Basic Linear Algebra Subprograms (BLAS) are standardized application programming interfaces for subroutines to perform basic linear algebra operations such as vector and matrix multiplication. We are the only company to actually propose a complete optimised BLAS library on GPU as we have the know how to implement double precision applications on the GPU.

Thanks to our technology and skills, we have implemented the most powerful BLAS library available on the market.

For matrix matrix multiplication, the performances on the GPU are around 10,7 times quicker in simple precision and around 3,7 times quicker in double precision compared to CPU performances.

In simple precision, the Nvidia GPU attained 80 Gflops while the ATI GPU 43 Gflops. We will redo this benchmark with an ATI R600 GPU as soon as possible.

In double precision, we surpass 14 Gflops whereas ACML Lib does not attain 4 Gflops.

The following comparisons were done on an AMD 64 X2 4600+ with 1 Go of memory (CPU) and an  ATI HD 2900 XT and a Nvidia Quadro 4600 (GPU).

Test it by yourself