There are 2 tasks that would require volunteer to do:

1) To determine NV drivers range that suport SIGNALS_ON_GPU OpenCL MB build.
2) To determine NV drivers range that support correct operation of usual OpenCL MB build.

As one can see both tasks require drivers reinstallation (cause different drivers versions needs trial).

If you able and interesting in doing such work please PM me. I'll send links to special binaries appropriate for this test.

Big thank you to all who responded on my call!
I sent you link to test case and instructions how to deal with it.
Will post link to compatibility table when first results will arrive.

Thanks to Mr. Kevvy we have first cells in NV drivers compatibility table for OpenCL app:
https://docs.google.com/spreadsheets/d/1gegAsoKPGiUnwidZUx7UFb9SaKXv-uNr-grB8QmVC4k/edit?usp=sharing

Thanks all who sent me data. Here is updated table: https://docs.google.com/spreadsheets/d/1gegAsoKPGiUnwidZUx7UFb9SaKXv-uNr-grB8QmVC4k/edit?usp=sharing
Not all recived data processed and included in this table still so if you don't see your results included please wait a little, they are pending for inclusion.

And some good news - it seems 347.09 and up are compatible so current 350+ limitation could be extended.

I don't know what "SoG/nonSoG" means, so it is hard to report results.

A note on the table would be very nice :)

My 750Ti seems to be working well, but I don't know what to tell you without a little more info.

I don't know what "SoG/nonSoG" means, so it is hard to report results.

A note on the table would be very nice :)

My 750Ti seems to be working well, but I don't know what to tell you without a little more info.

Please re-read opening post. Do you ready to change driver versions on your host and run offline testcase ?

Green means VALID, test passed.
Red means INVALID, test failed.
SoG means SIGNALS_ON_GPU build path
nonSoG means "usual" path used before SoG introduction.
For casual reader color marking is the most important part of that table.

If you have some documentation, I am willing to to try offline testing, but I don't know how.

If you have some documentation, I am willing to to try offline testing, but I don't know how.

PMed instructions.

Do you want these run without any other tasks running?

EDIT: I see now that it shuts BOINC down to run the test.

Thanks to EdwardPF and Mr. Kevvy there is big collection of nVidia drivers available:

http://filehippo.com/software/drivers/graphics/

and

https://cloud.mail.ru/public/JBX5/AkE3LxLAH (in upload process corrently)

There was some difference in behavior of different GPU models under 347.09 drivers.
So please check thoroughly drivers range from 346.00 to 348.99 - any driver revision you could find from this range.

Thanks

Table grown : https://docs.google.com/spreadsheets/d/1gegAsoKPGiUnwidZUx7UFb9SaKXv-uNr-grB8QmVC4k/edit?usp=sharing

As one can see only GTX 460 results differ from others.
So, if you have GTX 460 GPU and able to make tests with different drivers please PM me.

I would like to thank all who participated in this table creation, namely:
Cliff Harding, EdwardPF, Mark Lybeck, Michael W.F. Miles, Mr. Kevvy, Richard Haselgrove, Terror Australis, Urs Echternacht

and all who answered on this call but wasn't able to provide data so far.

Thanks!

I tried, but really didn't understand the instructions for running offline.

I tried once, and you said it was a bad return. I was busy so I just left it.

Oh well, I still willing to help next time.

I tried, but really didn't understand the instructions for running offline.

I tried once, and you said it was a bad return. I was busy so I just left it.

Oh well, I still willing to help next time.

Your text in letter showed that run was done and done right. It's just issue with archiver usage perhaps. Or log file was deleted before it was compressed into resulting archive. And yes, development continues so some opportunities for testing are unavoidable :)

Table grown : https://docs.google.com/spreadsheets/d/1gegAsoKPGiUnwidZUx7UFb9SaKXv-uNr-grB8QmVC4k/edit?usp=sharing

As one can see only GTX 460 results differ from others.

Just a thought - are you recording operating system in use? Not every variant, of course, but comparing XP with "Vista and everything later" might be worth a look, particularly for that GTX 460.

I might be able to help check out some others next week, now that the installer is finished, but unfortunately my GTX 470 (most likely to be of interest, from that chart) is in my main business workstation, which I want to disrupt as little as possible.

AFAICT that GTX 460 is Win7 x64 one.
But paired with ATi GPU. Maybe it's the reason, investigation in progress.

Initial observation from an extended test (in progress):

There was a PhysX version jump from 9.14.xxxx to 9.15.yyyy between 347.88 and 350.12

PhysX had remain unchanged at 9.14.0702 since 344.75

I'll dig out more minor version numbers (perhaps for intermediate releases) when I've finished the test.

I thought PhysX had nothing to do with crunching.

Completed my GTX 970 tests late last night - nothing really new to report. SoG didn't work with 345.20, started working at 347.88 (intermediate values not tested)

I did manage to get these timings for a representative set of FG_v8 test WUs and the cuda50 app:

Start		Finish			Duration		Driver		PhysX
11:37:31	12:30:06		00:52:35		344.75 clean	9.14.0702
13:00:21	13:53:16		00:52:56		345.20 clean	9.14.0702
14:04:55	14:59:18		00:54:23		347.88 clean	9.14.0702
15:10:23	16:03:17		00:52:54		350.12 clean	9.15.0324 new
16:10:14	17:03:47		00:53:33		352.86 clean	9.15.0428 new
17:23:56	18:18:42		00:54:46		353.62 clean	9.15.0428
18:24:06	19:18:43		00:54:37		355.98 clean	9.15.0428
19:24:06	20:18:24		00:54:18		358.91 clean	9.15.0428
09:56:22	10:51:12		00:54:50		359.00 OEM	
20:23:40	21:17:49		00:54:09		359.06 clean	9.15.0428
21:23:26	22:21:39		00:58:13		361.43 clean	9.15.0428
22:33:19	23:30:35		00:57:16		361.60 clean	9.15.0428

They definitely seem to be getting slower.

And deviance with one of 460 GPUs was caused by incompatibility between NV and ATi cards. After ATi card removal results are OK.

Note, that with more recent drivers (than 347.xx) even this incompatibility disappeared.

There will be new so called "SoG' OpenCl NV build for testing on SETI beta. Please participate in testing there.