Tesla C1070 vs Geforce GTX 295

Hi,

I am looking to but the most powerful CUDA enabled card I can and was looking at the specs for the Tesla C1070. Is it that much better than the Geforce GTX 295? Am I just better spending 1/3 of the money and getting the Geforce?

http://www.nvidia.com/object/product_tesla_c1060_us.html

http://www.nvidia.com/object/product_geforce_gtx_295_us.html

Cheers

Iain

AFAIK, the Tesla C1060 is identical to the GTX295 but for the quantity and quality of its memory. The Tesla series have 4GB of high quality memory, meaning they will run for weeks (I’ve tested) without memory errors. The gamer-level cards like the GTX295 are likely to produce memory errors on the order of 1 per day (assuming they’re running at full-pelt 24/7). Whether this is significant or not depends on your application.

The Tesla C1070 is just 4 C1060s clocked slightly higher in a big, loud box, designed to sit in a server room rack.

EDIT: It appears that the GTX295 is actually a dual-GPU board (I haven’t been keeping up), meaning it has processing power equivalent to two C1060s.

So all things considered I’m probably just better with the GTX295 as although it has about half the memory, it has more processing power and a faster memory bandwidth? Oh and it is a 1/3rd of the price.

Careful.

The Tesla C1060 has 4GB for one GPU.

The GTX 295 has 896 MB for each of two GPUs. This can make a severe difference for some types of computation, and it can be irrelevant for others.

I will be using it mostly for image processing applications where i’m looking for fast mem copys and V-fast computations keeping everything below 10ms (which I suspect most of which will be the memcopy). I already have a GTX 280 which perfroms very well but I’m looking for the next step. Is the 295 essentially the power of 2 280’s?

Yes, but the GTX295 will appear as two separate CUDA devices, not one ‘big’ one.