[Divyaprakash]

I am planning to buy a new laptop and I have just picked up some interest in GPU computing for it's applications in CFD. I had planned to buy a laptop with 2 gb graphics card, but now I am thinking about buying a 4 gb one. Should I make the extra investment? Also is CUDA supported on NVIDIA GeForce GTX 850M? I will begin to learn CUDA once I get the laptop.

What might be the best specification for a beginner like me.

[ssss]

GPU computation is worhtless in current CFD state. Eventhough if you want to use it you should go to a QUADRO card which would cost thousands of euros.

Why do yo say a 4GB video card is better than a 2GB video card? It has barely nothing to be with each other.

Last thing, a laptop for computing? nonsense

[wyldckat]

Greetings to all!

@Divyaprakash: If the laptop is meant to be used for academic development of CFD code using GPU, then the more memory equates to bigger meshes or more equations that can be solved.

The downside is that sometimes a card that provides more memory, might be using older RAM, such as DDR3, instead of GDDR5. This means that even though you can have more memory, you can't have as much speed using the memory.

But as "ssss" indicated, using a laptop for hard-core CFD is not advisable. Because for the same price of a laptop, you could get a desktop computer that is 2 or 3 times more powerful.


As for NVidia graphics cards that support CUDA, I think they have a list for that somewhere online. Try searching for:

nvidia CUDA-ready cards

Best regards,
Bruno

[saifmulla]

Well firstly I would like to remove your misconception that NVidia cards only do GPU computing it is place where their proprietary language CUDA works. Although GPU such as AMD, intel HD series cards are also build for compute processing. Besides they can be programmed using OpenCL, by programming OpenCL you code becomes portable to NVdia, AMD, Intel GPUs also multicore CPU and intel Xeon phi.

Considering your choice of doing GPU development on laptop isn't wrong, its better to have a portable hardware available with you anytime & everywhere.

If you are looking to only develop you applications on laptop and then run it on a server class GPU then it's a fair choice but if you are looking for running simulations on laptop then its a bit difficult. Apparently some one suggested to select Quodro GPUs, well actually Quodro GPU have less cuda cores than GForce although quodro's are more expensive than the former. Quodro's are mainly developed for visualisation and professional graphics while GForce are for gaming but Gforce are good for development as they have more CUDA cores so you get more compute performance the choice is always yours.

Apparently the class of GPU you looking is made for Gaming although all GPU (NVidia) presently supports CUDA & OpenCL, but problem with gaming GPU is they have few double precision floating point units on them and the application you planning to run are mainly CFD that demands deeper precision.

If you are seriously looking for both of worlds as in, development and simulating on same GPU then I would suggest go for desktop class GPU mainly higher range Quodro or a Tesla. Besides if you are considering OpenCL development then I would recommend AMD for GPU computation, recently they launched S series GPU with far better double precision performance than the current generation quodro and Tesla and also they support OpenCL 2.0.

Hope this helps.