Intel Tanglewood's first voyage
Sailed off only to find IBM, Sun already landed
Analysis The appearance this week of Intel's Tanglewood processor has added yet another high-end multicore chip onto the grand 64-bit roadmap and complicated the battle between Chipzilla, IBM and Sun.
We now know Intel will introduce its first attempt at a dual-core chip, called Montecito, in 2005 and then follow that with the more elegant Tanglewood, most likely in 2006 or 2007. Tanglewood should have more cores than Montecito and consume less power, making it a worthy competitor against IBM and Sun. The problem is that while Intel's high-end chips often run faster than the competition, they also tend to fall a few steps behind rivals with some of the more exciting technology in the 64-bit market.
IBM, for example, already cranks out the dual-core Power 4 chip en masse and has the multi-core, multi-threaded Power 5 chip coming next year. Power 5 is poised to handle mid-range and low-end applications well, especially with multi-threaded software running the various processor cores.
IBM also received millions from various institutions to build the Blue Gene supercomputer - a machine that could one day house thousands of low-power processors cores in a relatively small casing and still perform at the top of the high-performance computing ranks.
For its part, Sun will bring out the dual-core UltraSPARC IV chip this year. This will be followed by a multi-core UltraSPARC V in 2005.
Sun has been more vocal than IBM with "the other" part of its multi-core chip line known as the H-series. These designs will surround numerous low-power processor cores - namely lots of UltraSPARC IIs - with several gigabytes of memory. Sun believes that spreading out multi-threaded software across all of these cores will help keep them busy crunching data and make them less affected by lagging memory.
Sun's Marc Tremblay does a nice job of explaining the technology in a recent interview posted on Ace's Hardware . It should also be noted that the amiable Tremblay drives a dark blue Porsche with MAJC on the licence plate, showing his love for another Sun chip project and also the pleasant rewards of boom years now forgotten.
In an ideal world, Sun would roll out the first H-series product in early or even late 2004 and put some heat on the competition. Instead, it won't start shipping an H-series chip until 2005, meaning Intel and IBM will continue to use Sun as their poster child for a chip investment gone wrong. Even when the first H-series product, code-named Gemini, does appear, it will only place two UltraSPARC II cores on a single die. This is a far cry from the tens of cores we are promised down the line.
Still, in this befuddled 64-bit world, Sun is the only company vowing to offer various of multi-core processors to its customers.
Both IBM and Sun have the advantage of a hefty customer base tied to their Risc architectures and accompanying versions of Unix. They've presented their customers with a consistent and lengthy roadmap for these products and also drop hints with Blue Gene and the H-Series of interesting things to come.
HP, with whom Itanium's future lies, has a less pleasant story to tell its customers. Its tale is tainted by the murders of PA-Risc and Alpha, expensive software ports and a general hardware overhaul that already sends CIOs into paralytic shock. This situation is only worsened by Intel's slow move to dual-core technology, making HP's switch to Itanium seem more like a nightmare than a price/performance wet dream ®