An announcement about a new Core21 version is posted today on the Folding Forum:
- New Core21 v0.0.18 with NVIDIA bug workaround released!
 

by JohnChodera (Pande Group Member) » Mon Jan 30, 2017
 
We are pleased to announce the release of Core 21 v0.0.18 to full FAH.
 
This release contains a workaround for a change made in the NVIDIA OpenCL driver introduced in drivers starting with 375.57, as well as some improvements to error handling codes.
 
While a slight reduction in performance may be observed, this performance regression should be eliminated when NVIDIA removes the hotfix in forthcoming driver updates.
 
There is a known issue that this release will, for some projects, print sporadic messages to the log due to a known visualizer bug:
 
05:34:52:WARNING:FS01:Size of positions 64914 does not match topology 19
 
This issue does not affect the quality of the science, and a workaround is already in testing for v0.0.19, which we hope to release in a week.
 
Thank you all for your patience.
 
~ The FAH Core 21 crew

 
EDIT, March 27:
Update from JohnChodera on the Folding Forum, see:
- New core21 0.0.18 with NVIDIA bug workaround released! [Update]

by JohnChodera » Mon Mar 27, 2017
 
UPDATE: NVIDIA win driver 378.66 should be the first version where the NVIDIA FAH hotfix has been removed.
 
We have a test build of 0.0.20---which fixes the visualizer problems---in testing right now.
 
Thanks for your patience!
 
~ The FAH Core 21 crew

Still glad the released a new version of Core 21 v0.0.18 to full FAH.
This release contains a workaround for a change made in the NVIDIA OpenCL driver introduced in drivers starting with 375.57, as well as some improvements to error handling codes.
This would explain why I have to Folders now under the Client Folder.
\ProgramData\FAHClient\cores\fahwebx.stanford.edu
\ProgramData\FAHClient\cores\web.stanford.edu
Thanks for the update DutchForce  & Welcome Back!

Has anyone rolled back to one of the previously known effected drivers to test?
 
d

I'm waiting for our Beta testers to report the over all effect - Happy to stay with 372.90 - I'm not gaming on the Folding PCs
 
Thanks DutchForce

drougnor
bcav - Not really needing any specific version, was just curious if any of the known 'broken' ones has been tested against the new core. 

We should see an improvement on all drivers from 376.09 forward, so I guess just pick one and testing Pascal, as that was the primary 'break' would be fine.

Thanks!
 
d

I am using Driver 378.49 now so I can test the last WHQL Version 376.33 and report back.
 

tank1023
I've been folding on the latest drivers for two weeks and only had one failed WU.

If 376.33 then you have tested?
If that is Driver 378.49 then this Driver has NVIDIA's Fix and not the Core 21 Fix.
I have had a few fail as well on Driver 378.49 but it was due too to high of an overclock on the GPU.

PPD is over 164,000 and is a little higher then Driver 378.49.
Test Running now GPU One  GPU Two both Pascal Cards the links are to My HFM Site
 

DutchForce
Just updated to driver 376.09 and Core21 v0.0.18 works, no issues so far.
A few (early) results:
GTX 980 Ti - P11711 (Core21) is ~6% lower PPD (compared to my old driver 368.81)
GTX 1070  - P9151   (Core18) is ~5% lower PPD

Do you think that the 376.09 Driver is better to use then the 376.33 Driver?
Starting 376.33 Driver testing on my Maxwell GPU Rigs.
So All my Maxwell and Pascal Rigs are running under Driver 376.33
Testing some BOINC GPU Tasks ATM.

