NVIDIA reveals the first benchmarks of Tesla K20X
Last month, Nvidia announced its new parallel accelerators and Tesla Tesla K20X K20 , their first products based on the core second generation kepler “GK110″, promising a theoretical yield of 1.31 TFlops in double precision calculations (FP64). Today is Nvidia itself that gives us a preview of the performance of its flagship K20X Tesla.
The parallel accelerator “cGPU” Nvidia Tesla K20X consists of a total of 2688 shader processors organized in five GPCs, each composed of three SMX (each has 192 shader processors SMX), becoming the epitome of the family of accelerators Nvidia parallel intensive calculations for servers and supercomputers, Nvidia that compares itself to the intel xeon E5-2687W “Sandy Bridge-EP”, the epitome of the Intel family of microprocessors based on the Sandy Bridge microarchitecture (8-core 3.1/3.8GHz), and one of the microprocessors nominated among the best this year .
Here some of the Nvidia test:
According to the evidence we see Nvidia Tesla delivers performance K20X between 8-18 times (depending on the test used) than that offered by Intel’s microprocessor, which we know has a 198.4GFlops power in single precision calculations (FP32); compared to Tesla 3950GFlops K20X FP32 (20 times in theory).
As a curious note recall that a couple of years ago Tesla had a yield 14 times the Intel CPUs that time , so that things do not seem to have changed much since then.
We recommend reviewing the entire document from Nvidia.
Link: NVIDIA® Tesla GPU Accelerators K20-K20X Benchmarks (NVIDIA)Tags: accelerator, benchmark, cGPU, E5-2687W, gk110, intel, K20X, kepler, Nvidia, performance, sandy bridge ep, server, supercomputer, Tesla, xeon