Linux support for next CUDA release

Accelerated Computing CUDA CUDA Programming and Performance
1

In preparing for the next release of CUDA (after v1.1) , here’s the current plan for Linux distributions compared to those supported on CUDA v1.1:

Add support for:

  • Ubuntu 7.10
  • RHEL-4.6
  • RHEL-5.1
  • SUSE-10.3
  • Fedora 8

Drop support for:

  • RHEL-3.x
  • SUSE-10.1 (replaced with SUSE-10.3)
  • Fedora 7 (replaced with Fedora 8)

(continued support for: RHEL-4.3, RHEL-4.4, RHEL-4.5, RHEL-5(.0), SUSE-10.2, SLED-10(.0), Ubuntu-7.04 and SLED-10-SP1)

Schedule: beta ~late March.

And even though this is posted under the Linux heading: yes, the next release will support Vista 32 and 64-bit.

2

sounds good :)

3

Ahh, that will make my life on my vista laptop much easier. And the support for FC8 is also very welcome :)
Let march come ;)

4

Hadn’t noticed anyone shouting about RHEL3 support going away… just wanted to let folks know one more time in case this was missed over the holiday break

5

The linux 64 bit driver for cuda 1.1 is listed as being here:

[url=“Page Not Found | NVIDIA”]Page Not Found | NVIDIA

on this page:

[url=“http://www.nvidia.com/object/cuda_get.html”]http://www.nvidia.com/object/cuda_get.html[/url]

but the linux_display_amd64_169.07.html is broken. is there another link?

6

sorry about that. Looks like maintenance on the main site.
You should be able to go directly here for now

[url=“http://us.download.nvidia.com/XFree86/Linux-x86_64/169.07/NVIDIA-Linux-x86_64-169.07-pkg2.run”]http://us.download.nvidia.com/XFree86/Linu...169.07-pkg2.run[/url]

7

Sorry for asking this quite late, but I notice that none of the current set of supported distros use gcc 4.2 or better. Is there any plan for adding support for gcc 4.2+? I ask because I’m hoping to do some work with CUDA and OpenMP and OpenMP is really only supported after gcc 4.2.0.

Thanks.

8

I’ve been mixing the standard gcc 3.4 on RHEL4 with gcc 4.2 (RPMS with Barcelona patches from AMD) to use OpenMP and CUDA in the same program. I set my path to include the gcc 4.2 binary directory, which is in a non-standard location. Most of my code is then compiled with gcc 4.2. Then I pass -ccbin=/usr/bin to nvcc, which forces nvcc to use the system default gcc 3.4.

This solution has worked for me since I use OpenMP and CUDA in completely different places in my program. I have contemplated using OpenMP to help me manage threads for the separate contexts in a multi-GPU program, but I have not tried it yet.

9

Are we still on for the late March Beta?

10

getting kind of late in March… any word from the powers that be?

11

yupp looking very close now.
Note that beta release will come out for 32-bit and 64-bit for each of Win/Vista/RHEL4&RHEL5
It’s after beta is done that we fully support all the listed linux distros

12

GOD BLESS NVIDIA!!

13

Hi,

Any information on the release date of this beta version? We are waiting for

the new CUDA release for OpenSuSE 10.3 (gcc-4.2.1).

Jyh-Shyong

14

It’s been over 2 months since the Beta release. Is the final release with openSuse 10.3 support coming soon?

15

we’ve been aiming for about 2 months. Beta was posted on April 17th.

16

I’ve heard that fortran will be supported in this year.

the final version of next cuda will support Intel C/C++ compiler?

17

I’m working on Fedora 9, gcc 4.3
Is this due to be supported soon?

18

PLEASE add support for openSUSE 11.0

19

Ubuntu 8.04 is really, really nice to work on. Everything works, and some small stuff is significantly nicer in 8.04 than 7.10.

For instance, to name a small detail, GDB is better for CUDA debugging in 8.04 because you can turn off the thread start / stop messages.

7.10 was nice, but lacking in some areas. 8.04 is perfect, basically.

20

+1 :crying: External Media