64bit linux drivers?

Does nVidia have a plan for releasing an amd64 version of the linux CUDA drivers?

Yes, 64 bit OS support is planned for a future release.

Thanks,
Lonni

Is there a time line or estimation? The next few months? Sometime this year?

Thanks.

It’s safe to say sometime this year, but I can’t give more detail than that now. Adding a new OS requires a large amount of testing before we can release.

Mark

Mark,

Just FYI, I don’t know if the holdup is more related to testing the driver, or testing the compiler/SDK portions, but having the 32-bit compiled CUDA apps be able to run on a 64-bit host machine (running a 64-bit kernel and a 64-bit NVIDIA driver) would be a huge improvement in convenience in terms of deploying CUDA-enhanced software to real users. In our labs for example, every single machine is running a 64-bit kernel as they are used for a lot of calculations that require many GB of memory. Even on such a machine though, it would be very nice to be able to run a 32-bit CUDA app, until such time that the cuda compilers/SDK/runtime all work in 64-bit mode.

Cheers,

John

Can’t you already run 32 bit CUDA apps on a 64 bit kernel? The only restriction on running 32 bit code is that you can only link to 32 bit libraries, which most distros already come with. If there is some restriction on CUDA, then I definitely would be curious to know before we buy a GeForce 8800 for our lab, which is also all AMD64. (Not for memory reasons in our case, but because 64 bit code is faster on x86 due to the extra registers.)

You can’t run a real CUDA app on a 64-bit kernel, because you have to be running a CUDA-enabled display driver (9751) and thus far they are only available for 32-bit kernels. If you we had a CUDA-enabled display driver, then yes, I would think we could run a 32-bit app on a 64-bit machine.

John[

quote=seibert,Feb 23 2007, 03:59 PM]

Can’t you already run 32 bit CUDA apps on a 64 bit kernel? The only restriction on running 32 bit code is that you can only link to 32 bit libraries, which most distros already come with. If there is some restriction on CUDA, then I definitely would be curious to know before we buy a GeForce 8800 for our lab, which is also all AMD64. (Not for memory reasons in our case, but because 64 bit code is faster on x86 due to the extra registers.)

[snapback]163393[/snapback]

[/quote]

That is an excellent point I had not considered, and very sad as well. We were willing to compile and link our application in 32 bit mode, but we really don’t want to run the our AMD64 compute node with a 32 bit kernel.

I guess we’ll have to wait for the next 64-bit driver release.

Yes, even if we only have 32-bit CUDA to make use of, having 64-bit drivers would be a huge help.

I work for a company developping medical devices and we need huge computing and graphical efficiency.

We tested cuda on 32 bit and we want now test it on our 64 development devices. Our first test on 32 bits results are promising and we can increase our framerate and calculation performances by a x5 or x10 factor, but we need 64 bit driver/cuda ASAP for development and production integration.

Could you please let us know for your driver development timeline and scheduling ? even for beta versions.

Thank you

We have the exact same problem here and begining with a 64bit beta version of the drivers would be extremely useful.

Thank you.

Hi guys,

It’s more than just the display driver, there were issues in CUDA that needed to be fixed. So you’ll need to wait until the next release of CUDA.

Thanks,
Mark

We have a large-ish render farm of 8800s, currently using OpenGL to do image processing. But we can’t wait to port our code to CUDA. Currently we still use the CPU to do processing that is not possible through OpenGL, but most of it look very doable with CUDA.

Sounds like we have to wait, but just wanted to add my “vote” to this issue. We are running all 64bit linux here as well.

Thanks,
Ian

+1 for 64-bit version from me.

That was initial reason that i joined the forum.

I saw all the nice stuff about Cuda and how the development sw is free, but alas thing only works with nv88xx. :/

Well trip to the store solved that. I have caught the last 8800GTS 320 Mb on the shelf and ot replaced my trusty not-so-old 7900GT.

And then c*ap. Thing demands specific distro (but that is avoidable) and special kernel driver for the card.

I should have looked at that before, but it was obvious to me that anything new would be made 64-bit primarily…

Well, I’ll live, I suppose. :D

But I have one extra question about that kernel driver.
What is so special about it ? Is it because it is newer than any oficiall vanilla kernel module version or is it because it contains special functionality that vanilla drivers are not about to see ?

I ask this because my Gentoo ( 64-bit version) takes care of nvidia driver module by itself.
So, with next revision will vanilla module be adequate also for cuda ?

By simple logic it should, otherwise ordinary users couldn’t run your code, but it never hurts to ask, I suppose…

There’s nothing special about the 1.0-9751 graphics driver package other than it was tested & qualified to work well with CUDA.

Hello everybody!

we’re also waiting for the 64-Bit version. We’re currently running the 32-Bit one, but everything else is 64 Bit, so it would really help!

Some general feedback (hope it’s welcome?):

  • first of all a big compliment :thumbup: , with cuda it’s really very simple to learn for novices such as me! It took only 10 hours to get something quite useful, and I never got behind the shader programs!
  • I think the documentation is already quite good. For future versions I’d appreciate if you could put more emphasis on which function applies to which memory space in section 4. Also it’s not really clear how much memory is available for which space (except the shared and constant spaces). How about an example in the SDK using the constant space?

We are still waiting because our x86_64 platform does not supported (it’s also means we cannot order these G80 cards before it supported). Any ETA for the x86_64 public linux release?

The current 0.9 beta supports x86_64. It is currently in testing status for registered developers. I have it running since last week on a SuSE 10.2 box (a supported platform). I am not an NVIDIA engineer, but I would expect it to be out in a week or so.

Peter

Any update? Should I hold my breath?