by Anton Shilov
12/08/2009 | 12:01 PM
Intel Corp. admitted rather clearly that its code-named Larrabee graphics processing unit (GPU) and supporting drivers were much behind the schedule and could not be released as a successful commercial product. What Intel has not admitted is that the postponement of Larrabee may force it to re-invent its roadmap quite considerably.
Even though Intel has stressed many times that Larrabee was a many-core processor suitable for conventional graphics processing, ray tracing, general purpose computing on graphics processing unit (GPGPU) and many other useful things, the Larrabee has always been Intel’s next-generation graphics solution – for both discrete and integrated markets – with additional capabilities.
Without Larrabee in 2010, Intel will not only be able to address the lucrative market of advanced standalone GPUs, but will also not be able to address the growing market of GPU-based accelerators for high-performance computing (HPC), will be unable to offer advantages provided by GPGPU applications for consumers and will not be able to offer anything more powerful than its integrated graphics cores.
Without Larrabee, Intel will have to dramatically improve performance of its integrated graphics and media accelerator (GMA) cores in the short term and ensure it has its own GPU in the longer term future. Perhaps, elements of Larrabee will find their place inside GMA cores rather sooner than later.
Intel has already announced that its next-generation Arrandale and Clarkdale central processing units with integrated graphics cores will support hardware-accelerated video transcoding on graphics core and it is likely to be only the beginning. The company will need to support DirectX 11 along with Direct Compute and OpenCL going forward in order to stay competitive with AMD’s and Nvidia’s integrated platforms in 2010 and then with AMD’s code-named Llano CPU with integrated DirectX 11 GPU in 2011. Moreover, neither ATI, graphics business unit of AMD, nor Nvidia will stand still with their GPUs, which will provide even more benefits. All-in-all, the gap between GMA and GeForce/Radeon graphics processors will be getting even wider in the coming quarters and Intel will have to make its integrated graphics substantially more powerful both in general and relative terms. Perhaps, the company will even license as many third-party technologies as possible and also hire additional software teams to boost graphics performance.
Intel’s business prospers even discrete without graphics processors today. However, we do know that Larrabee was supposed to find itself inside central processing units sometimes in 2012 – 2013 timeframe. As a consequence of Larrabee delay, Intel may have to reconsider graphics cores inside processors like Sandy Bridge and Ivy Bridge – which were supposed to feature GMA – and, even more importantly, inside code-named Haswell processor – which was rumoured to feature built-in Larrabee. Even though Intel Xeon processors powered by Sandy Bridge or Ivy Bridge processors were not supposed to use stream processors for HPC applications, Haswell could well use Larrabee’s wide SIMD vector processor unit to accelerate data processing. Intel is likely to do its best to integrate a new-generation graphics core into its 2013 microprocessor; however, in order to do that, it will have to get Larrabee out of the door in early 2012 at the latest, or redesign the Haswell processor/roadmap.
Back in the mid-2000s Intel did a great job by developing the Core 2 micro-architecture and appropriate processors in just about a couple of years. Intel spent three years creating Larrabee without any obvious results, but maybe the next two or three will be more productive?