WARNING Do not upgrade to 375/376.xx drivers (for xx<48)

[foldy]

Does this mean I can run install the latest driver on my Windows machine for use with F@H? Not sure what this is saying.

Yes that is what I'm saying, the new latest Windows nvidia 378 driver includes the folding@home fix from previous hotfix driver.

[bruce]

I guess the question becomes: Does it include the performance degradation noted in the hot-fix driver?

[ComputerGenie]

I guess the question becomes: Does it include the performance degradation noted in the hot-fix driver?

Since I had 2 projects that were both ~50% complete I figured it was the best time to compare.
2x GTX1080 @ 2025MHz

My results:
376.48
Project:13500 - TPF: 1:31
Project:11710 - TPF: 2:06

378.49
Project:13500 - TPF: 1:27
Project:11710 - TPF: 2:02

[foldy]

Oh, new whql driver 378.49 is little faster than hotfix driver 376.48! And compared to previous good driver 373.06?

[neognomic]

Yes, using it for windoz 81 and it still fails:
ERROR:exception: Error downloading array interactionCount: clEnqueueReadBuffer (-5)
FS01:FahCore returned: BAD_WORK_UNIT (114 = 0x72)
This AM it continued failing about 6 times then it gave up and "FAILED" the FAH-GPU folding core. Normally, it stutters like that 3 or 4 times but finds a unit, folds and completes.

Operating System: Windows 8.1 Pro, 64-bit
DirectX version: 11.0
GPU processor: GeForce GTX 1060 6GB
Driver version: 376.33

My only questions are:
Were the FAH-GPU WU that have been completed and uploaded any good?
Does anyone know if the new (2 days) nVidia 378.49 driver actually resolves the issue(s)? nVidia claims it is fixed "[Folding @ Home] Work unit errors occur. [200262613/1831430]"
Does this occur for Linux 375.26 driver also?

[Joe_H]

Yes, using it for windoz 81 and it still fails:
ERROR:exception: Error downloading array interactionCount: clEnqueueReadBuffer (-5)
FS01:FahCore returned: BAD_WORK_UNIT (114 = 0x72)
This AM it continued failing about 6 times then it gave up and "FAILED" the FAH-GPU folding core. Normally, it stutters like that 3 or 4 times but finds a unit, folds and completes.

If it fails a few WU's then gets one it can process, that WU is probably using Core_18. Folding fails on Core_21 projects with these versions of the drivers. nVidia did release a hot fix version, 376.48, that has been reported to mostly work with Core_21 projects.

Operating System: Windows 8.1 Pro, 64-bit
DirectX version: 11.0
GPU processor: GeForce GTX 1060 6GB
Driver version: 376.33

My only question are:
Were the FAH-GPU WU that have been completed and uploaded any good?
Does anyone know if the new (2 days) nVidia 378.49 driver actually resolves the issue(s)? nVidia claims it is fixed "[Folding @ Home] Work unit errors occur. [200262613/1831430]"
Does this occur for Linux 375.26 driver also?

The WU's completed for Core_18 projects should have been good. Reports are just coming in for 378.49, so a bit more time is needed before saying it is completely fixed.

The Linux version 375 drivers have been reported to have the same problem with Core_21 projects as the Windows drivers.

[neognomic]

Thank you!
I got the 378.49 and will try it ...

The Linux version 375 drivers have been reported to have the same problem with Core_21 projects as the Windows drivers.

Bummer. Linux is my main OS.
nVidia is so slow with Linux drivers' fixes: the 378.09 is still beta. ...hopefully for not much longer.

[ComputerGenie]

Oh, new whql driver 378.49 is little faster than hotfix driver 376.48! And compared to previous good driver 373.06?

I had a saw blade tooth drop a carbide tip, so I didn't get my racks made yet. Over the weekend I'll finish the testing I intended to do today/tomorrow.
To be as apples-to-apples as possible, I'll compare 369.30(Cuda install), 373.06, 376.48, and 378.49.
Will post back as soon as I know...

[SombraGuerrero]

Hopefully Nvidia will merge the fix into the final Linux package, because as of the beta, they haven't. As for Windows, yes, the new driver works. It merges in the workaround they've had in since hotfix build 376.48.

[foldy]

@ComputerGenie: Testing the different driver versions performance may be obsolete because a new FahCore_21 update is currently in beta and will be released "soon".

[ComputerGenie]

@ComputerGenie: Testing the different driver versions performance may be obsolete because a new FahCore_21 update is currently in beta and will be released "soon".

Not a fan of holding my breath waiting for "soon".
That being said, since .18 is still not pushed to all betas, it's worth testing since I'm doing a clean install on this new rig (if for no other reason than "because we can").

[bruce]

Not a fan of holding my breath waiting for "soon".

This time, I would. The fact that the new core version has been placed in beta is a really good sign -- and the topic for reporting problems has been quiet. Don't forget that this problem severely impacts scientific production -- mostly because FAH can't depend on everybody with a GPU installing the hot-fix. Neverthelss, I'm not involved in the decision to release the workaround ... just observing what's happening.

[snapshot]

A quick test had a P9178 running at 3:00 per frame where 363.06 gave 2:30 and a P10494 running at 10:30 instead of 9:00. Not surprisingly, I've gone back to 363.06. I'll be interested to see if ComputerGenie sees the same.

[ComputerGenie]

Not a fan of holding my breath waiting for "soon".

This time, I would. ... just observing what's happening.

I'm not trying to imply that it will not happen, but as a programmer, I know what nightmarish setbacks can come from even one "small" fix being needed. Coupling that with what PPD I may get at time x vs what PPD I could get today are separate things.

[neognomic]

Folding fails on Core_21 projects with these versions of the drivers.

As I said I would I installed the 378.49 nVidia driver for GTX1060 on Win 8.1. FAH-GPU has been running without errors since then and, from the log, folding only Core 21.

I know that this is probably the wrong place to post that since this is essentially a useless/dead thread but do not see a "378.xx" thread to post 378.49 results.