Intel is the name on everyone's lips when it comes to exascale computing research - thanks largely to efforts like the Many Integrated Cores project and the resultant Larrabee-based 50-core accelerator cards - but AMD has declared its intention to bring its own expertise to bear on the problem - with a little help from $12.6 million of Department of Energy cash.

The US DoE is one of the biggest consumers of computing power in the world. As well as its work on civilian-friendly projects like nuclear power, it also has a major hand in some seriously civilian-unfriendly technology - like nuclear bombs.

Designing that sort of thing takes a lot of processing power for simulations and virtual experiments - thanks largely to international regulations which prohibit above-ground testing of nuclear munitions, preventing the DoE from just doing another Bikini Atoll or twelve - so the organisation has something of a vested interest in making sure that Moore's Law stays alive and well for the next few decades.

Chip giant Intel has been talking up its exascale - quintillion floating point operations per second (exaflop) computing - chops of late: a low-power chip running between 3MHz and 915MHz demonstrates the kind of efficiency improvement needed for the leap to exascale, while purchases of interconnect technology from Cray and QLogic's InfiniBand business unit indicate a desire to make a bigger name for itself in the supercomputing and high-performance computing (HPC) market.

AMD, by contrast, has been surprisingly quiet on this front - until now. Announced late yesterday, the partnership with the DoE Office of Science and the National Nuclear Security Administration (NNSA) will see AMD given $9.6 million in funds for processor-related research and a further $3 million for memory under the FastForward research and development initiative.

'This award from the DOE will fund critical research and development required to enable high-performance, power-efficient exascale systems,' claimed Alan Lee, AMD's corporate vice president of research and advanced development, of the funding. 'Additionally, AMD will undertake work to drive advances in memory bandwidth and communication speed, which are essential for heterogeneous architecture, exascale-class supercomputers with thousands of processors.'

What begins in the datacentre always trickles down to the desktop, so fingers crossed for some serious breakthroughs in efficiency and performance that could bring AMD back to where it needs to be in its fight against bitter rival Intel.


View this in the forums