Comments on: Intel Xeon Phi: a Teraflop supercomputer in a PCIe card

By: dsol

dsol — Wed, 18 Jul 2012 10:39:00 +0000

Is this the zombie remains of Intel’s “Larrabee” GPU project, rebranded as a HPC component? 512bit wide SIMD was definitely a Larrabee feature.

By: CyberGusa

CyberGusa — Wed, 20 Jun 2012 15:25:00 +0000

No confusion, how effective the product will be in real world use remains to be seen but it’s suppose to be cluster scalable and specifically intended for parallel computing environments.

By: digi_owl

digi_owl — Wed, 20 Jun 2012 14:08:00 +0000

Let us not confuse linear in the sense of increasing processing power, and linear in doing the computing with one step following the next.

By: CyberGusa

CyberGusa — Mon, 18 Jun 2012 16:07:00 +0000

In the server market highly parallel is pretty common, and is preferable for saving space and costs, mind the stated Linux support as well.

By: digi_owl

digi_owl — Mon, 18 Jun 2012 15:30:00 +0000

Not sure if it can be considered linear. It is almost a step backwards towards how the Altair shipped, there all the actual components was a daughterboards like the GPU and maybe audio is today. This seems to move the CPU back out alongside the GPU, meaning that a motherboard of all 16xPCIe could hold plenty of these. But to actually make use of that computing, the code needs to be highly parallel.

By: Tim Carr

Tim Carr — Mon, 18 Jun 2012 15:20:00 +0000

So…will this actually result in a substantial step forward in processing power, or is it another linear step? I can’t tell from the article.