I came across this benchmark comparison ( http://compubench.com/compare.jsp?config_0=14470292&config_1=18133965 ) between GTX-Titan and Tesla K40c, and the results seems very odd.Can someone please explain any possible reason behind this. How much would Tesla differ in performance if only single precision operations are being performed?

有帮助吗?

解决方案

I'd opt for the Titan because it has more up-to-date instruction set capabilities. The GK110 ASIC that it is based on has SM 3.5 capability (e.g. read-through-texture, funnel shift) and also supports dynamic parallelism.

Maxwell is coming soon, but it will support more capabilities, not fewer, so you should use the GK110 as a baseline.

许可以下: CC-BY-SA归因
不隶属于 StackOverflow
scroll top