GPU folding fails -- clEnqueueReadBuffer

[toTOW]

Be careful, because in the history of the failure of my 980 Ti, I first started to get random Bad States. Then I started to see some GPU resets with clEnqueueReadBuffer (-5) error. Then, I started to find my system being turned off automatically. And one day, after 9 months of operations, after powering it back on, the card blew up (VRM burned) when Windows activated it. The short circuit in the VRMs prevented the machine from even powering up.

Thanks for your answer. I take this problem really serioiusly.
Are there any tests/SW/tools you can recommend that prove that something is wrong with my GPU?

Any 3D intensive application will do the trick : choose the game of your choice or a benchmark like 3DMark or FurMark ...

[foldinghomealone]

toTow, as I mentioned in my previous post it is stable in daily life. Benchmarks don't crash my system.
But I lose WUs and you and Bruce argue (probably correctly) that this proves that something wrong with my system.

However, when I return my GPU within warranty period, the manufacturer will decline replacement, when I argue like you.

For that I need proof. Like a diagnostic tool that shows a specific problem with the gpu.

Thanks for your help.

[JohnChodera]

RABishop: I note that you have GTX-1080s and are using the FIRST driver introduced that supports GTX-1080s:

> looked, and my driver for all is the newest I know of or can find, which is the 367.27.

With OpenMM, we've had some issues with this early driver (released 2016.6.13) causing NaNs on GTX-1080s. The latest linux driver is 375.26 (released 2016.12.14):

http://www.nvidia.com/download/driverRe ... 2992/en-us

Any chance you could try this recent driver?

[Leonardo]

RABiship, I would install Nvidia driver 373.06. As of late, it seems to the best for stability, and generally enables very good performance. I'm using 373.06 with 1080s and a 1070.

[RABishop]

I'm having a problem similar to this. I'm trying to get a GTX 1080 FTW card to function in a new configured machine. Project # THIS time is 8207. Work Server #: 134.139.52.2. Collection Server #: 134.139.52.3. I have tried many times uninstalling, reinstalling, purging everything. Every time I try to get the first GPU to run, it tries, then fails, over and over again. I have the newest available NVIDIA Driver installed: 384.69, amd 64.

06:04:20:WARNING:WU01:FS01:FahCore returned: BAD_WORK_UNIT (114 = 0x72)
06:05:12:WARNING:WU02:FS01:FahCore returned: BAD_WORK_UNIT (114 = 0x72)
06:05:15:WARNING:WU01:FS01:FahCore returned: BAD_WORK_UNIT (114 = 0x72)
06:05:22:WARNING:WU02:FS01:FahCore returned: BAD_WORK_UNIT (114 = 0x72)
06:05:29:WARNING:WU01:FS01:FahCore returned: BAD_WORK_UNIT (114 = 0x72)
06:05:37:WARNING:WU02:FS01:FahCore returned: BAD_WORK_UNIT (114 = 0x72)
06:05:43:WARNING:WU01:FS01:FahCore returned: BAD_WORK_UNIT (114 = 0x72)
06:05:45:WARNING:WU02:FS01:FahCore returned: BAD_WORK_UNIT (114 = 0x72)
06:05:48:WARNING:WU01:FS01:FahCore returned: BAD_WORK_UNIT (114 = 0x72)
06:05:54:WARNING:WU02:FS01:FahCore returned: BAD_WORK_UNIT (114 = 0x72)


I am becoming a bit annoyed with this, to say the least.

[bruce]

ClEnqueueReadBuffer (N) errors are being reported by OpenCL. As a general rule, they're closely related to driver errors or resulting from flaky hardware or other instabilities. ( I really don't have a list of the varous values of N to any particular condition. As with all instabilities, you may or may not be able to fix it by simple underclocking or improving the cooling.

[foldy]

Please post your logfile.

[toTOW]

-5 is the code for CL_OUT_OF_RESOURCES error, which is logical when it happens while the GPU is reset by the driver, because when it's being reset, there no resources available ...