• <More on Intel.com

The Single-chip Cloud Computer

We are sorry, This PDF is available in download format only

Intel’s presentation of the 80-Core Tera-Scale Research Processor at ISSCC 2007 should have prepared us for the Single-Chip Cloud Computer, but it didn’t. Intel’s TRP—the Teraflops Research Processor, AKA Tera-Scale Research Processor—was built by the company for use by hardware and software designers as a learning chip. The PE, its processing engine, was a relatively simple high-speed dual-SIMD floating-point CPU. It was easy to envision this CPU’s integration along with the other 79 identical cores in the company’s experimental chip, because multiple-core chips employing simple CPUs were already in the market.

But with no more than four, or soon—six, of the pres¬ent complex cores implementing Intel’s Architecture (IA) in today’s PCs on one hand, and the simple CPUs employed in existing multiple-core chips on the other, it was difficult to imagine an on-chip integration of 48 dual integer supersca¬lar floating point Pentium5s, or more precisely P54Cs—the processors that were almost the equivalent to 96 486 CPUs. Almost, since the P54C’s two superscalar pipes are not iden¬tical: the V-pipe has limited functionality; only the U-pipe can execute all integer functions.

Considering the achievement in integration, it became even more interesting to understand the reasons why the SCC, the Single-chip Cloud Computer, should be intro¬duced by Intel, in December 2009, as yet another experi¬mental configuration.

Read the full The Single-chip Cloud Computer Article.