Folding Forum

Only Core 21 has issues, not core 17 or 18.

fredex wrote:I just installed a Nvidia update to 375.26. right after reboot, FAH GPU failed on two core A21 WUs, then got an A18 WU which it is now processing. Does that imply a (partial??) fix? or not?

No, if you go far enough back in this topic you will find that the issue only affects Core_21 WU's. The older Core_18 does not use the OpenCL code that was broken by the version 375 and 376 drivers. So the next Core_21 WU will also likely fail.

As has already been said, Core_18 does not exhibit the same types of failures that seem to only happen with {375.xx AND Core_21}. Thus we have no way of knowing whether we are waiting for a fix to Core_21 or a fix to OpenCL for 375.xx or to both.

At this point, my hunch is that it'll be both.

I'm feeling very smug at the moment as I've managed to hack 373.06 to get my GTX 1050 Ti working. It turned out to be rather simpler than the instructions pointed to suggested as I was able to copy the appropriate lines from the equivalent 375 file then paste them into 373.06 with a bit of minor editing. It is necessary to disable Driver Signature checking under W8 but that's hardly difficult.

At the moment it's 7% into a 11710 and is giving 220k PPD compared with 140k PPD from a GTX 950 so a worthwhile upgrade as long as it keeps going.

Care to share how?

Okay, the important thing is that the device code for the 1050 Ti is 1C82 (the 1050 is 1C81).
Open nv_dispi.inf from 373.06 and any 375/6 version in your favourite text editor. Find a line similar to
"%NVIDIA_DEV.1C82% = Section133, PCI\VEN_10DE&DEV_1C82 " in the newer file then copy and paste it into the older file after each instance of a similar line with 1C03 which is the code for the 6GB GTX 1060.
Change the Section number to match that of the 1C03 line each time.
Finally, there's a line near the end of the file that maps device ID to device name so copy that across too.
Save the modified file and install 373.06.

Snapshot

Thanks for the Great tutorial. It's working!! Currently Folding a P11403 work unit core 0x21
for 172,000 PPD on a Zotac GTX 1050Ti. I'm at 5% and it has not crashed.

I picked this card because it has a unique design. While the card itself is standard width - taking up two spaces on the motherboard.. the backplane is designed to take up just One.
So it can be installed in some of those older Antec and CoolerMaster chassis - the ones that had the little alignment Nubs between each rear card port. Effectively stopping the installation of any double wide cards. (unless you cut the little spacer nubs off with a Dremel)

To expand on your directions - I needed to post that "%NVIDIA_DEV.1C82.... data line in three or four different places in the nv_dispi.inf file.
Also, the newly modded drivers would not install until I rebooted the system. (though that might just be my issue)

A question for you software wiz kids.... I applied this modification to the nv_dispi.inf file of the 32-bit version of the Windows 7/8 373.06 drivers. Does anyone think I could just replace that .inf file in the 64-bit version of the same drivers - and have it work?

I wouldn't think that would be a totally safe assumption. I would use winMerge or a similar program to run a compare if you wanted to be sure.

FC from the command line is an old favourite.....
A compare will fail anyway as the file for the 64-bit drivers says 64-bit on its first line. I supposed the 32-bit file could be edited to say 64-bit but it's going to be much safer just to do the edit again.

I've been using the latest 376.33 with my 780 and my 1080 and so far i've yet to run into any problems.

With that older card though, are you getting core_21 work units or just 18?

For those of you who are folding on RHEL/Centos Linux, the kind folks at elrepo have compiled the previous working Nvidia driver for RHEL/CentOS 7.3. This resolves the repeated failures on core 0x21 work units. (If you do install that version, be sure to --exclude it when using yum to update your system subsequently, else it will just reinstall the bad one.)

look for kmod-nvidia-367.57-2.el7.elrepo.x86_64.

Fred

Folding on my ASUS STRIX 980GTX on the current 376.33 and had no issues so far in the past couple of days, and that's folding on the core_21 units

Using Windows 10 btw and have made no configuration changes to the FAH Client or anything else for this.

I haven't tested this against any gaming yet, but rather incredulously, I found an easy workaround for myself that I hadn't thought of before for some reason. Basically, I copied nvopencl.dll from my SysWow64 directory with 373.06 installed onto my desktop. Then I installed the latest drivers (376.33) and overwrote the new DLL with the old one. Then I rebooted my computer, which is a requirement of this process, and now I appear to be folding core_21 WUs without any issues. Sharing the 373.06 DLL from my GDrive for convenience. I should point out that while there is also a copy of nvopencl.dll in system32, the one that FAH (or any application) is going to read is going to be which ever one(s) are located in HKEY_LOCAL_MACHINE\SOFTWARE\WOW6432Node\Khronos\OpenCL\Vendors for 64-bit systems. Obviously, it *will* be the one in System32 no matter what for 32-bit systems.

Follow-up: I have performed CompuBench and 3DMark benchmarks. OpenCL and CUDA both appear to be perfectly stable with this tweak, and I completed a core_21 WU earlier this evening. I believe this to be a viable tweak.

Saw some people having success with the latest driver so i tried using 376.33. F@H failed immediately.