-65% PPD on projects 11728 & 11730 ???

[ifolder]

Hi, I notice a severe drop of the PPD on my 1080Ti these last two days. According to FAHClient the estimated PPD on the current WUs belonging to projects 11728 & 11730 is around 500k while it used to be around 1.3M the days before when WUs from these project didn't come yet. Maybe the base credit of these projects is heavily underestimated?

[foldy]

I have a gtx 1080ti too on Windows 7 and got a 11728 with 1.2M PPD (TPF was 0:57 min) on 2019-03-07 but no 11730 yet. Driver was nvidia 391.35 now I run nvidia 425.31 but it makes no difference except for games.

[ifolder]

I am running 7.4.4 under Linux and 418.56 driver.

In the logs I see that:

06:06:39:WU01:FS01:0x21:Project: 11730 (Run 4, Clone 428, Gen 159)
06:06:39:WU01:FS01:0x21:Unit: 0x000000bd8ca304e75bcd13d70a03b52a
06:06:39:WU01:FS01:0x21:CPU: 0x00000000000000000000000000000000
06:06:39:WU01:FS01:0x21:Machine: 1
06:06:39:WU01:FS01:0x21:Reading tar file core.xml
06:06:39:WU01:FS01:0x21:Reading tar file integrator.xml
06:06:39:WU01:FS01:0x21:Reading tar file state.xml
06:06:39:WU01:FS01:0x21:Reading tar file system.xml
06:06:39:WU01:FS01:0x21:Digital signatures verified
06:06:39:WU01:FS01:0x21:Folding@home GPU Core21 Folding@home Core
06:06:39:WU01:FS01:0x21:Version 0.0.18
06:06:47:WU01:FS01:0x21:ERROR:exception: Error uploading array posq: clEnqueueWriteBuffer (-4)
06:06:47:WU01:FS01:0x21:Saving result file logfile_01.txt
06:06:47:WU01:FS01:0x21:Saving result file log.txt
06:06:47:WU01:FS01:0x21:Folding@home Core Shutdown: BAD_WORK_UNIT

And nvidia-smi returns: Unable to determine the device handle for GPU 0000:04:00.0: GPU is lost. Reboot the system to recover this GPU

So looks like FAHClient or the driver or the GPU crashes but FAHClient reports it is still folding with a very low estimated PPD.

[bruce]

FAHClient does sometimes report "very slow" estimated progress when it's not actually happening. There are occasional reports of a WU stopping but the estimated percentage completion continues to increase until it reaches 99% -- or until the next checkpoint is written.

clEnqueueWriteBuffer errors indicate that the GPU could not allocate the necessary VRAM to perform the requested operation. Perhaps this is because the WU's context is no longer available (WU crashed?) or because the GPU has been reset due to some other error or maybe even because some other GPU process has filled up free VRAM. Somewhere there's a table explaining the (-4) but I don't know where to find it. If your GPU is overclocked, you might try underclocking it.

[foldy]

-4 = CL_MEM_OBJECT_ALLOCATION_FAILURE
if there is a failure to allocate memory for buffer object.
https://docdro.id/Ee5RTkS