My group is considering buying some GPUs. I found a website with some interesting comparisons.
3200+ bucks Tesla K20c doesn’t perform as well as 1000+ bucks gtx titan. Is it true? If we only perform single precision computation, then there is no advantage for Tesla over GTX, right?
http://www.game-debate.com/gpu/index.php?gid=1582&gid2=945&compare=geforce-gtx-titan-vs-tesla-k20
Are you in the market for a GPU to do CUDA development on, or are you looking for a GPU that can run production CUDA apps around the clock with high reliability? You may want to peruse the following two pages:
http://www.nvidia.com/object/why-choose-tesla.html
https://developer.nvidia.com/ultimate-cuda-development-gpu
That comparison seems to be focusing on games, and the K20 does not even have a video out.
The Titan is awesome, but for large, long(in duration), and precise multi-gpu calculations, the K20 is a better choice.
If you are sure you only need single-precision , and will not have a multi-GPU setup go for the Titan.
If you don’t need double precision and don’t need 6 GB of memory, then the GTX 780 might even be a better choice. It is “only” $650, nearly as fast, and is still a compute capability 3.5 device, so you can play with dynamic parallelism and the read-only cache.