Silicon Investor (SI) -- The First Internet Community

STOCKTALK

We've detected that you're using an ad content blocking browser plug-in or feature. Ads provide a critical source of revenue to the continued operation of Silicon Investor. We ask that you disable ad blocking while on Silicon Investor in the best interests of our community. If you are not using an ad blocker but are still receiving this message, make sure your browser's tracking protection is set to the 'standard' level.

Technology Stocks : Intel Corporation (INTC) -- Ignore unavailable to you. Want to Upgrade?

To: Dan3 who wrote (152822)	12/19/2001 1:41:38 AM
From: wanna_bmw	Read Replies (1) \| Respond to of 186894

Dan, Re: "SPECint is a measure of integer performance. Server processor performance is primarily integer performance." That's not quite true. SPEC does a reasonable job of testing integer based throughput in highly optimized libraries that tend to reside primarily in cache or memory. Real life server applications, however, benefit highly from larger memory bandwidth and a high performance I/O subsystem. Look, for example, at the 700MHz Cascades processor with 2MB of L2 cache, and compare it to the 1GHz Coppermine processor with 256KB of L2. specbench.org (SpecINT_base = 454) specbench.org (SpecINT_base = 414) On SPECint, the higher megahertz processor wins, but thanks to the L2 cache, the 700MHz processor isn't far behind. But look at how things change when running TPC. 4-Processor 700MHz Pentium III Xeon with 2MB L2 cache TPC-C Throughput 34,265. tpc.org 2-Processor 1GHz Pentium III with 256KB L2 cache TPC-C Throughput 17,336 tpc.org Despite being 300MHz short and scoring 9% slower in SPECint, the 4-Processor Xeon scores nearly twice as much as the 2-Processor Pentium III. This great scaling is partly due to the larger L2 cache, but the largest factor is in the platform. The Pentium III Xeon platform was probably proprietary from IBM to deliver exceptional performance and scaling. You don't usually find systems with double the number of processors scoring twice as high on benchmarks. You have to have a solid platform to make it happen. Similarly, the Itanium platform from Intel, the 460GX, was designed to work through heavy data loads while providing higher reliability features like ECC SECDED. These reliability features, called RAS, are just as sought after as performance (and sometimes more so). While the Itanium CPU itself may score poorly on SPEC (which isn't surprising, given the novelty of the platform), it will probably perform great in high-end server applications, which is where it is being aimed. I think that seeing the TPC scores will decide more determinately how good the performance will be. And you won't be seeing a single Celeron CPU tested in a benchmark like TPC. People already know that the limited cache size creates the biggest performance loss in these kinds of applications (and the small 256KB cache of the Pentium III is already becoming obsolete in servers). Remember that no one is buying Itanium machines to use as a front end. They aren't buying Itanium to manage their workgroup or run as an application server - at least not yet. Right now, Itanium is being offered for back-end jobs, which require heavy database work. That's why TPC will be a valuable benchmark to consider - not SPECint. I know for some people here it must be a monumental joke how low Itanium performs in SPECint. Hopefully, McKinley scores much higher, since McKinley and later cores will be meant for a much more general audience. For Itanium right now, though, SPECint just isn't that important. While it is still in the back-end, it will be back-end workloads that tell the real story. wbmw