I've got the rough first-pass performance numbers in hand. They're quite voluminous, so I need to cut them down before I can post them. While I'm waiting for this huge file to load into my HTML editor, I'll set the stage for the tests and the results. I plan to run SPECjbb, STREAM, and after a delay to get them compiled with the latest rev of Intel's compilers, SPECint_rate and SPECfp_rate. For burn-in, I used I’ve got the rough first-pass performance numbers in hand. They’re quite voluminous, so I need to cut them down before I can post them. While I’m waiting for this huge file to load into my HTML editor, I’ll set the stage for the tests and the results. I plan to run SPECjbb, STREAM, and after a delay to get them compiled with the latest rev of Intel’s compilers, SPECint_rate and SPECfp_rate. For burn-in, I used a shareware app called SiSoft Sandra, and I re-ran that application overnight to get its full performance report. I’m a benchmark snob, and as such I have to warn that until I prove them out with standardized tests, Sandra’s report results have to be taken with a grain of salt of approximately one foot in diameter. Sandra is built to test desktop machines, and in the past, I’ve used it purely for burn-in and to make sure that systems under test aren’t hobbled by poor BIOS defaults. Sandra’s code and its methods are opaque, so I don’t trust it to produce meaningfully comparable results for servers. I’m passing along only that raw data that a patient reader can interpret; analysis will follow the results of testing done with standardized, transparent benchmarks. Prior to running Sandra, I verified that the components used to build this Supermicro-based reference server are all commercially available, off-the-shelf parts. The 2 GHz Quad-core Opteron 2350 CPUs installed are the final retail version, not samples, and the system is in no way boosted above rated defaults with regard to frequencies, voltages and latencies. I cleared the BIOS to its automatic settings, and then applied AMD’s short and reasonable list of adjustments to those parameters. Sandra confirmed the vanilla configuration except for its confusion about the system’s front-side bus speed. Understandably, SiSoft’s CPU database had not been updated to account for Barcelona at the time of my testing. However, I am impressed with the details that Sandra was able to pull out of components’ configuration memory and snooping around the system management bus. Sandra didn’t know Barcelona from a hole in the wall, but it discovered what it’s made of. I installed Windows 2003 Server Enterprise x64, Service Pack 1 from scratch, using the default drivers supplied with the system and letting Windows run with its post-install defaults. I downloaded the 64-bit version of SiSoft Sandra XII 2008 directly from the vendor’s Web site. I used the default options for all tests. The results database that is included with the trial release is dated, and does not include Intel’s quad-core systems, providing me with another reason to withhold Sandra’s reported comparative results. So much for the explanations and disclaimers. I’ve got some serious HTML chopping to do, and I’d best get to it before jumping into a full schedule of meetings with AMD. Technology Industry