by Anton Shilov
12/09/2009 | 04:35 PM
Although a number of analysts welcome Intel Corp.’s decision to cancel the half-baked first-generation Larrabee graphics processing unit (GPU), they believe that the world’s largest maker of semiconductors will continue to develop its own graphics processor both for consumers and emerging market of GPU-based high-performance computing (HPC).
“Intel has made a hard decision and we think a correct one. […] We think this makes a lot of sense, and leaves the door open for Intel to take a second run at the graphics processor market. The nexus of compute and visualization […] is clearly upon us, and it is too big and too important for Intel not to participate in all aspects of it,” said Jon Peddie, the head of Jon Peddie Research.
According to Mr. Peddie, Larrabee silicon was pretty much proven, and the demonstration at SC09 of measured performance hitting 1TFLOPS (probably, double precision performance) on a SGEMM Performance test (4K by 4K Matrix Multiply), which was a computing performance measurement, not a graphics performance measurement. With 1TFLOPS double precision performance, Larrabee could rival both ATI and Nvidia latest and forthcoming graphics products in HPC applications, however, it is unclear whether the chip itself was free of bugs and whether drivers could deliver adequate performance in today’s and tomorrow’s video games.
"I am happier to see them recognize a problem rather than ship something that's subpar. It's incredibly hard to get a product right the first time, especially when you are doing something very new such a Larrabee," said Nathan Brookwood, the principal of Insight 64 market research firm.
Perhaps, not releasing subpar products for clients is a good idea. However, revealing that a couple of months before schedulled launch hardly is, another analyst believes.
“If they were having problems, why didn't that take the foot off the gas at the last IDF? It is disheartening when any company changes or cancels an effort that has had 30 months of disclosure. It makes you wonder how much of the rest of the road map is dependable,” said Richard Doherty, principal of consulting firm Envisioneering.
Mr. Brookwood also expects Intel to continue developing Larrabee project and its own graphics chips, which is a threat to ATI, graphics business unit of Advanced Micro Devices, and Nvidia. However, since there will be no many-core x86-based Larrabee in the next couple of years on the market, more and more software developers will have to actually take advantage of many-core GPUs from ATI and Nvidia.
“If you are at Nvidia or AMD you can breathe a sigh of relief. Over the next two years while Intel is re-architecting Larrabee, AMD and Nvidia will ship hundreds of millions of GPUs capable of parallel processing with OpenCL and DirectCompute. More and more we'll see application developers use those tools, and they won't even know whether the work is being done on an x86 or GPU core,” said Mr. Brookwood.
Intel clearly understands that the window of opportunity for x86 GPU is somewhat limited, which is why, according to Jim McGregor, chief technology strategist at In-Stat, Intel will still make the current silicon available as a graphics software development vehicle (SDV), and will also promote and support Larrabee as a SDV for a future co-processor/accelerator for high-performance computing (HPC) applications.
“Developing a new hardware architecture is difficult enough, but developing a new programming model is exponentially more challenging because it requires support from the rest of the industry. History in the electronics industry suggests that few new technologies meet their initial schedules and adoption of new technologies and methodologies takes two to three times longer than anticipated,” said Jim McGregor.
Intel has indicated that the setback will not have any financial or personnel impact on its investment in developing graphics technology. With Intel’s growing focus on consumer electronics, the company cannot consider ending graphics development, because graphics is a critical technology to all the major consumer and computing platforms.
Despite the efforts, however, In-Stat feels that this is still a major blow to Intel as the company looked to distance itself from both ATI/AMD and Nvidia with whom it shares partner-competitor (or love-hate) relationships. The plan was to enter the discrete GPU market on the high-end and migrate the technology down to the integrated graphics solutions. Now, Intel will have to rely on the older graphics architecture for integrated solutions while it regroups and re-enforces the Larrabee project. The strategy shift will, however, provide Intel’s rivals more breathing room in the higher margin discrete GPU space.
In-Stat does not believe Intel will give up on developing graphics technology because this function is too critical as the world moves to heterogeneous computing solutions, something that In-Stat indicated in 2005 that AMD and Intel would both have to do by the end of the decade.