CUDA 3.1 beta

Accelerated Computing CUDA CUDA Programming and Performance
1

The e-mail announcing cuda 3.1 on the developer site just arrived. Looks to me as if the Suse 32 and 64 bit packages are reversed…

2

I’ll ping someone.

3

Are the sdk examples yet to be uploaded or are they included with the toolkit, or…?

4

Are they not available? (I don’t even have an account to check these things, I just write parts of the driver…)

5

Unfortunately, no. Only cudatoolkit is available, but no cuda/gpucomputing sdk.

Also, cudatoolkit installation binary complains about driver version (less than 256), but only Tesla drivers are available for download from nvdeveloper site.

Will check again within several days.

6

Any updates with the SDK?
We are holding off the installation until the sdk is available

7

SDK is avaliable from nvdeveloper

8

I’ve recently bought a GTX 470 for my development box at home and would love to get some CUDA development done over the weekend. We’re going to start the beta testing phase for GPU acceleration in Indigo Renderer soon, so it would be great if I could do some testing and optimisation on Fermi before the 24th (planned release date for new drivers).

I’ve looked on nvdeveloper, the only drivers available are for Tesla cards External Image

Thanks for your time,

Thomas

9

I’ve recently bought a GTX 470 for my development box at home and would love to get some CUDA development done over the weekend. We’re going to start the beta testing phase for GPU acceleration in Indigo Renderer soon, so it would be great if I could do some testing and optimisation on Fermi before the 24th (planned release date for new drivers).

I’ve looked on nvdeveloper, the only drivers available are for Tesla cards External Image

Thanks for your time,

Thomas

10

Use CUDA 3.0, it has Fermi support.

11

Use CUDA 3.0, it has Fermi support.

12

Mister Anderson…

Thanks for the tip, I had been using CUDA 3.0 but assumed it wasn’t compatible with Fermi since my previously working code (with GTX 260) suddenly stopped working. However, I’ve found that compiling with -arch=sm_20 set in nvcc allowed the cuModuleLoad function to successfully complete.

Unfortunately, now cuModuleGetGlobal is failing… and so the quest for Fermi support continues… I’m going to look around a bit more on these forums in search of solutions. Thanks again!

13

Mister Anderson…

Thanks for the tip, I had been using CUDA 3.0 but assumed it wasn’t compatible with Fermi since my previously working code (with GTX 260) suddenly stopped working. However, I’ve found that compiling with -arch=sm_20 set in nvcc allowed the cuModuleLoad function to successfully complete.

Unfortunately, now cuModuleGetGlobal is failing… and so the quest for Fermi support continues… I’m going to look around a bit more on these forums in search of solutions. Thanks again!

14

I can’t help any further this this one. I only ever use the runtime API. The fermi compatibility guide might have something: You might be interested in the Fermi compatibility guide: http://developer.download.nvidia.com/compu…bilityGuide.pdf .

15

I can’t help any further this this one. I only ever use the runtime API. The fermi compatibility guide might have something: You might be interested in the Fermi compatibility guide: http://developer.download.nvidia.com/compu…bilityGuide.pdf .

16

That looks to be exactly what I need, thanks!

17

That looks to be exactly what I need, thanks!

18

The cuda reference manual is also missing in 3.1 beta