Pages Menu
Categories Menu

Posted by on Jun 27, 2013 in Hardware News |

AVX AVX 2.0 vs. 1.0 on Linux

AVX AVX 2.0 vs. 1.0 on Linux

Microprocessors Core fourth generation “Haswell” are just for their first month since it was released, and among its novelties have a slightly higher yield per cycle of its predecessor and new instruction sets :

  • Transactional Synchronization eXtensions “TSX” (not available in the K Series).
  • Advanced Vector Extensions 2.0 “AVX2″.
  • Fused Multiply-Add 3.0 “FMA3″.

Unfortunately no existing application leverages these new instruction sets (except the Linpack 11.0.3 ), so it has not been proven how beneficial they are.

From Phoronix we get some tests done to an Intel “Haswell-DT”, which proved with some applications compiled with 4.8.1 compiler (supports 1.0) and GCC (supports ), in order to check the gained when using the AVX 2.0 instruction set over its predecessor.

We have developed a graph gathering your results:

AVX AVX 2.0 vs. 1.0 on Linux image 2

The Himeno Pressure Solver is showing the largest increase performance by AVX 2.0 (51.45% higher), although performance is remarkable 8.14% higher on the test BLAKE2.

Much remains to make its appearance a large number of applications written to take advantage of AVX 2.0 and 3.0 FMA, which for now we have no benchmark showing performance.

Link: GCC 4.9 Running With Intel’s Core i7 “Core-AVX2″ (Phoronix)


Tags: , , , , , , , , , , , , , , , , , , ,