by Anton Shilov
11/07/2006 | 08:13 AM
Phil Hester, chief technology officer at Advanced Micro Devices, the world’s second largest maker of central processing units, said at a conference that the integration of graphics processing units (GPUs) into central processing units will allow personal computers to achieve performance of supercomputers eventually.
<%BANNER[article]%>“Get ready for round two of the "attack of the killer micros. By combining graphics processing unit (GPU) and CPU functions in heterogeneous cores, microprocessors will bring supercomputer performance to the desktop,” said Phil Hester, in a keynote speech at the International Conference on Computer-Aided Design (ICCAD) in San Jose, California, reports EETimes web-site.
The chief technologist at AMD believes that in order to achieve tremendous computing power on the desktop, central processing units (CPUs) should start utilizing heterogeneous multi-core design, where each of the cores will be able to perform certain types of tasks very rapidly. Given that theoretical peak power of modern GPUs is much higher than that of CPUs, it is natural to built in GPUs into CPUs to increase performance.
“A step increase in microprocessor performance per watt per dollar is needed. But simply adding more homogeneous CPU cores to a baseline architecture is not good enough. The solution is to adopt a heterogeneous architecture with GPU/CPU silicon-level integration,” Mr. Hester is reported to have said.
Mr. Hester also called integration of graphics processing engines into AMD’s chips as the “biggest microprocessor evolution” since the introduction of x86-64 concept back in 1999. Advanced Micro Devices proposed 64-bit extensions to x86 architecture seven years ago and has managed to transform the x86-64 technology into an industrial standard since then, which stresses how significant the idea to combine CPU and GPU is.
However, according to Mr. Hester, there are two significant design challenges in developing heterogeneous architectures (that combine CPU and GPU) – power management and memory hierarchy.