Hi there,

as the title say's, every Nvidia driver after version 337.88 keeps failing on my system with constantly restarts. I have absolutely no clue what the problem is, so maybe someone can give me a clue, what's the problem? :? The system: Core2Quad Q9550S on ASUS P5B-E running stock settings, just a bit undervolted to 1,1375 V. That should not be the problem, cause is also occurs on standard voltage (1,2V). Any ideas? :?

[edit]
Ok, i've forgot to mention, my system is a hybrid system with a PCI-E card with GT 640 and a PCI(yes, PCI!)-card with GT 430. Both cards are running absolutely perfect on WinXP(!!), but also on Win7Pro with driver 337.88. Every driver after that version fails with timeouts or other odd stuff. I admit, this is a strange system to run, but to the specs it should be running fine! :?

---

Aloha, Uli

I'm running 344.11 on both of my systems, with no problems.
C2D E6600, Win Vista 32bit, 2*GTX 750Tis.
I7 2600, Win7 64bit, 2*GTX 750Tis.
Lunatics, MB only.


I'm thinking running something like Latency Monitor might be worth a go, or DPC Latency Checker.
Random spikes in system latency caused by a dodgy driver could be the issue.

---

Grant
Darwin NT

The Q9550S and chipset described, is definitely within a time period of Intel Drivers that could be stuck on some dicey versions. Manually looking at the dates via device manager could turn up something. I would check DPC Latencies with DPC Latency Checker, then if looking dicey under load LatencyMon to help isolate.

In my case chipset was stuck on some ~2009 or so Intel chipset drivers, despite running detecttion and update tools. Manual update of the drivers (CHipset, PCIe and RAID) was necessary.

There are also several NVidia services introduced since, that I get soem benefit turning off, mainly Shadowplay and the Network STreaming servince (both relatively new)

Also I'd recommend for that system, go through and turn off the recent MS spyware (backported from Win10). Spybot Anti Beacon + disabling specific diagnostic tracking services as a subset of what's needed on 8/8.1/10

All combined those things made a huge difference on my main dev system of similar Generation, though individual effects were mostly minor.

---

"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.

Hi there,

thanks a lot for your answers and a happy new year to all! :)

I installed the latest chipset drivers directly from intel (Version: 9.2.0.1030) to no avail...

Here the results from LatencyMon:


Here the results from DPC Latency Checker:


I don't get it, one prog say's it's screwed, the other say's it's ok! :?

[important edit]
I only installed the Nvidia driver and the Physx driver, nothing else, no HD-Audio, nor Nvidia experience etc...

[edit 2]
Power management is off, i. e. the system is running on full throttle.

---

Aloha, Uli

Hey Bro

I would try taking off Hyperthreading and bump the voltage a bit. 1.25 should be good
Make sure speedstep is off
Are you reserving a core for each NV task?
Your combiation is a bit curious. I would be thinking the driver is getting a bit messed up controlling both PCI-E and PCI
Try removineg the 430 and see if that helps

Also the sound component may be the problem. Reinstall driver as Admin an do a clean custom install with everything except Nvidia experience

Just suggestions Bro

I have had no problems running 337.88 on my 460
I also run a little ati HD and no problems

Michael

I don't get it, one prog say's it's screwed, the other say's it's ok! :?

My guess, one is just checking for DPC (Deferred Procedure Calls) latency, they're OK.
The other is checking for other latencies, it's the one showing issues.
Or one was running at a time issues were occurring, the other wasn't.

As Jason mentioned in his post- while the video driver is having issues, those issues can be caused by other drivers poor performance impacting on the video driver.

I installed the latest chipset drivers directly from intel (Version: 9.2.0.1030) to no avail...

I'd suggest going the board manufacturer's web site & getting their specific drivers for that board.

---

Grant
Darwin NT

Hi again,

i tried the chipset drivers directly from ASUS with the exact same result.

I even disabled AV software and the TV streaming device, always the same result. The moment i start BOINC, the latency issues start. All system drivers (except the NVidia driver, which is version 377.88) are up to the latest. I think, i just have to live with the 337.88 version for the graphics card. :?

---

Aloha, Uli

If DPC Latency issues are showing up in the tools, it can take some digging indeed.

LatencyMon, with some practice, should reveal the exact driver or process sticking. That can be a result of corrupted or bodgy driver, or something physically failing.

---

"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.

Ok, i've forgot to mention, my system is a hybrid system with a PCI-E card with GT 640 and a PCI(yes, PCI!)-card with GT 430. Both cards are running absolutely perfect on WinXP(!!), but also on Win7Pro with driver 337.88. Every driver after that version fails with timeouts or other odd stuff. I admit, this is a strange system to run, but to the specs it should be running fine! :?

