Latest update on CUDA programming on Laptops 8600M GT

Dear Friends, :)

I am relatively new to GPGPU and CUDA. I have planned to do some communication related research using GPUs with CUDA. Basically the algorithms involve the following computations:
Compute 64 sets of 64-point FFT
Computation of 33 covariance matrices in frequency domain.
Computation of eigenvalues and eigenvector
I want to do all this on my laptop. I did a thorough research and found that 8600M GT (has 32 Stream Processors, while 8600M GS has only 16) will be both economical and efficient among the available mobile versions of CUDA-enabled GPUs (at least for some prototyping and to check that my algorithms are indeed working. The actual testing will be done later on Tesla). Since I want to develop the applications on my laptop I have zeroed on DELL Inspiron laptop with Windows Vista/Linux and NVIDIA GeForce 8600M GT, 256 MB.
I have following questions my mind:
1-While buying a laptop, what are the features we must look into if we want to do some serious GPGPU for non-graphical applications (like memory, number of stream processor, GT versus GS etc). What if we want to do some graphical GPGPU?
2-How much system RAM is sufficient (I am planning for 1 GB; it will cost me a bit cheaper)? Shall I go for 2 GB or more? What about GPU memory (GeForce 8600M GT has 256 MB).
3-What is the latest status of support for CUDA on 8600M GT by NVIDIA (driver’s availability etc) with Windows Vista/XP and Linux.
I want to know all this before buying the laptop, since I do not want to invest huge cash in buying a toy laptop!

Any suggestion will be highly appreciated.

Kind regards,


Two short pointers -

Get 2GB of RAM, preferably 4.

Check out the new 9700 M GT cards, they’re similar to the 8600 M GT but more power efficient, and are IMO probably going to be more widely available.

That said, I just got my hands on a very nice Macbook Pro for a good price, CUDA seems to run very well under Linux on it.

Thank you for your reply.

I think 9700 M GT is not CUDA compatible. What do you think??

List of CUDA compatible GPUs:

Kind regards,


Unless you run a 64-bit OS or a 32-bit OS with PAE available, 4GB of RAM could be a waste. I am really not sure, if PAE (Physical Addres Extension) can be used by 32-bit OSes. Can some1 enlighten?

Why I say this is : With 4GB RAM and a 32-bit OS, IO devices have to be memory mapped only from the 32-bit 4GB space. This will cause the OS to hide some RAM inorder to accomodate other devices…

Best Regards,

Sarnath aka Kandi

All GPU’s after G80 are CUDA compatible. The list is not always up-to-date.

Thank you very much Sarnath and E.D! :)
Humm… Sarnath could you please elaborate your point more precisly. Do we need to have PAE with 32 bit OS’s? I am sorry I could not understood your point,especially when you say:
“This will cause the OS to hide some RAM inorder to accomodate other devices…”


32-bit operating systems normally only have 4 GB of address space to work with. This is shared between system memory as well as I/O to graphics cards and other PCI/PCI-Express devices. For example, on our systems, when we installed 4 GB of memory, only 3.1-3.3 GB was actually available to the operating system. The rest was hidden by the address block assigned to device I/O.

Current motherboards can “hoist” this hidden memory up to a higher address (beyond the 32-bit/4GB limit), but to see that address space, the operating system has to run the CPU in 64-bit mode. (PAE also gets around this problem while in 32-bit mode, but I don’t have any experience with that.)

I’m using 64-bit Linux on a Macbook Pro. Works fine. There may be drawbacks to using CUDA on an 64-bit os, though, but it’s worth it for me.

BTW, Mac OS X sees the full 4GB as far as I know.


I am also a CUDA developer here in India. At the labs in our institute, we have good hardware to do actual performance testing, but at home I use my new Dell XPS M1530 with the GeForce 8600 M GT.

This is decent enough to develop and test CUDA programs, and u can always deploy them later on better hardware for actual benchmarking. the 8600 M GT has 32 stream processors, and is CUDA 1.1 compute capable.

The fact is that it’s a very decent price here in India, I got my laptop from Dell India at around 66K INR. Trying to look for any laptop with a faster video card that is CUDA compute capable will end up costing in excess of 1 lakh (100K) INR here in India.

I fully recommend the 8600 M GT and Dell XPS M 1530. You can also check with HP who also provides the same card on some of their pavilion notebooks. If you really don’t want to spend that much, the minimum spec required to deploy CUDA on a laptop is the 8400 M GS, u can get this card on a laptop here in India for under 45K INR.

Go for a minimum of 2 G RAM. As far as development is concerned, I use Linux, and 8600 M GT is fully supported with CUDA, NVIDIA’s driver and im using Fedora 8 64 bit.

Let me know if u need any additional help here…

You might want to check on the Gateway Model: P-6860FX. It has an 8800M GTS GPU with 64 stream processors and 512MB of ram. The laptop comes with 3 GB. In the U.S., BestBuy is selling them for $1250.

Thank you everyone for your comments and sugggestions! :)

Warm regads,