Clock speed is only half the story

Penryn is the codename for Intel’s latest processor generation, the first based on the firm’s 45nm fabrication technology. This manufacturing process results in smaller chips that pack in more transistors, generate less heat and require less energy. The first Penryn chips are due today. These were formerly known by the codenames Harpertown, Yorkfield and Wolfdale, referring respectively to a quad-core Xeon for twin-socket servers; an Extreme Edition quad-core desktop processor; and a dual-core Xeon for twin-socket servers.

The new parts add architectural features, including SSE4 instructions primarily for media acceleration, database and data mining purposes; the Super Shuffle Engine and Fast Radix-16 divider for speeding certain complex calculations; and a Deep Power Down state for saving energy. Larger on-chip caches, support for higher bus speeds, and faster context switching in virtualisation round out Penryn’s capabilities.

A key figure behind these changes is Steve Fischer, Intel’s Penryn lead architect, who said that the first Penryn chips will debut at clock speeds of up to 3.2GHz.

“We’ve provided a mid-life kick in terms of clock rate and another step change in functionality that comes from microarchitecture enhancements,” he said. However, he did not commit to the 4GHz-plus clock speeds that Penryn is said to be capable of reaching.

The issue for chip developers today is whether cranking up clock speeds comes at too high a cost in terms of energy consumption.

“Clock speed is important but we’re definitely more sensitive to it if we’re consuming more power,” Fischer said. “We can’t think of it as we did with the Pentium 4: ‘If you want more performance, clock up the power’. There’s a vicious circle. 4GHz is not something we’ve committed to, but it’s something we keep an eye on and it depends on how aggressive we get with future tape-outs.”

Recently, Intel’s strategy has been to ratchet up performance while holding firm on existing power and thermal characteristics, but that could soon change. In laptops, for example, “equivalent performance at a lower power point” could be attractive, Fischer suggested.

Already a complex business, designing processor chips has become more difficult in recent years as a multiplicity of operating environments, virtualisation technology and graphics needs have changed the way systems are used.

“We’re going to create some interesting computing paradigms and I continue to be surprised by the usage models that take advantage of virtualisation,” Fischer said. “We’re trying to get 25 to 75 per cent faster [switching] between virtual machine and virtual machine. Virtualisation is always going to introduce a [performance] tax, so anything you can do is important, but it’s still in its infancy.”

As for graphics, bringing together teams within Intel has been “a challenge”, Fischer conceded, but perhaps with half an eye on rival AMD’s work with last year's acquisition of ATI , he added that Intel should benefit from the fact that both CPU and GPU groups “have been working at Intel for some time”.

45nmclockhalfhardwareisonlypenrynspeedstorythe