Quick easy test you can do.
First remove the PCI card and see how the system runs, then replace the PCI card and remove the PCIe card and repeat the test.
I had a bad experience trying to mix newer and older NVidia cards on the same machine on Win 7-64, especially with a PCI card in the mix. I believe that you're really limiting the 640's performance weighing it down with that 430. One driver, but I think two different code streams....
Just a thought, easy to test and see if that removes the symptoms you're seeing.
Regards, Jim ...

---

Hmmm, I wonder what would do that.... Oh Crud, yes unified memory model in later drivers+OS will probably be dog slow over PCI, hence deprecation of Pre-Fermis (as well)

---

"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.

Hi again,

i tried both options, only the PCI-E and only the PCI card. On both choices the system behaves the same, only on the PCI card it is crunching slower (obviously).

After that i tried an older driver version 327.23, which was running very very good on WinXP.
Now it got really interesting: BOINC told in the log, it was running one GPU task on GT 640 and the other GPU task on the GT 430. In REALITY it ran both on the GT 640 and i couldn't do anything to get it working correctly. There was never such an issue with WinXP. Meanwhile i think, the Windows 7 driver interface is completely borked in comparison to WinXP and i finally regret taking the upgrade to Win 7. Things were so smooth on XP...

...back to version 377.88 and i will have to live with that mess. :((

---

Aloha, Uli

Hi again,

believe it or not, but the latest driver from Nvidia (version 361.43) is running fine again with my unusual configuration.
Definitely something must have changed under the hood - strange...

---

Aloha, Uli

Maybe it's related to this (excerpt from BOINC log):

03/01/2016 17:42:16 | | CUDA: NVIDIA GPU 0: GeForce GT 640 (driver version 361.43, CUDA version 8.0, compute capability 3.0, 2048MB, 1969MB available, 692 GFLOPS peak)
03/01/2016 17:42:16 | | CUDA: NVIDIA GPU 1: GeForce GT 430 (driver version 361.43, CUDA version 8.0, compute capability 2.1, 512MB, 460MB available, 269 GFLOPS peak)
03/01/2016 17:42:16 | | OpenCL: NVIDIA GPU 0: GeForce GT 640 (driver version 361.43, device version OpenCL 1.2 CUDA, 2048MB, 1969MB available, 692 GFLOPS peak)
03/01/2016 17:42:16 | | OpenCL: NVIDIA GPU 1: GeForce GT 430 (driver version 361.43, device version OpenCL 1.1 CUDA, 512MB, 460MB available, 269 GFLOPS peak)

Do you see the difference? ;)

---

Aloha, Uli

Maybe it's related to this (excerpt from BOINC log):

03/01/2016 17:42:16 | | CUDA: NVIDIA GPU 0: GeForce GT 640 (driver version 361.43, CUDA version 8.0, compute capability 3.0, 2048MB, 1969MB available, 692 GFLOPS peak)
03/01/2016 17:42:16 | | CUDA: NVIDIA GPU 1: GeForce GT 430 (driver version 361.43, CUDA version 8.0, compute capability 2.1, 512MB, 460MB available, 269 GFLOPS peak)
03/01/2016 17:42:16 | | OpenCL: NVIDIA GPU 0: GeForce GT 640 (driver version 361.43, device version OpenCL 1.2 CUDA, 2048MB, 1969MB available, 692 GFLOPS peak)
03/01/2016 17:42:16 | | OpenCL: NVIDIA GPU 1: GeForce GT 430 (driver version 361.43, device version OpenCL 1.1 CUDA, 512MB, 460MB available, 269 GFLOPS peak)

Do you see the difference? ;)

The 430 will only do 1.1 for OpenCL as that was all it was coded for.

---

The 430 will only do 1.1 for OpenCL as that was all it was coded for.

Yes, finally the two GPUs are handled differently, as it should be and therefore they work well together again! *happysmilesagain* :)

---

Aloha, Uli

There are reports coming in that the 361.43 driver in combination with Windows 8.1 and Windows 10 and BOINC (7.6.22) starting up causes blue screens of death with the error: KERNEL_SECURITY_CHECK_FAILURE.

Returning to a previous driver, like 359.06, fixes this in all cases.

---

Thanks for the warning!

I run that driver on Windoze 7 64 bit without any issues.

My Windoze 8.1 laptop is running driver 354.56 without issues on an NVS 5200M chip, but i was tempted to try version 361.43, which is still considered Beta for the NVS 5200M. Thanks for the heads up, i will avoid that version on my laptop.

---

Aloha, Uli