For a lot of years now computer systems with a lot of processors, each often with their own cache, have used a common cache coherence protocol to allow all of them to access their shared memory and the contents of each others cache. Esoteric enough for you? What this has meant, though, is that all of the processors in such systems are essentially identical processors. High capacity systems can be built by having a lot of such homogenous processors, but it happens that there tends to be a limit in how large these system can get. More recently, specialized compute accelerators have been showing up on the market, with these producing outstanding performance in specialized situations, but being different, they have not been able to access the system's memory in the same way as the homogenous SMP's processors. Wanting the best of both words, a consortium of companies have formed the CCIX with the intent of creating still faster heterogeneous systems of both generic processors and accelerators. Outlining what they are up against is largely the purpose of two articles I've recently had published on TheNextPlatform ....
- Drilling Into The CCIX Coherence Standard ... (July 13, 2016)
- Weaving Accelerators Into The Memory Complex ... (July 14, 2016)