AMD is more pragmatic than Intel for teraflop systems
AMD's first teraflop performance server may come just week's after Intel's 80-core processor announcement, but AMD gains from its use of product silicon, the Opteron dual core CPU and the R600 Stream Processor.
Indeed, Intel said it had no plans to bring the 80-core chip to market, but it expected chips with 20 to 40 cores could hit the market within 10 years. But no one expects to wait that long for teraflop performance from a commercial CPU?
AMD has made no secret of its plans to use co-processors interoperating with x86 microprocessors to provide supercomputer performance for specific applications.
Does this approach make more sense than Intel's single CPU-centric approach to super server performance?
Intel may grab the headlines with a single processor made up from 80 cores, but perhaps AMD's more pragmatic approach gains from being closer to a real data processing application.
For AMD, accelerated computing begins with an x86-based processor to which is added a separate accelerator chip, the Stream Computing processor.
Clearly, further integration is likely, whioch will inevitably bring the approaches of Intel and AMD clsoer together. But right now AMD's accelerator approach is all about beating Intel to market with commercial products.
"Going forward, AMD will integrate at the device level with accelerated processors, and will integrate a CPU and GP called Fusion, focused on mobile applications," said AMD senior fellow and chief platform architect Steve Polzin.
The Torrenza Stream Computing processor is AMD's first effort in this area and is shipping now, he noted, which performs general-purpose computation on the graphics processing unit (GPU) and allows the floating point performance of the GPU to be an order of magnitude faster than the CPU. Changes to GPU architectures allowed this.
This has real commercial applications, one of which is running oil exploration software from PeakStream.
