Intel Unlikely to Release Larrabee As Discrete Graphics Processor - Company

From X-bit Labs: Intel Corp. is unlikely to release a discrete graphics processor based on the Larrabee architecture, said a director of graphics architecture at Intel architecture group. Moreover, the executive admitted that Larrabee had a number of architectural caveats that eventually made it inefficient in terms of performance and consumed power.

"I just think it is impractical to try to do all the functions in software in view of all the software complexity. And we ran into a performance per watt issue trying to do these things," said Thomas Piazza, the director of graphics architecture development at Intel architecture group, at an event during Intel Developer Forum, reports Techradar web-site.

The Larrabee graphics processor featured some special-function graphics hardware inside and was more flexible in terms of programmability than current graphics chips from ATI, graphics business unit of Advanced Micro Devices, and Nvidia Corp. Apparently, Intel failed to find the right balance between fixed-function and programmable blocks inside its graphics processing unit (GPU); for instance, the company tried to implement programmable rasterization (possibly in order to encourage various approaches to programmable graphics), but it turned out to be very inefficient.

"[We were trying to find] what's the right level of programmability and what's the right level of fixed function. [...] Naturally a rasterizer wants to be fixed function. There is no reason to have the programming; it takes so little area for what it does relative to trying to code things like that," added Mr. Piazza.

Earlier this year Intel released its Knights Ferry development board that presumably features the chip that was supposed to become the first Larrabee GPU. The board is aimed at software developers trying to utilize the power of many Intel cores (MIC) architecture in high-performance computing and similar applications. The Knights Ferry has 32 x86 cores clocked at 1.2GHz and featuring quad-HyperThreading. The unit, aimed at PCI Express 2.0 slots, has up to 2GB of GDDR5 memory. The chip itself has 8MB of shared L2 cache, which is quite intriguing by itself since highly-parallel applications do not require a large on-chip cache nor do graphics applications.

View: Article @ Source Site