Thoughts
Look around the web for pre-Nocona Xeon vs Opteron articles and you'll generally find just one style of conclusion; the Opteron is great, the Xeon less so. Therefore testing the Nocona Xeon, based as it is on the desktop Prescott core which affords it a faster system bus, new cache layout, SSE3, tweaked HyperThreading and access to DDR400 memory officially for the first time, was a means to see if the update was worth Intel's time and effort.And my opinion is yes, very much so. While it doesn't overtake and outrun the Opteron at any point in our benchmarks, there's seemingly a lot of improvement to be had from bumping up the new entry-level Xeon's basic specs to 800MHz bus with 1MB L2. 3400MHz doesn't do it any harm either.
Looking at older Xeon systems, the criticisms are generally levelled at its bus sharing system and the memory controller. Neither of those things change with Nocona and its host core logic. Rather the bus and memory controller just run faster, in tandem with the CPU's new speed increases. So while those things are still the main limiting factors in Xeon's performance, especially on the small cache models, turning everything up a notch is enough for a new turn of speed.
However, and it's a pretty big however, Opteron still remains the performance SMP platform of choice for professional applications such as content creation, 3D rendering and image analysis. The long held belief that you buy Xeon if you're doing professional media encoding is a fallacy. It's largely encoder-dependant. You can engineer it so a Xeon looks like it's the fastest in that field of application, but then you can also turn that on its head with a well chosen encoder suite, allowing the Opteron to show off.
And with more general purpose muscle than Xeon, the only reason you wouldn't buy it is if your most widely used application performs better on Xeon. And those applications seem increasingly few and far between.
As for Tumwater, it's the first generally available means to dual PEG. However that's really the only thing that makes it stand out. It's still paired with ICH5/R. AMD systems will have dual PEG slots in due course and there's simply nothing to put in the second slot just yet, with NVIDIA yet to release any official SLI implementation in terms of supporting hardware and a driver.
To sum up, small-cache Xeon just got usefully faster, but Opteron still easily holds the lead.
Many thanks to Boston and AMD with their help in putting the test systems together.
