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 : Compaq -- Ignore unavailable to you. Want to Upgrade?

To: jim kelley who wrote (71250)	11/8/1999 8:19:00 AM
From: rudedog	Respond to of 97611

Jim - You are right - Intel will eventually get a fix for the motherboard problem. Intel currently claims that OEMs can ship 2MB cache and 1MB cache versions. However, independent testing (in particular performance tests at both MSFT and ORCL) show that the Intel "fix" imposes a "tax" on their version of the profusion architecture which cuts performance by as much as 10% in memory intensive applications, including the database benchmarks. The problem stems from maintaining cache coherence when a memory access is across the "hemispheres" of the profusion memory architecture, which essentially links 2 4-way systems through a crossbar. When the cache subsystem "snoops" the far hemisphere, it stalls on update of its own cache waiting for the status. In some cases this caused the system doing the snoop to wait forever, effectively killing the system. Intel's "fix" appears to put a limit of 50 cycles or so on those accesses before declaring a miss/refresh. Since the problem is relatively infrequent under many system loads, this did not impact performance for most applications. But for applications which have broad memory access, where there is not much locality to the access, the difference between 9 memory cycles for a local access vs. 50 for a far begins to take its toll. Actually fixing the silicon on Intel's version of profusion will take more time. DELL and others using the Saber board have held back on volume shipments awaiting Intel's "fix". They will probably start shipping in volume with the modified motherboard, but I would not expect performance numbers until the new silicon is out. CPQ designed their own version of Profusion which does not have the problem, and the CPQ field has been well-versed on how to demonstrate the performance differences in competitive situations. DELL (and all the vendors of Intel-based servers) are moving up the food chain but the rate of movement has slowed a lot this year. Most market research shows that NT growth is "horizontal" - customers are deploying more NT in areas where they are happy with the functionality but have slowed deployment into areas traditionally held by Unix. Sun has made the most of administrative and management issues with big NT boxes. That may change with Windows 2000 but not until late 2000 when the Datacenter version is available. I was not holding CPQ up as a model of innovation. But I guess we can not talk about DELL without a reference to how much better they are than CPQ...