As a result, the time has come for us to turn our attention once again to server/workstation-class processors and see how these latest offerings compare. To that end, we've employed a revised set of extensively multithreaded benchmarks and new methods for testing energy efficiency. What we've found is excellence on several fronts from both camps, in a market that is more tightly contested than ever. Read on for our results and analysis.
Genesis of a comparo
This article grew out of several different sets of plans, and fortunately, those things have at last converged in a pleasant manner. When AMD's Socket F Opterons debuted, we asked AMD for some of these new chips for review. They responded by issuing a sort of power-performance challenge: they would be willing to loan us a pair of servers, one based on the new Opterons and another on Intel Woodcrest Xeons, if we would be willing to test them head to head for both performance and power efficiency. The servers were to come from an actual server OEM and be identically configured with the necessary exceptions of the CPUs, motherboard, and memory.
I believe this offer was prompted by AMD's frustration with seeing IT-oriented publications test systems with widely varying configurations against one another for performance and power efficiency. That is, of course, not how we operate here at TR, so we were more than willing to conduct an apples-to-apples test. We accepted AMD's challenge, and they shipped us a pair of systems from server vendor Colfax International.
We inspected the boxes to confirm that they were, indeed, identically configured. Happily, we didn't find any grand incongruities between the configurations. Both boxes were equipped with dual processors, 8GB of RAM in eight DIMMs, a pair of WD Caviar RE2 hard drives in a RAID 0 array, the same brand and model of power supply, and motherboards with built-in ATI ES1000 graphics chips. We even swapped the removable units in the power supplies and checked our Watt meter to make sure the PSUs weren't somehow different. Satisfied with the general hardware configs, we then flashed to new BIOS revisions, wiped the disk arrays, and reinstalled the operating systems with the latest driver revisions. (I'm not paranoid, exactly, but I haven't forgotten the recompiled 3DMark DLL file that AMD sent out with its original Athlon test systems, either.)
The power-performance challenge boxes lingered in Damage Labs for a while after we received them, and before long, Intel's new "Clovertown" Xeon 5300 series came along. Intel sent us a Clovertown review system, a pre-production version of a Hewlett-Packard xw8400 workstation based on the workstation-class "Greencreek" chipset. This system came with dual Xeon 5355 quad-core processors, 8GB of RAM, an Nvidia Quadro FX4500 graphics card, and a pair of HP-branded SAS hard drives.
With apologies to the Quadfather, I think this may be the single baddest computer ever to set a rubber foot in Damage Labs. Despite having eight processor cores inside, the system's 120mm rear fan and smart overall cooling design makes its noise levels admirably low.
Sadly, though, the HP workstation box wouldn't work for an apples-to-apples comparison against the other two systems. Fortunately, the quad-core Clovertown Xeons are intended to be drop-in replacements for dual-core Woodcrest Xeons, so we flashed the BIOS of the SuperMicro motherboard in our Colfax box to the latest revision, popped in the Clovertowns from the HP workstation, and were almost instantly in business. The system booted up and worked with nary a hitch. Even the power management features of Clovertown, such as C1E halt and Demand Based Switching (also known as SpeedStep), worked properly on the SuperMicro board. So, at the end of the day, using these two Colfax boxes as our test platforms, we should have a resonably good comparison between similarly configured systems based on the latest processors from Intel and AMD.