NVIDIA GPU recommendation: Mac (OSX 10.6.x - 64bit server)

I'm trying to wrap me head around the differences between the; NVIDIA Quadro 4000 (Mac) & NVIDIA Quadro FX 4800 (Mac)

The 4000 offers 256 cores, while the 4800 offers 192 cores. The 4000 is significantly less ($780 USD vs $1450 - amazon.com).

The 4000 is compute capable 2.0, the 4800 is compute capable 1.3.

(the parallel toolbox requires 1.3 or greater last I checked).

Both support OpenGL

Is the main difference (besides the # of cores) the compute capability then? Is the compute capability tied to the card, or the driver (chicken or the egg, I know)?

I see there are some threading performance advantages to the 4800, http://www.geeks3d.com/20100606/gpu-computing-nvidia-cuda-compute-capability-comparative-table/

But the main question is; what is the advantage of spending almost twice the cost for the 4800?? (Especially when the 4000 offers 64 more cores!)

Both usable on 64bit osx/matlab?

Any help would be, well, a lot of help!

Thanks!

-ashton

Answer 1

[Walter Roberson]

The Quadro 4000 is newer than the Quadro FX 4800, and has better specifications in most areas. See NVidia's comparison chart here

I do not know what the price difference is due to.

Answer 2

[Jill Reese]

Both cards can be used on Mac OSX with MATLAB and the Parallel Computing Toolbox, although you should check out each card's spec page on the NVIDIA website for more details on the OS and driver versions required. I wasn't able to pull up the page for the 4800, but the specs for the 4000 are here:

http://www.nvidia.com/object/product-quadro-4000-mac-us.html

Look at the "Drivers & Downloads" tab for more information on the necessary set-up. Note that this is the set-up required by NVIDIA in order for this card to work at all (independent of MATLAB).

Answer 3

[Joan Puig]

I would suggest you look at the amount of RAM your problem needs and make sure your choice exceeds that amount, otherwise you will have to introduce unnecessary code to break the computations down to a manageable size.

While compute capability 2.0 might not be required for this particular project you are working on, I think it helps to ensure your hardware will stay current for a longer period of time.

Answer 4

[Thomas]

I would suggest going for the newer 4000 card as it has a better specs and occupies less space (only 1 slot as opposed to the 2 slots for the 4800). Also it will help you stay current longer because of the higher compute capability.

Answer 5

[King Fisher]

If you are o.k with 1.3 capability and not going to work for enormous amount of data, 4800 is enough. But you think that your algorithm needs higher GFLOPS then go for a costly one !