CUDA on mobile 8x series...?

…specifically, the 8600m GT as found in the new Mac Book Pros… (Yes, I am a Mac person… :) )

I know that the memory size is relatively small (256megs), but it would be incredible if I could tinker on projects while ‘away’, even if it were limited.

Drop into Bootcamp, do a little Slipstreaming and… whoa-hoe! There’s Fred!

Possible?

dave

Yes, CUDA should work on mobile G8x parts, assuming you can find a CUDA-compatible driver for your particular laptop.

We haven’t tried it, but CUDA should also work on the new MacBooks if you boot Windows using Boot Camp.

If you’d like CUDA to be supported on MacOS, please ask Apple.

Would these MacBooks be devices of type 1.0 or 1.1?

1.1

Hmm, it seems the answers here are sort of confusing me… shouldn’t the general series 8 driver work fine on laptop models as well?

I suppose it’s better to be safe than sorry, so here’s my specific inquiry: can anyone here confirm that CUDA works fine on the A8SC laptop by ASUS? This is the one that I am considering buying. The low performance of the 8400M G isn’t much of a problem from a developer’s point of view, and may even be preferable if it increases the battery life of the laptop.

It sure would be nice if we could get some solid information (telling us if CUDA works on this particular laptop or not) for each of the laptop models listed in this link:
http://www.nvidia.com/object/wtb_notebooks.html#eightSeries

You want 256MB of memory on the GPU.
We will try to compile a list of models known to work.

Thank you, I am looking forward to this list. :)

But… is there some important reason that I don’t understand why I would want 256MB of memory on the GPU? The laptop I am considering only has 128MB GPU memory, and that sounds adequate for my needs. I reckon that with only 8 streaming processors, you don’t want to process very large data sets anyway.

What is the reason for wanting 256MB for CUDA in general? It seems to me this would only depend on the developer’s personal needs? Or is there some extra overhead with CUDA that I am not aware of?

A bigger concern IMO is that it only has a single multiprocessor, so bugs due to unwanted block-dependencies wouldn’t show up, since it would only execute a single block at a time. But I figure this will be easy to avoid with a little forethought and regular testing on a more capable member of the Geforce 8 series.

CUDA needs some extra memory.

Is there a CUDA compatible driver available? (or planned)?

We are testing CUDA with mobile GPUs.

Could you please give a little more information on what this means? What “invisible memory usage” is there to consider? Does the CUDA driver itself eat up several tens of megabytes, or even more?

Even as little as 64MB of available memory sounds ok to me for development purposes… so I really would like to have some estimation of how many megabytes that won’t be available for allocation.

Any progress on this yet? I’m going to my country’s capital this weekend, and I would love to return with a CUDA compatible laptop. But to do that, I really need to know which models will work, and which are known to fail.

Thanks in advance.

The list is not going to be ready by tomorrow.

Any idea of when this list is expected to be done? Are there any particular models you have been able to test yet, which you could name at this time?

Will these be subject to the 5 second limitation?

Thinking of getting the upcoming Thinkpad T61p. Will the GFX card be supported for CUDA?
Quadro FX 570m is not listed on the supported card list, but I figure its pretty much the same hardware as the new Geforce8M lines.
Will it be supported too?

Yes, the Quadro FX 570M is based on the G8x architecture and should work with CUDA.

The issue with laptops is that they sometimes require customized drivers to support special features of the laptop. We aren’t currently distributing drivers for GeForce 8M or Quadro FX M on our website, so you should check with the notebook manufacturer that they provide a driver that is compatible with CUDA.

any results yet?

Before reading Simon Green’s post above, I placed an order for a Zepto Znote 6224W laptop, thinking that an 8600M GT card with 512MB should work without problems.

It arrived yesterday, and my CUDA experience in XP is a mixed bag. I have tried running the executables in the SDK, and the good news is that it seems like regular CUDA is working fine. The bad news is that I have had no luck with the executables that use OpenGL interoperability to draw stuff.

DeviceQuery reports my 8600M GT just fine, and the bandwidthtest gives impressive figures. Executables such as MatrixMul, scanLargeArray and simpleTexture work too. But SimpleGL just gives the error message “GPU interoperability on multi GPU systems currently not supported”, and so does fluidsGL. Interestingly enough, simpleD3D seems to work fine while fluidsD3D just prints the error message “failed to find a G80”.

As Simon Green hinted, the Nvidia drivers refuse to install, telling me that no supported hardware could be found. So I am stuck with the drivers provided by the seller.

I noticed the comment about OpenGL interoperability not working when using multiple displays and was hoping I had found the issue. But setting this option to “single display performance mode” did not change the behavior in any noticable way.

Is there any issue that I have overlooked that could be the cause of OpenGL interoperability not working? Or is there a way to make the 8600M GT pose as a regular 8600 GT card and install the regular drivers?

I invested a fair amount of money in this laptop expecting to use it for CUDA development, so the thought of not being able to use the hardware due to driver issues is quite depressing to me. I would really appreciate any sort of guidance that could help me get CUDA working fully on this laptop.

What is the driver version installed on the laptop?