Folding Forum

That''s why when they began this whole thing, I was assuming that they would disperse a fully regression tested release at regular intervals between the GRs, but alas, this does not seem to be.

SombraGuerrero wrote:That''s why when they began this whole thing, I was assuming that they would disperse a fully regression tested release at regular intervals between the GRs, but alas, this does not seem to be.

I think to do that adequately requires massively more resources than even a company such as NVIDIA could really afford.

That could very well be. That's why I feel that unfortunate as it is, this "Game Ready" path is one from which there is no return.

Regression testing shouldn't require a lot of programmer-hours. It should be an automated script that just keeps getting longer and longer (in terms of processing time).

I suspect the problem here is that Core_21 found a new bug.

OpenCL is an API that's supposed to work as long as the external code doesn't violate the concepts of OpenCL. Validating OpenCL isn't as simple as calling each API entry once. External programs can use those calls in much more complicated ways and might uncover a particular sequence that was not part of the basic test.

Maybe nvidia could use FahBench to reproduce the error and then add FahBench to regression tests for the future.

FAHBench needs to be upgraded so support the current OpenMM but once that's done, it's a great idea.

It is updated for current Core 21 OpenMM with FahBench version 2.2.5 now
https://fahbench.github.io/
If configured with a long runtime and running a real work unit then this is a good driver stability test.

still no fix other then rolling back the driver?

fn_duud wrote:still no fix other then rolling back the driver?

Yes, that is still the only fix available at this time. The latest version driver reported to work is 373.06. Unfortunately that leaves buyers of some of nVidia's most recent releases of video cards unable to run folding.

fn_duud wrote:still no fix other then rolling back the driver?

Or editing the driver files of the older driver to work on the newer models.

Joe, 373.06 should work without issue on all recent Nvidia GPUs. For me, it works perfectly with series 1080, 1070, and 980 GPUs. Perhaps you were referring to limitations with games?

Leonardo, the recently released 1050, 1050ti and possibly other recent ones are not supported by nVidia with the older drivers. For example, the oldest driver listed for a 1050ti was 375.63 when I checked their downloads site and listed all available drivers.

Not sure it'll do much good since it seems no one's been reading the Dev forums anyway, but I did post on my thread to Nvidia the idea of using fahbench for defect replication.

Leonardo wrote:Joe, 373.06 should work without issue on all recent Nvidia GPUs. For me, it works perfectly with series 1080, 1070, and 980 GPUs. Perhaps you were referring to limitations with games?

As Joe_H says, my GTX 1050 Ti is an expensive door-stop at the moment as it won't work with 373.06.
If someone can point me to instructions on how to hack the relevant files then I might give it a try.

I certainly stand corrected. Sorry about this issue, snapshot.