AMD Opteron tiptoes into Intel server space
Server Briefing AMD's recent admission that the performance of Opteron, its upcoming server-oriented processor, formerly codenamed Sledgehammer, isn't up to snuff will have pleased Intel executives. Opteron marks AMD's attempt to go beyond the 32-bit PC market and into the more lucrative 64-bit server and scientific computing arenas.
Quite apart from the effect Opteron might have on the 64-bit processor market, Intel's own contender, Itanium, has had its share of performance trouble. However, the poorly received first version of the chip, codenamed Merced, will soon be replaced by a second-generation Itanium, McKinley. Intel is already gearing up its marketing machine: a few days before the Opteron performance revelation, Intel claimed McKinley-based systems will out-perform rival 64-bit kit by 50-100 per cent.
Intel's definition of 'rival' is product from IBM and Sun. Intel's other 64-bit competitor, Hewlett-Packard, is committed to replace its own PA-Risc chips and the Alpha CPU acquired with the Compaq takeover with Itanium. Intel's performance claims were made against Sun's UltraSparc III processor.
Itanium 2, as McKinley is officially named, has had a long gestation, growing out of development work at HP (hence its support). It's a totally new architecture, built specifically for the 64-bit world, with its requirements to access vast amounts of memory (40 billion times more than a 32-bit CPU can) and crunch bigger, more precise numbers. Itanium's clock speeds are low compared to the Pentium 4, but its EPIC (Explicitly Parallel Instruction Computing) design allows it to do far more work per cycle. No wonder it has taken Intel over five years to get it right. But Itanium 2 looks to be the first version of the chip that can be offered commercially. It will go on sale mid-2002.
AMD has had to take a less monolithic approach. Opteron extends AMD's existing 32-bit x86 processors with new 64-bit registers and data pathways. Its goal is to build a next-generation desktop processor that happens to offer the power high-end computer users need too. This 'one size fits all' approach is borne out of AMD's limited resources - it simply can't afford to develop and produce a dedicated chip for servers and scientific workstations.
The trouble is, the 64-bit market has always seen itself as very distinct from the low-end 32-bit arena. In that respect, vendors and buyers may distrust AMD's 'super x86' approach. Intel also touts 32-bit backward compatibility, but it knows that few Itanium buyers are likely to want to run old 32-bit code, and has placed less emphasis on it. Support for 32-bit apps is there, if users need it, but they probably won't. Unlike Opteron, Itanium isn't intended to replace 32-bit processors.
Opteron, however, has two powerful weapons. First, Microsoft's endorsement - 64-bit Windows will be optimised for Opteron. It'll be optimised for Itanium too, but Microsoft's support makes AMD a player.
More importantly, though, will be AMD's pricing strategy. With Opteron targeted at desktop PCs as well as eight-way servers, AMD needs to get the chip out of the door at a very low price. Just as the Athlon MP has a premium over the Athlon XP, so too will server Opteron be more expensive than desktop versions.
The difference will be bigger this time round, as AMD has far more space to move in - 64-bit CPUs aren't cheap - but it's still likely to compete on price, just as it did when it muscled in on Intel's 32-bit dominance. That will put a lot of pressure on the incumbent 64-bit players, who are going to have a hard enough time competing with Intel.
As for performance, high-end Opterons aren't due until early 2003 - plenty of time to get that right. ®