Now the dust has settled following the launch of Intel's new 64-bit Xeon, code-named Nocona, it's time to take stock of what this chip actually offers, and how well it compares to AMD's 18-month-old Opteron in the dual-CPU server market.
Performance improvements have to be Nocona's main benefit yet, curiously, Intel's marketing machine has made little of this. Instead, it's chosen to focus on additional features such as its ability to address a larger amount of memory - up from a measly 4GB (an amount seemingly limitless at the time of the introduction of 32-bit-ness in 1986) to 16GB. It also allows access to faster bus and memory technologies - get more detail both in our news story about other benefits offered by the Xeon with EM64 (Intel's title for the 64-bit extensions), and on Intel's site.
Benchmarks have been run - see here and here - and the results show that Nocona is clearly faster by some 20-odd per cent, as you'd expect, than the older Xeon, when running 64-bit software. It's also quicker than AMD's Opteron. This is all much as you'd expect, given that Nocona uses newer technology.
Nocona uses a 90nm process and, as ever when a new, smaller geometry is introduced, the chip designer can both shoehorn more transistors into the same space and drive the chip at a lower voltage. This then helps to offset the heat and power draw associated with higher clock speeds. What's more, Intel has exploited the additional on-chip space by raising the Level 2 cache size to 1MB from the Xeon DP’s 512KB. And by bumping up the size of the Level 2 cache, Intel can remove the Level 3 cache in late-model Xeon DP processors. And it's the doubling of Nocona’s Level 2 cache - which runs at the CPU’s full clock speed -- that's responsible for the performance boost when running Xeon-optimised applications.
However, it's not all roses. The initial Nocona release doesn't include support for the NX (no execute) security feature introduced in Windows XP's SP2. That arrives in September. NX allows areas of memory, such as the default heap, various stacks, and memory pools, to be flagged as data-only, prohibiting code execution there. Designed to prevent viruses and worms gaining back-door entry to computers, AMD's latest Athlons - available now - already support this feature.
Additionally, the AMD chip needs no additional glue logic to get the CPUs talking, unlike Nocona. Also unlike Nocona, Opteron both talks directly to memory without intermediary silicon, and can scale its memory bandwidth upwards with the number of CPUs.
What's more, as a 32-bit x86 processor, Nocona instantly renders its predecessor, the 130nm Xeon DP, obsolete.
AMD trumps Intel?
So it's arguable whether Intel has actually managed to steal a march on AMD. On the contrary, it looks more as if the Santa Clara giant has been rushed into responding to AMD's Opteron release. It even looks as if the company is running a little scared of its relatively tiny rival. For example, its marketing materials for the new Xeon used benchmarks that compared it to the older Xeon rather than processors from its rivals. For Intel, that's a departure from its established practice.
So on the face of it, Intel has managed a stealth product launch which, unlike most, doesn't include a comparison with its chief rival, and whose subject, while faster in raw terms, is behind in terms of functionality and features. Given AMD's lead in the dual processor systems market, it's probable that the older Xeon will be the biggest casualty following Nocona's launch, with Opterons remaining ahead.
Zoom out to the bigger picture, however, and Intel canters well ahead of AMD and looks set to remain there - if a little more nervously.