Preferences above
---------------------------------------------------------

I am running Nvidia driver 344.75 for my Titan Z, which is obviously not a 900 series gpu. I have noticed issues with WU's and it does not seem to be functioning as it was before with an older driver and my sli 780's. Any advice?
____________

I am running Nvidia driver 344.75 for my Titan Z, which is obviously not a 900 series gpu. I have noticed issues with WU's and it does not seem to be functioning as it was before with an older driver and my sli 780's. Any advice?

I am running Nvidia driver 344.75 for my Titan Z, which is obviously not a 900 series gpu. I have noticed issues with WU's and it does not seem to be functioning as it was before with an older driver and my sli 780's. Any advice?

I tried uninstalling BOINC client and reinstalling, letting BOINC benchmark run, then running GPU Grid to no avail--I am getting about 70% through long-run WU's and then crash! Worst part is, I seem to be starting over on WU's when i reboot the system and not completing the WU's that were almost finished. This has to be a driver issue, because my sli 780's were running fine on 340.52 driver, and I did not update to driver 344.75 until I installed my Titan Z. I have not tried my Titan Z on 340.52 because I heard GPU Grid will not give me any WU's, but I know this new driver has to be the issue!

Will i get work units if I roll my 344.75 driver back to 340.52, the last known good working configuration I had? 340.52 driver is only 6-8 months old, but I cannot get 344.75 to run. As soon as i start boinc client, my driver crashes, restarts, then it will let me run wu's, but then the whole system crashes 3/4 of the way through long run wu's.

Are you doing any overclocking? Also, are you making sure to keep your GPU temps below 70*C? Does it error in Heaven 4.0 at maximum settings for 5 hours? Have you tried downclocking it until it stops erroring?

Will i get work units if I roll my 344.75 driver back to 340.52, the last known good working configuration I had? 340.52 driver is only 6-8 months old, but I cannot get 344.75 to run. As soon as i start boinc client, my driver crashes, restarts, then it will let me run wu's, but then the whole system crashes 3/4 of the way through long run wu's.

I have tried downclocking, default clocks and overclocking, same results. Temps never go above 65c with aggressive fan profile, cpu is overclocked to 4.715 Ghz, fx8350, but thats the same overclock i have always used for gpu grid, and previously with my sli 780's

I have not ran haven benchmark for 5 hours, but have run full benchmarks with no issues, as well as normal gameplay with no issues. The system does not become unstable until I start up BOINC client, as far as I can tell anyway

You can try other 344 dirvers:
344.11
344.48
344.60
344.65
Did you checked all power connectors in your PC (including the 24 pin MB power connector) for burn marks, or loose connections?
Do you use any power connector converters, or extension wires inside your PC?

EDIT: you can try uninstalling the GPU driver, then use driver sweeper to remove any remaining driver files before installing different drivers.

There is only 1 Titan Z present in that host, as dual GPU boards are always treated by BOINC like 2 GPUs. So if there were 2 Titan Z's in that host, the host's page should show "[4] NVIDIA GeForce GTX TITAN Z (4095MB) driver: 340.52".

Yes Zoltan. Although there were two computers, it's just one.
I'm just curious how many of one Titan z is able to be calculated rac for one day.
If everything works fine.. Because that man is here for the first with that monster GF card :-)
Suferbus has similar problems as when here came some new model gf card into the project. Also there fell nvidia driver and the task. Had nothing overclocked and the clean install os there .. Project scientists probably surely do not have a few Titans Z to test job before releasing tasks out.
If it works this way. I do not know. -))

Hi! I'm a fairly ignorant user of BOINC and so, I don't understand if this information affects me. I have an NVidia gtx 750 Ti SC graphics card, which, to my understanding, uses Maxwell architecture) and its drivers are always up to date. Are these the drivers you're referring to that should be updated? If not, please provide me with a fairly "for dummies" explanation of how to update said Cuda drivers.

Thank you very much! :)
____________

Hi! I'm a fairly ignorant user of BOINC and so, I don't understand if this information affects me.

I have an NVidia gtx 750 Ti SC graphics card, which, to my understanding, uses Maxwell architecture) and its drivers are always up to date.

Are these the drivers you're referring to that should be updated?

If not, please provide me with a fairly "for dummies" explanation of how to update said Cuda drivers.

Thank you very much! :)

Hi Robert,

I tried to put together a comprehensive list of what is necessary to receive GPU workunits. You can either do step 1 and post back the results or take a look at your settings for steps 2-5 and see if anything needs to be changed.

- Happy crunching,
Steve

4. Project preferences
http://www.gpugrid.net/prefs.php?subset=project
- Use NVIDIA GPU = on
- ACEMD short runs

I have not ran haven benchmark for 5 hours, but have run full benchmarks with no issues, as well as normal gameplay with no issues. The system does not become unstable until I start up BOINC client, as far as I can tell anyway

A program you might want to install to make it easier to see your CPU and GPU temperatures:

http://efmer.com/b/