





To increase a chip’s performance, the industry engineers works with multi-core processors. Ericsson can easily take advantage of chips with few cores, but beyond that several challenges have to be addressed.
September 29, 2008

Five years ago there was just one core on a chip; now there are already chips with several 10s of cores on the market. And later there may be 1000-core chips. The development speed is comparable to the well-known Moore’s law, but there is no consensus in the industry on how to use these “beast” processors.
The technical background is that it is no longer possible to increase performance on a chip by increasing the frequency. It becomes too hot. But it can be done by putting more cores on one chip, all running on lower frequencies. Usually, up to 16 cores and it is still called multi-core chips; beyond that, usually the term many-core is used.
Andras Vajda, Senior Specialist at Ericsson Software Research, says: “The paradigm shift that the emergence of multi-core chips will bring does influence the way we will do software. Multi-core chips are no real problem for us, we have done parallel software for 30 years. But for 32, 64 or more cores, different problems occur, related to operating systems, scheduling and securing high availability and reliability.”
It is difficult to program these chips and, for instance, make sure that several cores can access the same data consistently while the system still maintains a high level of reliability. Without true parallelism in software there is no gain.
To meet this challenge the computer and telecom industries have cooperated. In the US, for example, Intel, Microsoft, IBM, SUN and others are investing USD 26m in research projects (at Berkeley, Stanford and the University of Illinois) for the coming five to six years.
Ericsson also co-operates with the academic institutions and supports the Swedish Multi-core Initiative and the Swedish Multi-core Research centre.