CUDA 2.1 When? (Now!)

Any tentative date for CUDA 2.1 release? THank you

Anyone? I’m wondering this myself, as I just had to reformat/install my dev box with Vista x64, and I’d like to wait to install CUDA until 2.1 arrives to take advantage of the new/better integration with Visual Studio 2008.

Should be sooner rather than later.

I’m keeping my fingers crossed that speed issues will be fixed with the final version. My NVISION code runs at 35 gazillion flops compared to 40 with 2.0 half a year ago. I unfortunately haven’t had the time yet (don’t go to academia in Germany, you spend the entire November and December writing grant applications) to post a repro on nvdeveloper…


@ Tim,

I am running 32-bit CUDA applications with 32-bit cudart.dll over WinXP 64-bit.

I can’t run CUDA applications more than 14 to 15 times… After which cudaMalloc() fails… This is irrespective of the amount of allocation. I had filed a bug report earlier and Simon acknowledged it and said it will be fixed soon.

I really hope things will be fixed with CUDA 2.1. We chose XP 64-bit to replicate our client’s setup…Hence…

Best REgards,

Does anyone know if they’ve fixed up the VS2008 integration (build paths, templates, etc.) or is that awaiting the final release? I wanted to wait to re-install the SDK and toolkit until that was handled, but if it’s not yet, I’ll wait a little longer for the 2.1 final. I had a ton of trouble with getting the 2.0 beta working on my XP-64 box, so now that I’ve got a clean Vista-64 install, I don’t want to have to deal with that again.

2.1 beta should work with VS2008 as far as I understand. It is in the releasenotes.

Sorry, I should have been more specific. I know that CUDA 2.1 is designed to work with VS2008 (as in, it is built for that compiler), but my issue is that when I installed the 2.0 beta on my XP-64 machine, it took a huge amount of toying with build paths and settings to get everything working properly because there are some differences in the folder names on an x64 machine (verses a normal 32-bit installation of Windows). tmurray mentioned a while back that they were planning to spend some time working on this issue so that CUDA 2.1 would have much better integration with VS2008 (project templates, build settings, etc.) on both 32- and 64-bit versions of Windows.

Since I just did a fresh install of Vista-64 on my development box, I’ve lost all those settings and I don’t want to go insane changing all of that stuff around again, so I’m hoping that nVidia makes good on this and gets everything fixed up properly for the 2.1 final release (if it’s not already in the beta, which was my question).

I don’t think I ever promised this for 2.1, I said we were working on it for an indeterminate future release. AFAIK if you install the 64-bit version on 64-bit Windows it should work okay, although obviously you have to add a few library/include paths.

I don’t think I ever promised this for 2.1, I said we were working on it for an indeterminate future release. AFAIK if you install the 64-bit version on 64-bit Windows it should work okay, although obviously you have to add a few library/include paths.

Ah, sorry tmurray, I didn’t mean to put words in your mouth. I thought I had read somewhere that this was going to be a part of the 2.1 release. I guess I’ll just go ahead and install the sdk/toolkit and see if I can get it all working properly.

it’s up: [url=“”][/url]

Thanks a lot!

I just un-installed WinXP 64 and installed XP back again :(

Anyway, Would be happy to try this new one out!

THanks a Bunch!

BEst Regards,


Is the Mac OS version coming soon as well, or should we expect to wait a few months?


it is probably going to be a while.

That doesn’t sound good… Considering there’s probably very little difference between linux and OS X in terms of your CUDA API, and most of the differences would in fact be in the driver - and most of ‘those’ differences wouldn’t even relate to CUDA (just wrapping around OS X’ kernel driver api) - that would leave me to believe that most of your driver-related CUDA code is in fact ‘common’ (same code) across all platforms…

Which means… apple is holding you back from releasing CUDA 2.1 on OS X for some reason?

Either that, or CUDA is still a bit of a hack in your driver code base, thus not really running a ‘unified’/common code set across all platforms.

Of course, it could just be OS X related issues (I mean, PhysX + OS X hasn’t happened for similar reasons I’m guessing) making development tedious on that platform for such a little gain.

Believe me, CUDA is a first class citizen of the entire stack (from the user-mode driver all the way to the silicon) just like OGL and D3D. Supporting OS X is a whole lot more complicated than you probably think it is. Right now we’re not entirely happy with some technical aspects of the OS X implementation, and we want to fix those before we release another version. It is coming, just not quite yet.

Is NVIDIA also limited to Apple’s OS update schedule for releasing new drivers? The CUDA download page (whenever I have looked) has simply pointed to a minimum OS X version required for CUDA. Does this mean when the new driver is ready, it has to go through Apple QA as well and be rolled into one of their updates?

Just curious if beta drivers can ever be released directly to the community.