Intel Corp (INTC)

[Unkwn]

I am saying that if I have a hugely parallel task that will benefit from acceleration from a GPU or GPU-like product (i.e. Xeon Phi), then a GPU that can deliver much higher peak performance is probably preferable.

There's a reason that Xeon Phi still has minority share in the HPC accelerator market compared to NVIDIA's Tesla GPUs.

The question as to whether Xeon Phi will be a success or not depends on how many computing tasks performed on such computers (which is not just supercomputers) will benefit from the theoretical compute power of GPU accelerators. I have no answer to this, but for broad market application, ease of programming, existing tool support, general applicability and so on are much more important than any theoretical flop figure.

So, in the end, the broad success of Xeon Phi won't depend on flop figures but on tool support, general application and so forth. I don't know whether it has an advantage there, but I just wanted to question the general statement that beating NVIDIA will be hard due to some GFlops here and there.