CUDA 2.0 Release information?

It seems kind of strange that CUDA 2.0 appeared without any announcements here in the forum. I’d like to switch to the official version 2, but I’d like to know whether or not there are going to be any more changes in the short term before I put in the effort.

I have to say that I’m pretty disappointed with how the release with version 2 has been going so far. Beta 2 was never even announced in the announcements forum. Bug reports were generally greeted with a cryptic response of “This is a confirmed bug and is fixed in CVS” and no indication whatsoever of when these bug fixes might be made available.

Agreed, their communication on this forum is a hit or miss… the latter more often than the former. But hey, after all CUDA is free for everybody, so I won’t be complaining.

High profile commercial customers (i.e. not myself) probably have access to better support channels anyway.


Unfortunately, I have to second your opinion. Very unfriendly and rare responses on bug reports, no press-releases and information on the newer version, and, the worst thing, old bugs in the new version.

I have already reported it in the ‘programming’ branch without an answer: windows watchdog problem for the secondary GPUs is still there regardless the fact that Nvidia promised to fix it in 2.0.

In my case, this problem is crucial: I can’t migrate to Linux rapidly and my kernels are time-consuming. Actually, it is hard to understand why things go as they go when nvidia counts on non-typical usage of their GPUs with so mediocre CUDA-related development.

What is currently posted is the final 2.0 release.


Can you say a single word on the problem of watchdog, please ?

One of nvidia guys (tmurray) confirmed this bug (Confirmation), why it is not fixed and what windows user should do about it ?

I’ve been waiting for final 2.0 only because of watchdog problem and this issue wastes the whole work in some cases.

I reopened the bug yesterday, it was supposed to be fixed in the final release but it slipped through.
We are working on a solution.

Thanks a lot!

Are there any approximate estimations on the time the fix will require ? Days, weeks, months ?

Hmm, the Mandelbrot Sample still randomly flakes out for me with the Cuda 2.0 release toolkit and SDK. I am going to add some CUDA_SAFE_CALL macros to see if any calls start throwing errors.


What card are you using?

Other thing you should consider is to release the CUDA2.0 libraries for all the C CRT models ( /MD, /MT, /MTD, etc ). Currently is almost impossible to link a CUDA program using MC++/CLR and due to linking conflicts… and to fusion x86 and x64 into the same installer to allow cross-compiling easier.

8800GT on Windows Vista 64 Bit. I installed the 32 bit toolkit and SDK because my Visual C++ Express edition can only generate 32 bit binaries.

the symptom is the graphics stop updating and the FPS indication in the title bar goes nuts, randomly after several seconds (or minutes) of Mandelbrot zooming/panning. I shall try to update from 177.79 driver to 177.84 to see if that improves anything.