NVIDIA Benchmarks.

[lambchops511]

Here's some quick benchmarks for those interested. Benchmarks were repeated 32 times and the average number reported.


GT 520
Memcpy within GPU (saturated): 3.2 GiB/s
SP GEMM 2048x2048: 78.1014 GFLOPS

GTX 580
Memcpy within GPU: 80.8081 GiB/s
SP GEMM 2048x2048: 1 065.42 GFLOPS

GTX 680
Memcpy within GPU: 72.7273 GiB/s
SP GEMM 2048x2048: 1 450.54 GFLOPS

Tesla C2070
Memcpy within GPU: 55.1724 GiB/s
SP GEMM 2048x2048: 692.388 GFLOPS

Tesla C2075
Memcpy within GPU: 50.000 GiB/s
SP GEMM 2048x2048: 691.517 GFLOPS

Tesla M2075
Memcpy within GPU: 58.1818 GiB/s
SP GEMM 2048x2048: 695.014 GFLOPS

Tesla K20c
Memcpy within GPU: 70.1754 GiB/s
SP GEMM 2048x2048: 2 421.83 GFLOPS

Tesla K20m
Memcpy within GPU: 78.0488 GiB/s
SP GEMM 2048x2048: 2 432.55 GFLOPS


For reference (other benchmarks):
http://www.anandtech.com/show/6774/nvidias-geforce-gtx-titan-part-2-titans-performance-unveiled/3
http://blog.accelereyes.com/blog/2013/02/21/benchmarking-tesla-k20/
http://blog.xcelerit.com/benchmarks-nvidia-kepler-vs-fermi/

 

[nenforcer]

Interesting to see the Fermi flagship take the Memcpy performance within the GPU. I actually want to upgrade to a 560TI for the higher compute performance but the prices are too close now and the Kepler cards perform better for gaming.