Workstation GPU: Quadro K6000 or Teslta k20x or GTX titan?

Accelerated Computing CUDA CUDA Programming and Performance
1

Which one is better? And a perfect combination of visual design and parallel computation?

Thanks.

2

Unless youre doing some very specific things (multi gpu with heavy P2P usage, interaction with other hardware), the Titan will get the job done. It’s the second fastest out of the three and by far the cheapest. The only other reason you might get a K6000 is that you need 12GB of memory (like I do :P)

3

Depends on what you are doing: For private projects on a single workstation a GTX titan might certainly suffice:

4

The only reason to pick a K20Xm over Titan is if you need TCC mode/ECC memory support and you want the warranty/support that comes along with it. Like sBc-Ramdom mentioned, the K6000’s benefit is the additional 6GB VRAM along with TCC mode/ECC memory and PCI-E 3.0 support, as well as warranty/support that comes with it. Otherwise, just go for the GTX Titan which has 6GB VRAM, PCI-E 3.0 support, and nearly the same computational power as the K6000.

5

I use Titan for some time. Pretty fast and unless you really need ECC you can go along with it. The k6000 has 12 GB.

6

Another option is to wait for the K20 successor the K40 (Atlas) that like the K6000 also has 12GB and 2880 cores.

7

Like people mentioned the Titan is an excellent option regarding price/performance.

The K6000 is actually a bit more powerful with its 2880 FPUs and relatively high clock it delivers roughly 5.2 TFLOPS of SP performance. All in an impressive 225 watt power envelope.

Not bad.

8

And going down the sliding price scale, the soon to be released GTX 780 Ti is likely to have the same single precision performance and memory bandwidth as the K6000, but with only 3 GB of device memory.

So, to summarize everyone’s suggestions, deciding on the best card requires you to answer a few questions about your needs:

  1. Do you care about CUDA performance on Windows? If so, you want a Tesla card so you can run TCC drivers. (On Linux, it doesn't matter which card you use. The driver is fast for all of them.)
  2. Do you care about GPU acceleration in various proprietary CAD/graphic design/animation programs? If so, check their documentation because many only support these features with a Quadro.
  3. Do you care about memory error-correction, GPU acceleration inside virtual machines, or simultaneous bidirectional card<->host transfers? If so, you want Tesla or Quadro.
  4. Do you need a lot of device memory? If so, you want Quadro (for 12 GB), Tesla (6GB) or Titan (6GB).
  5. Do you care about double precision performance? If so, you want Tesla, Quadro or Titan.
  6. Do you only care about memory bandwidth and single precision performance? If so, then a GTX 780 is nearly as fast as a K20X or Titan, and the GTX 780 Ti will likely be as fast as the K6000.

Hope that helps without being too overwhelming. :)