CUDA (and GPUs in general) seem to perform well only with SIMD workloads. I think Larrabee is likely to be more flexible than GPUs at the cost of a little SIMD

CUDA (and GPUs in general) seem to perform well only with SIMD workloads. I think Larrabee is likely to be more flexible than GPUs at the cost of a little SIMD performance. I also think the advantage of being on 22nm with the FinFET process will prove to be significant because NVIDIA and co. are likely to be stuck on a conventional 28nm process for a while. It has the potential to be a good part.