AMD Boosts Dual-Core Opteron Past 3.0 GHz: Will It Be Enough?

Yesterday's launch by AMD of dual-core Opteron processors at the 3.0 GHz and 3.2 GHz clock speeds substantiates what company officials had been saying at an analyst day meeting the week before: Maybe clock speed no longer matters in the consumer desktop space, but it's still a factor in the IT department.

Last year at this time, AMD processors were taking a pounding from newly introduced Intel Core 2 Duos. Since then, there have been two successive upgrade generations to Intel's new Core Microarchitecture, and it's already well on its way to replacing it with a design that eliminates its dependence upon an external memory controller...like AMD's had for a few years already.

Late last year, AMD lost market share in the critical server CPU space to Intel. That hemorrhaging may have subsided early this year. Still, AMD must continue to be seen as competitive in this higher-margin space - its new 3.2 GHz 8224 SE for eight-way systems sells for $2,149. To substantiate that margin, AMD needs to turn up the heat - using whatever dials it has at its disposal. So it's back to clock speed.

But will that be enough? Yesterday, AnandTech's Johan De Gelas presented the first early tests of new AMD dual-core Opterons matched against similarly clocked, dual-core and quad-core Intel Xeon MP processors. Opteron's quad-cores are on their way, though the 3.0 GHz speeds and higher won't be the first off the block.

The initial verdict is mixed - although some might say hardware reviewers' initial verdicts are always mixed, if only to stay on the safe side. But De Gelas' findings do point to some interesting possibilities for future Opteron performance, perhaps validating some of AMD's arguments about 1) its microarchitecture performance paying off in the long run; 2) today's benchmarks being insufficient for gauging modern-day performance.

In a number-crunching test, a 3 GHz Opteron 2224 SE-based system (standard performance, priced at $873) outperformed a similarly equipped 3 GHz Xeon MP-based system by 5% overall. We could stop there, and perhaps AMD might be happy if we did.

But AnandTech found that a 2.33 GHz Xeon 5345 DP-based server outperformed the same 2224 SE by 19.2%. The reason - though the review could have made this just a little bit clearer - is faster memory access, even though the DP is an older product line. The faster front-side bus on the 5345 (1333 MHz) enables it to outperform even newer MP-based CPUs at 1066 MHz in some respects.

So why can't AMD make a similar tweak? If you'll recall, it doesn't have a front-side bus - its memory controller is on the CPU, not separate. That's not at all a negative for AMD, as De Gelas' numbers indicate. The absence of an FSB increases the Opteron's overall memory bandwidth, as much as 46% over comparable Xeon MPs, enabling it to read from and write to memory significantly faster. Thus the Opteron's comparatively smaller L2 memory cache doesn't work against it; the Xeons actually need the higher cache to compensate. In short, latency is much less of a problem with Opteron than for Xeon.

It will therefore be most intriguing to see what AMD can accomplish with a quad-core, Barcelona-based design at 3.0 GHz and above. On some tests, however, the gain won't be felt as much, as De Gelas determined many benchmarks based on everyday utilities such as WinRAR don't take into account processor scalability after the second core, or NUMA memory architectures such as AMD's (since WinRAR apparently has yet to be compiled for NUMA). This validates some of what AMD's spokespersons have been saying about performance gains that customers will feel even if the testers have yet to actually see it.