Inexpensive Nvidia cards that support CUDA programming inexpensve hardware

Accelerated Computing CUDA CUDA Programming and Performance
1

I’m at PhD. student studying computational fluid dynamics and just beginning to get my feet wet with parallel computing. Right now I am learning MPI but I’m interested in learning CUDA and GPU coding as well.

I’m wondering if I can build an inexpensive PC that will allow me to experiment with CUDA and GPU programming. Can someone tell me what the the least expensive Nvidia card is that will allow me to practice programming with CUDA? Does anyone have any other advice for someone who is interested in building an inexpensive box just for learning GPU programming?

My thinking is that if I build an inexpensive box and get to a point where I can demonstrate the ability to write efficient code in CUDA then I might be able to convince my adviser to front the money to put together a TESLA box for our research group.

2

Just ordered http://cgi.ebay.com/ASUS-1015PN-PU17-N550-1-5-1-250-10-1-W7S-NT-BU-/320691489190#ht_9479wt_1141
Haven’t received it yet but the ION2 is CUDA compatible. Here’s a link to a the device properties http://www.pixel.io/storage/post-files/deviceQuery.txt
With a RAM upgrade and installing windows 7 home you have a tiny netbook that acts like a laptop and is CUDA capable for about $300.

If you have working code and want to test it on a large scale then places like UC Berkeley have GPU testbeds and are open to outside applications.

3

Absolutely you can get a cheap card. CUDA is designed to scale nicely when you later want to use more powerful GPUs.

The cheapest card that will work is probably only $65 or so. (You can get even cheaper ones but I’d recommend using a Fermi based card even at the low end just so you have the most modern capabilities).
You would get a powerful, nicely balanced, modern GTX560 card for $230.

4

em… if you are looking for something with the best bang per buck, it is without doubt the GTX 460. It has 7 MPs so it brings you closer to the reality of more powerful cards. Price is around 150 usd

5

Is it free? What cards they use? Could you please give the link? Sorry, I couldn’t find it…

By the way, if your program need more memory, GT 430 is a good choice (CC2.1, 1GB, 48x2 SP, GF108). It’s cheaper too (around 75 USD)!

6

Thanks everyone for the good advice

7

Wessey - Thanks for the idea. I can see where just getting a cheap netbook with the ION2 would have it’s advantages. For one I wouldn’t need to build a system around it. On the other hand if I did build a desktop system I could always start with a very inexpensive card and then upgrade the Nvidia card if my programming skills and code require it. I’m not sure what direction I will go but probably won’t do anything for a month or so, as I get up to speed on MPI.

But I would like to know more about the netbook option. Once you get it and have a chance to play around with coding and running code on it, do you mind replying back with your thoughts of how well it works? The Nvidia page for the ION2 says it has ‘up to’ 16 cores. Do you know how many cores their are in the ASUS-1015PN-PU17?

Thanks again for your thoughts.

8

second that! got it also :)