Managed code is just-in-time compiled by the .NET framework. Unfortunately, the JIT compiler can only use the x87 floating point unit. Most native numerical libraries can use special SIMD instructions (SSE and beyond) to vectorize the typical floating point operations that are used in matrix multiplication and other linear algebra methods. There is not much we can do about this except offer you Math.Net with a native provider.

Perhaps you can vote on Microsoft's Connect site on the suggestion to add access to SIMD instructions from .NET. That way, our managed implementation could become as fast as any other native implementation.

Last updated on May 21, 2010 by Math.NET Team