Intel Tigerton - first look at current and upcoming enterprise server propositions

Thoughts, HEXUS.right2reply and related reading

There is no denying that the Caneland platform, with its Tigerton processors, packs an incredible amount of computing power into one box. Where we do have our doubts however is in how well equipped Intel's ageing front-side bus architecture is to utilise it. This is a technology that dates back to the Pentium Pro - first introduced in 1995 - as a way of providing relatively cheap multi-processor systems to compete with the RISC based machines of the day.

The 'Clarksboro' 7300 chipset, while no doubt a big advance over the E8501, positively reeks of the sweat from a team of chipset engineers trying desperately to find ways to grapple with the front-side busses limitations. But implementing four independent busses is not an elegant fix and neither is a 64MiB Snoop Filter. Combined, these workarounds push up the chipsets pin-count, die-size and power requirements in a seemly desperate bid to mask the interconnects inherent flaws. It appears little more than an expensive way to save face in a niche market segment.

Quite frankly Intel needs CSI equipped platforms, and in the four socket space they needed them a long time ago.

That's not to say it's all bad news. The Tigerton can almost match the Clovertown in absolute clock-speeds, and if your workload is sufficiently threaded it should be Intel's fastest performing solution simply by virtue of having twice the number of cores.

Applications such as POV-Ray, which scale relatively well with additional cores allow it to win benchmarks by wide margins. The worst case scenario appears to be if you are simply I/O limited you'll see little or no improvement compared to Intel's two-socket, quad-core configuration at the same clock-speed. Here more than anywhere however, the big threat will be Barcelona.

Our final gripe would be an issue which Intel seems to carry across its platforms - particularly in the server space - and which AMD has been focusing on with Barcelona.

For existing multi-processor Xeon users there is no drop-in upgrade path to the Tigerton processors. No previous Socket 604 chipset supports its 1066MHz front-side bus, nor the four independent busses. It's a complete new box or make do with your existing Paxville MP or Tulsa based system.

It has taken the best part of a year for Intel to bring a four-socket, quad-core, Core Microarchitecture based product to market when really this should have been the product that launched last August, at least in dual-core form - not Tulsa.

So when Intel potentially begins to introduce octo-core Nehalem architecture based parts, where is the upgrade path for a corporation looking to double the computational density of their year old four-socket boxes? While Caneland is without a doubt the fastest Intel platform out there at present for highly multi-threaded workloads it seems very unlikely it will ever receive octo-core parts. Penryn based quad-core parts are expected next year however.

Overall therefore, it would be our advice for system administrators to consider very carefully the kind of workloads such a system would be used for. It is really worth investigating whether you may be better served by, for example, AMD's Barcelona based Opterons, or multiple Clovertown based systems, and to remember that the product you purchase now will, in terms of processors, most likely stay constant until the server is retired.

HEXUS Right2Reply

At HEXUS.net, we invite the companies whose products we test to comment on our articles. If representatives from Intel choose to respond, we'll publish their commentary here verbatim.

HEXUS related reading