Intel's cancelled Larrabee is set to get a new lease of life in the world of supercomputers.
For those who came in late, Larrabee was Intel's project to develop a graphics chip to work alongside a conventional computer processor.
It was officially canned ages ago, but it seems that the design is not quite dead.
According to the Australian, Intel has been watching Nvidia and AMD use GPUs in the supercomputer market and has been thinking "I will have some of that!".
To do that, Intel needs to bring out a chip that can speed up performance and compete with GPUs will be crucial to defending a market.
Now it seems that the new chips will be based on Larrabee and a prior research effort called the "single-chip cloud computer."
The initial chips, code-named "Knight's Corner" will offer more than 50 processor cores. It depends on a manufacturing process that won't be ready until the second half of 2011 so we are unlikely to see it until 2012.
This is a long wait, but some software writers might think it is worth it. Intel's chips will be based on technology it popularised in PCs and standard servers.
If they use graphics chips from Nvidia or AMD that have more cores they will have to redesign software applications that will be expensive and time consuming.
Larrabee was never cancelled, the only part of it that was back-burnered was the consumer graphics adaptation. Intel always intended to fab a version for HPC.
Anonymous: we are talking multiple x86 cores, which can be readily handled by existing dev tools, unlike OpenCL or CUDA, which require new tools and new paradigms of thought, leading to rewrites of applications. Larrabee would allow a currently multi-threaded x86 application to run with no real change.
total Incompetence, Little Likelyhood of Overcoming AMD Patents, let alone Know much about desktop Processing.
x86 IS Over, Gone, Done. Yet, after 50+ years of
Night of Living Dead. Artemus Special. Wasting Ones Time on Hopes with larabee Is Like Hoping Handicapped Olymics Will Produce Athletes.
Computing Projects with code name come & refresh, plenty are similar to larabee, NO Results, Just More Cost, Burdened Upon Simple Contract Signatories. Now theREADER Knows of One, Hundreds+ Remain in isolation, just printing more & more precious Cash, for Nothing. You Pay TAX, You Get Nothing In Return. Larrabee Scam that Public Can see now, theres Too many to even guess at vastness in computing, in every direction, throughout industry. maybe its graphene, maybe is liquid nitrogen poured on head. maybe Its room of broken equipment or even dead personel. No Limit, apparently, To gullibility of Peoples.
Signed:PHYSICIAN THOMAS STEWART von DRASHEK M.D.
HE-EL-LO-O. DO-OC-TE-ER. NA-AI-ME.
CO-ON-TI-IN-NU-UE-YE-ES TU-UR-DA-AY,
TO-O-O-MO-OR-RR-O-OhhhhhhhhhhhhhhhhhhWWWWWW
For starters, Larrabee is massively parallel too. So, it doesn't so much represent the paradigm we're used to - although admittedly this changed in the last few years.
But more importantly, GPU computing is already becoming more prevalent. SDK's and other development tools have already been made and will see a stronger coming still due to backing from nVidia's own work and AMD's Fusion Fund. AMD's Fusion chips will further accelerate this, as it makes GPGPU commonly available. Just imagine where we are in two or three years time...
With these trends in mind, as well as the performance difference, don't you think that Larrabee might be a case of "too little, too late"? Or perhaps more like a stopgap along the way to true GPU compute, which may be where we really want to go?