Over the past month or so a couple of article's have come to may attention that seem to indicate the beginning of a new computing architecture on several fronts.
Part of it is the failure of Intel's Itanium 64 bit chip set, which is not lighting the market a fire, while AMD's 32/64 Opteron chips doing very well in both performance and sales. Well enough for Intel to bit the bullet and introduce their own, Prescott, which has been plagued with problems. For a change Intel looks to be chasing AMD's tail for at least the next couple of years. A transition to 64 bit computing on Servers make a real difference in that it (at minimum) allows for addressing more that 4 Gb of ram. So we have begun this Transition to full 64 bit computer via the Hybrid of 32/64 where even the 32 bit programs and OS's are faster even before waiting for all your essential software to be re-written as 64 bit. And this transition will also occur (faster than they think) in desktop (if not laptop) in the workstation class machines (financial , engineering, and gamming situations where time is money).
But there are 2 other big sea changes going on in the non-Intel chip world.
One of these is the move to new multicore chips. This is more than just hyper-threading; this is two chips on one. The reason this is important is because the standard (old) way of doing Dual processor machines never actually got you as much as you thought it would. That second processor only got you 40% to 60% more bang for your buck, because of bottlenecks in inter chip communication. (and the more processors per machine the worse it got, with 4 way machines averaging only 20% improvement over the first processor) In a multicore chip with 2 cores (processors) on a chip it looks like a 90% to 110% increase over one core, and with more cores being additive in a similar fashion.
Sun Microsystems has cancelled its next planned processor, the UltraSparc V, in order to go with new multicore designs named Niagara and Rock. IBM is also adding multicore designs to it Power5 designs.
The other thing that opened my eyes was a series of articles on Linux Insider :
Apple: Up the Market Without a CPU
Grid vs. SMP: The Empire Tries Again
The Computing Landscape Has Changed
Fast, Faster and IBM's PlayStation 3 Processor
Linux on Intel: Think Dead Man Walking
The summary of this boils down to:
Sunís new chips are both multicore and multithreaded and continue their tradition of SMP (Symmetric Multi Processing). Sunís work promises major performance gains.
IBM is building a new architecture with Sony and Toshiba, called the Cell, which is based on a both multicore and multithreaded Grid-on-a-chip philosophy. The processor itself appears to be a PowerPC derivative with high-speed built-in local communications, high-speed access to local memory, and up to eight attached processing. The actual product consists of one to eight of these on a chip -- a true grid-on-a-chip approach in which a four-way assembly can, when fully populated, consist of four core CPUs, 32 attached processing units and 512 MB of local memory. The performance gains promise to be astounding. The initial use of the chip, at the low end, will be Sonyís PS3 (Play Station 3) at the end of 2005. This will also allow production scales that will (require) high volume fabrication and quickly drive down costs.
The articles suggest that Apple might have difficultly porting OSX to the new Cell platform, but can (more) easily move to Sunís new platform, and that IBM will quickly rollout laptop based on the Cell platform and Linux.
If any of these new chip deliver there performance gains, on schedule, Intel looks to be in trouble in the short term. I would also wonder where this leaves MS Windows and the WinTel monopoly.
But it does appear we have a major sea change on our hands.
(see also The Computing Landscape Has Changed By Bill Nicholls