When will multicore processing truly be mainstream and is it simply a software issue?

Last week one of my colleagues at NMI ran an event entitled Multi-core/Multi-processor systems - an event I took a personal interest in having cut my teeth on transputer technology way back in the 80's. As a freshly minted engineer I marvelled at how simple it was to create systems with scalable performance simply by designing code that could be partitioned based on good old CSP (communicating sequential processes). The default language was Occam of course - it was all very neat and efficient - IMO.

Inmos (the transputer manufacturers) found that they had to introduce parallel C to help adoption levels - a move I felt (as a young and somewhat naive engineer) had no major advantages... except being more attractive to the mass body of "C talking sequential coders" - tcchhh! I also recall at the same time an opinion from my boss (and engineering mentor) that "programmers are programmers - they just talk a different syntax"... a view I thought rational at the time - yet now I am not so sure.

If we roll the clock forward we find that the transputer was undoubtedly ahead of its time and only now that Intel have run out of Ghz are multiprocessor systems back on the table as a scalable solution. Intel still dominate the PC processor market and that puts them in the driving seat of course. This means that the hardware adoption issue is somewhat different from the past - multicore will break through. The processors may become multicore but you can still write your code in time honoured sequential fashion (and take the potential hit on system performance). Having recently looked at reviews on PC's recently it appears that many applications still resist taking advantage of multiple cores - its all a bit complex and technical. So we're back to the coding paradigm again - surely the industry is poised to see a shift in the way we write code?

Maybe not tomorrow but I reckon it is highly likely as Intel will want to encourage the industry to adopt the benefits of ever-better technology. I am also wondering whether it may also mark a change in the professional status of programmers; there was a time when you could just jump in and hack away - its not so simple any more.

However, some of those guys that architected the original transputer are back with a different slant on things - if you have not come across Xmos (a Bristol based start up) I suggest you look them up. They have applied multicore in a user friendly way to appeal to the mass of conventional software developers out there - and at consumer friendly price points too (another criticism of the transputer). We're watching their progress with great interest here at NMI - with all of the Silicon engineering talent in Bristol it is curious we have not yet seen a start up emerge on the scale of CSR.

Anyway... the transputer did make it into mass production eventually - it got stripped down a little and was reborn as an embedded processor core and used in apps such as set-top boxes.

I still have a few transputers in the cupboard for nostalgic reasons... and I am still in love with the concepts it was built upon 20 years on - just as multicore is set to be the next big thing (again).

What do you think?