Intel is lacing up its jackboots to go Nvidia stomping.

Seems we can expect the semiconductor giant to commercialize a better-than-50-core co-processor code named Knights Corner, a descendant of its sorta aborted Larrabee GPU adventure, in 2012-13 to compete against Nvidia's Tesla GPU accelerators that currently own the HPC co-processor space.

Intel's cores will all be x86 apparently packed on a single ring bus and reportedly programmable with existing x86 tools that - allowing for parallelizing the code - are supposed to be easier to use than Nvidia's proprietary CUDA platform although Tesla would probably result in faster machines.

Next year Nvidia should introduce its Kepler GPU, reportedly three times faster than what it's got now.

Intel's expected widgets represent what it calls its Many Integrated Core (MIC) architecture resulting in a general-purpose accelerator for highly parallel applications.

MIC is supposed to lead to exaflop computing, 100 times faster than current machines, long about 2018. Speculators suggest it's thinking of an integrated MIC-Xeon part for exascale supercomputers.

So far what Intel has been able to salvage from Larrabee is a prototype called Knights Ferry that it and six of its OEMs showed off in systems at the International Supercomputer Conference this week.

The vendors included Dell, HP, IBM, SGI, Supermicro and Colfax International. They've been working with the part for the last year or so and Intel has got more than them, with more expected.

The 45nm single-precision 32-core Knights Ferry, a MIC design called Aubrey Isle using 1.2GHz cores, will never go commercial. Intel wants to use its newfangled 22nm Tri-Gate process for Knights Corner and support double precision floating point operations.

Intel hasn't confirmed much else about the thing.