Should be quick, given the fact that the releasenotes mention 64bit linux support in a previous version. So I would guess as soon as NVIDIA is sending 64bit conformance drivers :)
In addition, you might be able to work around that by using -m32.
Oh thank you for releasing the OpenCL SDK to all registered develepers, I think.
I also have another request, or suggestion, I hope it is not too intrusive.
With CUDA nvidia for some reason made the 64 and 32 bit SDKs mutually exclusive and never listened to the complaints of some of us wich had problems with this.
This is a very difficult problem for those who use 64 bit system for development but also distribute 32 and 64 bit SDKs.
There are some of us that do distribute small SDKs to a small group of people.
for us realizing one solution for 64 bit and not for 32 or vice versa create a great deal of complain and forum wars.
I refrained from releasing CUDA support for my SDK for that reason.
Please this time around made 32 and 64 bit opencl SDK coexist on the same system.
I just checked the Developer site. Really now, really, who on this planet still uses 32-bit XP. I’m assuming that CUDA people would be most interested in OpenCl, and from what I can see in the forums, the most common configurations are 64-bit Linux, and 64-Bit Server 2008. It’s nice to light your cigar with a $100 bill that was previoulsy lit by a $20, but when you’re not rich, 32-bit XP doesn’t make much sense either…
hi, tmurray
i have registered and filled the form, several days elapsed, but also no response.
so i want to know how long do you need to check my informations? :mellow:
I have registered several days ago, how long does it normally take to get access.
Thanks,
One and a half year ago it took more than a month, but I believe it is shorter nowadays.
It only took me 6 hours o get accepted, back when CUDA 2.2 beta was the hot stuff.
Alex.
I have registered for a long time, but until now, have no response.
does anyone can help me to download the OpenCL driver on windows and share to me?
i will be appreciate!
I believe registered developers re not allowed to redistribute the OpenCL Candidate, according to the license agreement. Here’s something to try: if you have a CUDA project, set up a website describing it; that might get you in faster.
did you receive e-mail from NV after filling the registration form? I didn’t.
The question is: has our form arrived???
:ph34r:
Here we are working our butts off trying to improve mpp on hybrid, heterogeneous compute clusters to help scientists and engineers. The cuBLAS library is inefficient, the cuFFT library is WAY inefficient, and now OpenCL (which doesn’t seem to improve the efficiency or effectiveness, anywhere) is being dribbled out developers. I don’t know, it plain doesn’t make sense.
We have put ALL consideration of OpenCL on hold for the forseeable future, and are only developing technologies that allow highly effective, efficient CPU / GPU compute fabrics. The old mainframe, traditional cluster approach is going the way of GM - and those who don’t understand the use of their own products may be left in the dust.
When NVidia wants to know about these new technologies, they can register on our developer’s site, and we will release it to them eventually.
The problems that you might have with imoplementing GPGPU clusters is most likely not nVidia’s fault, I am sorry to say. I have gottenA lot of people, with a bit of patience, and help from other forum members have obtained excellent results.
The PITA with CUDA is writing the software, but anyone with a bit of imagination can come up with fantastic ideas.
Do not take this as a personal attack, but I see this lack of imagination on GM’s part as one of the reasons GM’s R&D is a piece of crap nowadays (again, take no offense).
64-bit conformance candidates are now available.
Mr. Nuke,
I take no offense. And I certainly don’t fault NVidia for the CUDA platform. I have been working with CUDA since beta .8, and have invested mightily in learning what it takes to develop efficient parallel programs and data for that platform. An example of what we can do is being presented at various venues this summer, using our demonstration hybrid, heterogeneous cluster (a little screamer based generally on Rocks/CUDA). So, I’m not complaining AT ALL.
My rub is dribbling out OpenCL (which I don’t see as bringing much to the table) only to “select” developers. Bad form.
Prerelease means just that. And it’s not like it’s that hard to become a registered developer at this point anyway–it’s probably worthwhile so you’re not stuck using some old OpenCL conformance candidate when we release full official drivers or even incremental improvements down the line.
What I see OpenCL doing is bring a little more compatibility with a little less power. It’s nice that, theoretically, the same program should run on ATI cards as well, maybe even Intel, but as long as nVidia cards stay cheap, I see no reason in making the effort to port to OpenCL. And it’s still in its beta stage, so I concede that the potential headaches are not worth it… maybe later down the line. Right now CUDA has all the goodies.
Alex
just look at the assembly code of your opencl kernels. there’s still a lot of improvement which can be done with further development of the OpenCL compiler. because this is just a prerelease i assume that nvidia is going to increase the performance of opencl to be competitive to CUDA. afaik, the first releases of CUDA had similar performance issues.
Its been forever! Any idea when NVIDIA will release OpenCL publicly? Theres been no response to my “application” which was irritating in itself. Why does NVIDIA make it such a royal pain for early adopters? You guys suck at software.
I’ll try to check on this for you. What email address did you use on the registered developer application form?