New 375.57 GeForce driver is bad for folding

[GEHORC]

Short story:

Don't upgrade your Nvidia driver to the 375.57 version.

 

Long story:

I just discovered this morning, that my GPU wasn't folding properly.

F@H was reporting every single WU as a bad one.

After a quick lookup on the googles i found out that i wasn't the only one.

http://forums.evga.com/Anyone-keep-getting-bad-work-units-after-updating-to-lastest-37557-drivers-m2567207.aspx

After going back to the 373.06 version everything was fine again.

If you did not upgrade yet - don't do it.

[Imakuni]

4 hours ago, GEHORC said:

Short story:

Don't upgrade your Nvidia driver to the 375.57 version.

 

Long story:

I just discovered this morning, that my GPU wasn't folding properly.

F@H was reporting every single WU as a bad one.

After a quick lookup on the googles i found out that i wasn't the only one.

http://forums.evga.com/Anyone-keep-getting-bad-work-units-after-updating-to-lastest-37557-drivers-m2567207.aspx

After going back to the 373.06 version everything was fine again.

If you did not upgrade yet - don't do it.

Can you tell us what your GPU is, please?

 

This may / may not bother people on the BOINC side as well (anyone remembers the 364.xx OpenCL fiasco?), and if so, it will require action. I'll test this myself as well; in the meantime, all the info we can gather will be useful, if not for the sake of at least bug reporting it.

 

EDIT: for you F@H folks out there, how / where can I check for Invalid WUs?

 

EDIT 2: I gave a quick try at 2 PrimeGrid OpenCL apps. And at first glance, nothing's wrong with the new drivers. Will begin testing on F@H.

 

EDIT 3: Well, it seems like the new driver indeed has problems with folding. I tried it with my Gtx 970 for a little bit and every WU would fail, whereas before it would crunch no problem. Back to 372.90 for me, then...

[SnowyMus]

Yep! I've gotten the same issue as of updating to the latest driver. I've only managed to successfully fold one WU correctly last night with the GPU before getting the below log message after downloading the second one. The WU after that completely failed.

 

The GPU, an MSI GeForce GTX 1070 Gaming 8G, is not overclocked aside from the factory overclock, and it has always worked 100% of the time before updating drivers where it's not working at all now, so I can only conclude that it's the driver.

 

05:20:39:WU01:FS01:0x21:*********************** Log Started 2016-10-22T05:20:39Z ***********************
05:20:39:WU01:FS01:0x21:Project: 11707 (Run 15, Clone 28, Gen 49)
05:20:39:WU01:FS01:0x21:Unit: 0x000000438ca304f357a491256f8182e2
05:20:39:WU01:FS01:0x21:CPU: 0x00000000000000000000000000000000
05:20:39:WU01:FS01:0x21:Machine: 1
05:20:39:WU01:FS01:0x21:Reading tar file core.xml
05:20:39:WU01:FS01:0x21:Reading tar file system.xml
05:20:39:WU01:FS01:0x21:Reading tar file integrator.xml
05:20:39:WU01:FS01:0x21:Reading tar file state.xml
05:20:39:WU01:FS01:0x21:Digital signatures verified
05:20:39:WU01:FS01:0x21:Folding@home GPU Core21 Folding@home Core
05:20:39:WU01:FS01:0x21:Version 0.0.17
05:20:42:WU01:FS01:0x21:ERROR:exception: Error downloading array interactionCount: clEnqueueReadBuffer (-5)
05:20:42:WU01:FS01:0x21:Saving result file logfile_01.txt
05:20:42:WU01:FS01:0x21:Saving result file log.txt
05:20:42:WU01:FS01:0x21:Folding@home Core Shutdown: BAD_WORK_UNIT
05:20:45:WARNING:WU01:FS01:FahCore returned: BAD_WORK_UNIT (114 = 0x72)
05:20:45:WU01:FS01:Sending unit results: id:01 state:SEND error:FAULTY project:11707 run:15 clone:28 gen:49 core:0x21 unit:0x000000438ca304f357a491256f8182e2
05:20:45:WU01:FS01:Uploading 2.41KiB to 140.163.4.243
05:20:45:WU01:FS01:Connecting to 140.163.4.243:8080
05:20:45:WU01:FS01:Upload complete
05:20:46:WU01:FS01:Server responded WORK_ACK (400)
05:20:46:WU01:FS01:Cleaning up

 

373.06 works, though.

[GEHORC]

2 hours ago, Imakuni said:

Can you tell us what your GPU is, please?

Like you I use a GTX 970.

2 hours ago, Imakuni said:

EDIT: for you F@H folks out there, how / where can I check for Invalid WUs?

Not sure what you mean by this. The logs will tell you.

[Imakuni]

2 minutes ago, GEHORC said:

Not sure what you mean by this. The logs will tell you.

I thought the error wasn't as bad. You see, I thought it was something along the lines of the WU finishing and outputting a result, but that result being incorrect. Logs will tell you about the unit failing immediately, but if it was to crunch to end and then fail validation when checked against another user, how would you determine that such thing ever happened?

[GEHORC]

2 minutes ago, Imakuni said:

I thought the error wasn't as bad. You see, I thought it was something along the lines of the WU finishing and outputting a result, but that result being incorrect. Logs will tell you about the unit failing immediately, but if it was to crunch to end and then fail validation when checked against another user, how would you determine that such thing ever happened?

Ah ok. No it was crashing right at the beginning and instantly requesting another WU from the server.

[Ithanul]

You not the only ones.

 

Several others on different forum board having the same issues.  Even the latest one 375.63 is no good.

 

Whole reason I rock old drivers.

[Imakuni]

12 minutes ago, Ithanul said:

Whole reason I rock old drivers.

I follow a different logic: keep as up to date as possible so that if a problem ever shows up (this 375 F@H bug and the 364 OpenCL mess come to mind) it can be detected as soon as you can.

[Ithanul]

4 hours ago, Imakuni said:

I follow a different logic: keep as up to date as possible so that if a problem ever shows up (this 375 F@H bug and the 364 OpenCL mess come to mind) it can be detected as soon as you can.

I understand that as well.  Since no other way of the rest to know if it alright to update to the latest driver.

 

Reason I tend to sit put is I don't like dealing with a unstable system or issues since I tend to run my folder and main rig 24/7 non stop.  I pretty much don't update drivers unless I'm doing a clean install.  Which I'm a bit over due, tend to do that about once a year.