All it takes is for the ComputeCache to get corrupted and then compute is done for and your cache will just empty out on instant errors. Only solution is to stop BOINC, delete the ComputeCache and reboot.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

Last night i noticed that my machine with the 397.31 driver had 6 GPU tasks "waiting to run" which means they had started and been "postponed" This is a fairly new machine and has never had that before.

So I rolled back to 391.35 and the problem seems to have gone.

Also I use this machine for gaming and yesterday evening I was totally unable to get Fallout 4 to start, now however having the old driver back it started with no problems.

I think perhaps with 397.31 your mileage may vary.

I installed 397.31 yesterday and so far no problems here.

Apparently there are problems with this driver.

https://hothardware.com/news/nvidia-fix-recent-buggy-drivers-geforce-gtx-1060

---

Kevin

I had to reinstall it using GeForce on my Windows 10 PC. After the first installation all Einstein@home GPU tasks crashed after a few seconds.
Tullio

---

Greetings,

The fixed driver can be downloaded here: GeForce Hotfix Driver Version 397.55.

Siran

---

CAPT Siran d'Vel'nahr - L L & P _\\//
Winders 11 OS? "What a piece of junk!" - L. Skywalker
"Logic is the cement of our civilization with which we ascend from chaos using reason as our guide." - T'Plana-hath

Mike, thanks for your response.

I've had no issues since reverting to an older Nvidia driver either (approximately 3 days ago). Yes, time will tell. Both of the episodes of errors occurred within 36 hours of installing the ver. 391.35 drivers. I see a driver hotfix is out but I'm going to stick with the current version that I'm using until the official new version is released to test. Best of luck to you!

---

All it takes is for the ComputeCache to get corrupted and then compute is done for and your cache will just empty out on instant errors. Only solution is to stop BOINC, delete the ComputeCache and reboot.

Thanks for your assistance Keith. I want to verify which ComputeCache folder to delete in case there could be more than one that I didn't locate. It would be the one in the path "...Users\username\AppData\Roaming\NVIDIA\ComputeCache" correct?

---

Yes, that is correct. Just delete the folder and all the subfolders when BOINC is not running and you are not doing anything else compute wise with the cards. The Nvidia driver will recreate the folder upon reboot. And the first compute work creates the necessary compute primitives for each task all automatically. The ComputeCache contains all the the CUDA and OpenCL primitives. I have had corruption on both CUDA special app work and SoG OpenCL work. As long as the necessary drivers and capabilities are called out at the beginning of the BOINC event log and the cards are seen, any stderr.txt error message on a failed task alluding to missing capability or failure to initialize CUDA or OpenCL almost always points to ComputeCache corruption or system pointers that have gone missing. Easiest solution is to delete the ComputeCache and reboot.

You have to be fast to catch the problem because on fast systems with lots of cards, it only takes a couple of minutes to completely error out your cache as each task start errors out in seconds and then grabs another task to repeat the failure.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

Thank you Keith for your very clear explanation! I appreciate it.

---

latest driver and just in time compilation feature can give you a boost.

Petri, are you referring to how the current Nvidia drivers work? Or to your latest special app with the embedded CUDA 9.1?

NVIDIA apps in general. OpenCL too. New driver can have a new compiler and new optimisations. OpenCL (SoG) is compiled whenever a driver changes. CUDA 9.1 version has pre-compiled code for existing hardware and new hardware is supported through JIT compilation.

Petri

---

To overcome Heisenbergs:
"You can't always get what you want / but if you try sometimes you just might find / you get what you need." -- Rolling Stones

Thanks for the lucid explanation Petri. I got it all in one take. Really appreciate your efforts to maximize our hardware's potentials.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)