Intel's mysterious 'Yamhill' technology - the chip giant's answer to AMD's AMD64 architecture - is a runner, but won't appear before 2005, a report from investment research organisation American Technology Research (ATR) has claimed.
Rumours that Intel is working on Yamhill, a set of 64-bit extensions to the 32-bit x86 instruction set, have abounded for as long as AMD has been known to be working on similar technology, if not longer.
AMD's version, originally dubbed x86-64 but now known as AMD64, finally made it to market in April when the company shipped its first Opteron processors for workstations and servers. The technology will make it into the mainstream in September when the consumer-oriented Athlon 64 ships.
AMD64 allows Athlon 64 and Opteron chips full backward compatibility with 32-bit operating systems and applications. Essentially, if you run a 32-bit OS, the chip operates just like a traditional Athlon but faster. Run a 64-bit OS, and you have access to 32-bit apps and 64-bit software.
64-bit computing has traditionally been the province of specific high-end applications, such as database management and scientific modeling. AMD64 allows AMD to tap into those markets with the same chip family that its uses to target more mainstream applications that are work perfectly well in the 32-bit domain. We can see some consumer apps taking advantage of the memory space made possible by 64-bit pointers to stretch their muscles - games, for example, and video editing - but most consumer software simply won't need it.
But that doesn't matter. AMD64 is more about extending the market reach of AMD's processors than dragging mainstream into the 64-bit domain. It may yet have that effect, but not for quite some time. It needs 64-bit desktop software to be widely available, and that's not going to happen until 64-bit desktop processors are widely available. Athlon 64 will set the ball rolling, but we're not yet convinced that it's going to grow AMD's market share dramatically enough to create a 64-bit mass market.
However, a similar technology from Intel would accelerate the process, if only because of the weight of the company's marketing machine. Where AMD largely has to wait for market momentum to build up slowly behind its 64-bit technology, Intel has the power to get the market moving itself.
Yamhill is how it may do so. ATR reckons it will ultimately do so, and "eventually supply a 64-bit extension of its P4 [and] P5 [aka Prescott]", according to the ATR report, cited by Semiconductor Business News.
"Any 64-bit variant of P5 [is] not in evidence until 2005," the ATR report says, Yamhill "would seem to present the only real challenge to AMD's Athlon 64, which should more or less enjoy the power desktop by itself in H1 2004." Meanwhile, "Opteron, which should be ramping aggressively in late Q4 2003, should have the low-end server market more or less to itself throughout 2004", the report adds.
In short, the P5 will come, and it will stay 32-bit for the time being. Reports on Yamhill from early 2002 suggest it would be incorporated into Prescott, but turned off. Should a market for 64-bit computing on the desktop emerge, Intel can then tweak the die and enable that extra circuitry. Back in early 2002, there seemed much less need to: AMD's 64-bit chips kept being put back, and IBM had yet to make its work on the 64-bit PowerPC 970 - what Apple calls the G5 - public.
But both processors - and AMD more than IBM, thanks to its focus on the x86 world - really only challenge Prescott at the very high end, where the desktop market blurs into the workstation business. Arguably these processors are more of a threat to Intel's Xeon. While the case for 64-bit in the consumer space has yet to be made, there is a need for it in the workstation and low-end server arenas.
Opteron is far more competitively priced than equivalent Xeons. IBM is putting the 970 into blades; Apple is putting it into workstations/pro desktops. In that respect, Yamhill seems far more likely to appear in Xeon chips before Intel activates it on the desktop. But if Intel is holding Yamhill off until 2005, it clearly believes its challengers won't pose enough of a threat to Pentium and Xeon before then. If so, AMD had better not waste the time Intel has granted it. ®