Thanks DutchForce  I will keep moving with 376.33
Forgot to clear out the old files under the old Core 21 file.
Cleaned them out and now running 376.33 on the new Core 21.
So at least this test show the Old Core 21 was still failing on Driver 376.33.
  My HFM Running All NVIDIA GPU ATM.
*********************** Log Started 2017-01-31T23:09:28Z ***********************
23:10:17:WU00:FS00:0x21:ERROR:exception: Error downloading array interactionCount: clEnqueueReadBuffer (-5)
23:10:18:WU01:FS01:0x21:ERROR:exception: Error downloading array interactionCount: clEnqueueReadBuffer (-5)
23:10:19:WU03:FS03:0x21:ERROR:exception: Error downloading array interactionCount: clEnqueueReadBuffer (-5)
23:10:21:WARNING:WU00:FS00:FahCore returned: BAD_WORK_UNIT (114 = 0x72)
23:10:21:WARNING:WU01:FS01:FahCore returned: BAD_WORK_UNIT (114 = 0x72)
23:10:22:WARNING:WU03:FS03:FahCore returned: BAD_WORK_UNIT (114 = 0x72)
23:10:30:WU00:FS00:0x21:ERROR:exception: Error downloading array interactionCount: clEnqueueReadBuffer (-5)
23:10:33:WARNING:WU00:FS00:FahCore returned: BAD_WORK_UNIT (114 = 0x72)
23:10:34:ERROR:WU03:FS00:Exception: Server did not assign work unit
23:10:40:WU04:FS01:0x21:ERROR:exception: Error downloading array interactionCount: clEnqueueReadBuffer (-5)
23:10:43:WARNING:WU04:FS01:FahCore returned: BAD_WORK_UNIT (114 = 0x72)
23:10:47:WU03:FS00:0x21:ERROR:exception: Error downloading array interactionCount: clEnqueueReadBuffer (-5)
23:10:49:WARNING:WU03:FS00:FahCore returned: BAD_WORK_UNIT (114 = 0x72)
23:10:51:WU01:FS03:0x21:WARNING:Console control signal 1 on PID 6276
23:10:51:WU02:FS02:0x21:WARNING:Console control signal 1 on PID 4204

Just got snagged by this Panda Group update - Core21 to version 0.0.18
 
FahCore returned: CORE_OUTDATED (110=0x6e) Took the GPU into some sort of STUCK - Limbo
 
As my WU for the other two GPUs were almost done, I chose "pause finish" -- when those WU finished the Core21 Update auto-finished
 
We are back to full power, now to check the other Rig
 
I'm running Nvidia Driver 372.90 - by the way

Just to add to core 21  . I only have 1 GPU and it did not update after WU was done . Had to restart client . So keep a eye on it if yours has not updated yet .

This Driver 376.33 really Sucks when it comes to PG, Just and FYI
 
Going back to Driver 378.49
I will test the below later today
GeForce Hot Fix driver version 378.57Answer ID 4378
        Updated 01/31/2017 06:23 PM
    

GeForce HotFix driver version 378.57 ( A Great Driver for Folding Maxwell and on Pascal)

---

  This is GeForce Hot Fix driver version 378.57 that addresses the following:
-        Fixed crash in Minecraft and some other Java-based titles
-        Resolved 'Debug Mode' as default option on Pascal based GPUs
 
Click the appropriate link to download.
Windows 10
64-bit version
32-bit version
 
Windows 8.1/Windows 7:
64-bit version
32-bit version
 
The GTX 1080 Cards are installed an Windows 10 X79 Rig. You can view the current status on my GTX 1080 and 980 here My HFM
 Not I have no Red Limit Reach on my GTX 1080 Cards with this Hot Fix Driver. Testing the other GTX 980 cards later tonight.
Still have the Limits on my GTX 980 Cards
I have Not seen NOT having a Limit of some kind on my GTX 1080 Cards before while Folding.
BTW the bottom card is my Lemon Card, the Second one I now have. I have one 980HC Lemon Card as well.
 
Also "Resolved 'Debug Mode' as default option on Pascal based GPUs" It did not Resolve on my Pascal Cards.

I am running .18 core 21 on my two muli gpu machines but I started last night with a clean install of FAH on one machine and a fresh start with no work units on the other.  both are folding fine as best I can tell. 
 
Machine 1 is on the latest driver(378.49) and machine 2 is running the latest hot fix driver (378.49)
 
Machine 1, 980 TI, 680, 670, Windows 10
Machine 2, 670, 570, 570, Windows 10