EBFE (x.delete@this.y.com) on August 4, 2012 6:37 am wrote:
> However, I am assuming the following disadvantage, so I was expecting
> more flops and flops/W.
> 1. Coding for a target performance level is harder for
> KNC than GPU.
> KNC has to close the gap of low raw flops by significantly
> higher effciency, which could be impossible or make coding harder.

Eh? No. Just no. People have ported full on top end super computer programs to MICs in under a week involving extremely large code bases. Many of them. It takes orders of magnitude longer to port small subsections to run on gpus. One of the primary advantage of MIC is that it is the easiest target to get running and optimize for with the most supported tool sets in the world.

> 2. KNC is
> more expensive than GPU. (SNB-EP is already near $2k)
>
Not in the market it competes in. A Tesla M2090 runs $2500.


> 3. On-borad RAM is
> smaller than competition, which may lower real efficiency.

Both Intel and Nvidia are talking about a max of 8GB for their top end products.

> 4. SP performance is
> incompetent.
>

And no one really cares.


> 4.5 KNC trails GPU in some GPU-favorable workloads.(e.g. where
> GPUs are used for now)
>
you mean games...