The difference between [the P4] and the [Athlon] die size is frigging huge - AMD's Jerry Sanders III
CHIPMAKER Intel has hit out at closed standards in high performance computing (HPC), saying that it is a costly detour.
Kirk Skaugen, VP of Intel's architecture group, told journalists that Intel wants to "democratise highly parallel computing" and he was "urging the industry to avoid costly detours down proprietary paths". Talking up Intel's Many Integrated Core (MIC) architecture, which allows programmers to use languages such as C++ and Fortran to make use of its single instruction multiple data (SIMD) units, Skaugen criticised the use of proprietary, closed standards.
Skaugen said that firms do not want to "spend all the money to port [applications]" to specific hardware and that one of MIC's biggest selling points is the fact that it can run existing code and make use of SIMDs with relatively few code changes. Dr Eng Lim Goh, CTO of SGI backed up Skaugen's point, saying that "a large portion [of customers] are realising that this [moving code to the GPGPU] is tough", adding "they don't want to spend the same amount of money to port software as on hardware".
As for what exactly Skaugen was referring to when he said "proprietary paths", he did not say. However The INQUIRER asked Joseph Curley, director of MIC marketing at Intel to clarify whether Skaugen was referring to Nvidia. Curley would not identify Nvidia by name but said that any firm that only allows code to run on its own hardware can be classified as proprietary.
Curley explained that Intel wants to bring high level programming language portability to many core programming. That puts Intel at apparent loggerheads with OpenCL, a language driven by the Khronos consortium, of which Intel is a member. Curley said that "OpenCL is an exception" and it will continue promote OpenCL as it is an open standard that will allow code to run on multiple vendors' hardware, including AMD's GPGPUs and accelerated processing units (APUs).
Although Skaugen and Curley would not mention Nvidia and its CUDA programming language by name, it was pretty clear that Nvidia was the intended target. Curley did mention that other technologies such as FPGAs and DSP chips also fall into the "proprietary path", but pointed out that Intel has not been in competition with those chips for a very long time.
Nvidia, which has had a relatively easy ride in the HPC market with its GPGPUs, looks set to face not only AMD but Intel as a competitor to both. µ
Sign up for INQbot – a weekly roundup of the best from the INQ