Will C++ AMP run on a machine without a compatible GPU?

Question

I understand that C++ AMP is accelerated by GPUs that support DirectX 11.
However, my question is, if the compiled C++ AMP program is run on a machine without a DirectX 11 compatible GPU, what happens?

Does it get emulated by some software implementation of DirectCompute?
Does it get executing on the CPU (perhaps using SSE style instructions)?
Or, does it just fail to execute?

Normi

I am trying to run a CUDA application on 2 different PCs where the first one has NVIDIA GTX 550 Ti and the other one has NVIDIA GTX 560. Both graphic cards are CUDA-enabled and DirectX compatible. The current version of DirectX in these PCs is DirectX11. However, in both cases the application returns the message: [Link.CUDA.GaussDerivative] CPU implementation fallback (kernelRadius = 120). Have you any clue of what is going wrong? – Catherincatherina

Answer 1

Indeed C++ AMP has a CPU fallback (multi-core plus SSE) implementation called WARP (aka "Microsoft Basic Render Driver"): http://www.danielmoth.com/Blog/Running-C-AMP-Kernels-On-The-CPU.aspx

Scabrous

Thanks Daniel! Good to see you are patrolling stackoverflow. :) – Normi