During a recent webinar one of our attendees asked us the question:
“Which hardware is faster – the GeForce GTX 690 or the Tesla K10?”
The short answer: GeForce GTX 690.
That isn’t the entire story though. The specifications for both cards are shown in the table below:
||Number of GPUs
||Memory BW (GB/s)
||SP Performance (GFlops)
While the raw throughput on the GeForce GTX 690 is higher than the Tesla K10, the memory on the GTX 690 is half of the size of the K10. Depending on your application, memory size may outweigh the benefit of the slightly higher throughput on the GeForce.
Additionally, NVIDIA’s Tesla product line has features not available for GeForce hardware:
- ECC Memory Protection – improved memory reliability
- Manufacturing Quality – hardware is manufactured and tested by NVIDIA to guarantee highest reliability
- 1.5 to 2 Year Availability – useful for enterprise customers who need to replace or add hardware after initial deployment
- Cluster and GPU Management Software – diagnostics useful for monitoring hardware in clusters
- Advanced Driver Support – TCC drivers allows the use of Windows Remote Desktop and avoids the Windows watchdog timer when running large kernels
- Form Factors – Tesla is supported in certified servers, blades and workstations
The full list of Tesla benefits can be found here: www.nvidia.ca/object/why-choose-tesla.html
In almost all cases, our customers deploy Tesla hardware in production environments. The memory size, reliability and long availability outweigh the modest performance benefits of the GeForce product line.
Finally, for double precision computations you will want to consider the K20X or GeForce GTX Titan since the double precision performance on the GTX 690 and K10 is approximately 1/24th the single precision throughput. But if you can get away with single precision calculations, the GeForce GTX 690 and Tesla K10 are great hardware choices!