Apple multicore future comes sooner rather than later

19 November 2008

When Apple's director of the Unix technology group Jordan Hubbard spoke at the LISA 08 conference for big-iron server admins, he flashed up a slide that indicated Snow Leopard is coming earlier rather than later. The result, naturally, is plenty of speculation that release 10.6 of Mac OS X might make an appearance at Macworld in January 2009.

There are some clues pointing to an earlier rather than later release. One is the 25th anniversary of the Macintosh itself. It seems unlikely that Steve Jobs is going to get on-stage at Macworld and talk about an iPhone with more flash memory or iPods in dayglo orange and green. Mac Pros, at the very least, with Intel's Nehalem processors together with faster GPUs and probably multiple GPUs seem likely.

Then there are the machines that won't get a refresh in January: the MacBook and 15in MacBook Pro notebook machines. The Pro machines have two graphics processor units in them which, right now, cannot be used together.

As Snow Leopard is the release that is meant to make the GPUs accessible to regular programmers and make better use of all the processor cores in a machine, I think it's a fairly safe bet that Apple will want to bring the software in for those machines sooner rather than later.

And then you have the timeline in Hubbard's table. The slides are available from Usenix, and were noticed by MacRumors which flagged up the content.

The slide shows 10.6 (with the Snow Leopard portion showing as black on black for some reason) going live in the first quarter of next year, 14 months or more after Leopard first shipped in late October 2007.

After apparently talking about security improvements in Leopard, Hubbard then turned to the multiprocessor future of OS X. Or, as the presentation puts it, "Our scary future" complete with the image of an asteroid hurtling to Earth, followed by a cheery T800 Terminator, indicating that Shirley from Garbage is going to be building her empire on ATI and nVidia chips.

The next part is about the rise of the GPU - "GPUs are becoming insanely fast and capable" and "OpenCL is an important development in this space". This is point at which nVidia takes a back seat as Hubbard's slides claim: "Convergence with CPUs is not that far." This is followed by "The future: Intel".

Rather than assume a combined Intel and nVidia roadmap, Hubbard concentrated on the plans for Intel's 'GPU' Larrabee. As I've argued before, this is unlikely to be any great shakes as a GPU in its first incarnation, but is CPU-like enough make programmers take notice. Hubbard is looking to a point in 2010 when, with successive die shrinks, a basic Intel processor could deploy more than 32 reasonably powerful cores on top of the standard x86 CPUs with potentially many more being in the system by 2015.

This, argues Hubbard, will radically change PC design. It will not be possible to have one shared memory space. They will have to be split up, which makes life easier for the hardware designers but makes life tougher for software developers. On the basis that the kernel will need to mediate how applications run across different memory spaces and activate cores at the right time, Hubbard said new programming interfaces and techniques will be needed. What isn't clear is whether this is what developers will find in Grand Central once Snow Leopard appears, or whether there is a second wave of technology that Apple intends to deploy.