Published in Processors

Larrabee ipso facto will half not bee


GPU not interested in graphics
Chipzilla has lifted the kimono on its Larrabee 50 core GPU but is telling us that it has no intention of using the technology to run graphics.

Larrabee was part of a cunning hardware software GPU effort that tanked in 2009. At the time Intel said that the hardware part of the project would be resurrected in a new project.

Now it looks like the Larrabee GPU will finally go into commercial production next year, but as a glorified coproccessor. Its 50-core goodness will be fused on Intel's 22nm chip and targeted at giving Nvidia a heart attack. Nvidia has also been trying to flog its Tesla  GPU's to data centres and high performance computing manufacturers with no interest in graphics.

It is now looking like 2010's Knight's family of HPC coprocessors was just a test run for Larrabee in this market. The Knight's family was shipped in limited quantities and Chipzilla started to say that it was based on Many Integrated Core (MIC) architecture.

Today's announcement is that the Knight's family is now ready for heavy commercialisation. First off the bloke is the 50-core Knight's Corner chip on 22nm. Intel said that it will work with GI and other system integrators that plan to build commercial HPC systems around the MIC silicon.

It puts Nvidia in a tricky place. On one hand Intel's Knights corner will give its Tesla a run for its money and on the other integrated processor graphics (IPGs) like Sandy Bridge and AMD's Llano are carving up the discrete GPU market.

Intel claims that MIC is better than Tesla because its x86 cores is easy for users to port their existing toolchains to it.  If you want to use Tesla you have to use Nvidia's proprietary CUDA platform.

So far no one in Intel has been talking about the performance differences between the two technologies. Jon Stokes at Ars Technica pointed out that Intel is not changing its interconnect architecture for Knight's Ferry.

This means that the beast will have 50 cores hanging  off a single, high-performance yet power-hungry ring bus. We will probably know more details soon, but if that is the case, Nvidia can argue that Tesla gives a higher performance with a lower power draw, which is exactly what a database manager wants to hear.

Rate this item
(12 votes)

Read more about: