I want to share my experience with this round of HCC for GPU Beta because of several issues I got.

I have only two computers with GPUs modern enough to be able to crunch GPU WUs. Both are laptops and I have used them in the past for Collatz.

The first one has an Nvidia GTX 260M which is my fastest GPU, good enough for some games. But, as I stated already, it gets the “cl_khr_local_int32_base_atomics extension required by this program is not supported” error, so it’s out of the race for this project.

The second one has a more modest, but newer, Nvidia GT 330M. With this I managed to complete successfully several Beta WUs, but it hasn’t been all rosy. This laptop has a Core i7 Quad which process 8 jobs in parallel. Win7 64-bit, Nvidia drivers v257.21, BOINC v6.10.58. Left on its own, the GPU jobs starts but do nothing. The elapsed time increases, but progress remains at 0.000% complete (and GPU temperature remains low, suggesting no activity there). I managed to make them start by suspending all other tasks but six (there were seven running plus the beta), then suspending and restarting the beta job, which brings back the elapsed time to 0. Then the GPU temperature rises and at ~45 seconds the progress bar starts to increase. At this point the machine becomes very unresponsive, almost unusable. Fortunately the job finishes OK.

BTW I got a gazillion CPU WUs after receiving the betas. I aborted most of them, sorry.

Restarting task BETA_X0000038271479200409250004--201209052227020_0 using beta3 version 649 (opencl_nvidia_hcc1)

get this one. At 99% tthis one everytime is freezing. After Boinc restart it starts witch 0% normaly up to 99%

*edit*
same Problem with:
BETA_X0000038271291200409250007--201209052227033_1
BETA_X0000038271045200409250012--201209052227033_0

aborted all of my 3 Betas

System:
Processor: 4 GenuineIntel Intel(R) Core(TM) i5 CPU M 450 @ 2.40GHz [Family 6 Model 37 Stepping 5]
Processor: 256.00 KB cache
Processor features: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss htt tm pni ssse3 cx16 sse4_1 sse4_2 syscall nx lm vmx tm2 popcnt pbe
OS: Microsoft Windows 7: Home Premium x64 Edition, Service Pack 1, (06.01.7601.00)
Memory: 3.68 GB physical, 7.36 GB virtual
Disk: 452.66 GB total, 130.79 GB free
Local time is UTC +2 hours
NVIDIA GPU 0: GeForce GT 540M (driver version 301.42, CUDA version 4.20, compute capability 2.1, 1024MB, 961MB available, 258 GFLOPS peak)
08.09.2012 22:19:33 | | OpenCL: NVIDIA GPU 0: GeForce GT 540M (driver version 301.42, device version OpenCL 1.1 CUDA, 1024MB, 961MB available)

My Win-machine with an AMD/ATI HD5770 never encountered a freezing beta-GPU-WU situation. I have a non-hyperthreading-capable quad in that machine and had set them to use all four cores. When the latest beta-HCC-WU was downloaded to my machine, I saw situations where I have 4-WUs (3_CPU-based + 1 beta-HCC-GPU) or 5-WUs (4_CPU-based + 1 beta-HCC-GPU) simultaneously running. BOINC seems to have adjusted the (hardware and/or software) parameters so it can operate at either of the modes. I do not have to do anything like reserving cores and how many cores to reserve and when to do all those things. It was a definite improvement over a situation where parameters, if having to be adjusted, have to be pre-done by the cruncher before a machine set-up accommodates and thereby able to crunch a beta-GPU-WU alongside a CPU-based WU (from other WCG projects) -- all to a hassle-free conclusion.

I've read a number of posts that talks about a beta-HCC-GPU WU freezing at some point during a crunching run in the latest and also in the older beta-HCC-GPU runs. I'll hazard a guess that the for-nVidia part was not yet fully addressed* in the current beta-HCC-GPU run, and that WCG has only the AMD/ATI part covered thus-far.

Notes
*The bug that caused the exitCode 293 error and the issue of very large download of WUs for other applications -- have been found, but I'm not sure if the correction (on account of that bug) also covers the for-nVidia part.
;
; edit1_2012.09.08Sa.2222.utc
; -- typos
; edit2_2012.09.09Su.0115.utc
; -- added some necessary devil, I mean, necessary details along with the necessary re-arrangement of words to accommodate the devil, I mean, the details...
;

I have been taking a look at my results for the GPU beta since this is my first foray into GPU processing. I am usings an HD 5970 Black Edition and have allowed both GPUs to run the test.

Using a basic 'out of the box' setup, I seemed to always have six tasks waiting to run (and when a new wave became available, I always seemed to grab six at a time).

What I find strange is that if the first of the six errored out, thay would all do the same. Similarly, if the first passed, they would all pass and seem to have all been validated.

One other oddity I have noticed is that some wingmen would get the same errors whilst others would pass. In only a few cases did I note that everyone got an invalid result.

I would really like to know if al those who got the same result were using IDENTICAL GPUs. I guess that is something that the techs will be considering

It seems pointless to list all my results because there are too many and the techs will get the info anyway. I am sure that eventually all the data will be aggregated and help to produce a definitive list of 'good' and 'bad' cards and even better. GPU beta testing is far more than testing WUs and for this reason, I am enjoying being able to take part in what is truly developing as we participate.

I don't think the forum paradigm does the job in keeping people informed about the beta tests. I think we need a semi-static page where the latest announcements and status are posted. Sure, use the forum for trouble reports etc. but one page to get the latest skinny on the current betas. As it is now, I bounce around all over the place trying to figure out what is going on.

This is not intended as criticism but rather as a suggestion to make the beta testers job easier. Just a suggestion. Regards.