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: Tenchusatsu who wrote (140387)	7/29/2001 5:52:37 PM
From: Dan3	Read Replies (1) \| Respond to of 186894

Re: claim that 2-way Palomino beats 4-way Xeon in MANY benchmarks. There aren't a lot of even vaguely impartial tests in this area. The only test attempting to compare Dual 1.2GHZ Athlons with Dual 1.7GHZ Xeons in a server application showed Athlon beating Xeon pretty convincingly. anandtech.com The GTL+ bus has certainly been a lot easier for implementing SMP boards - Intel, VIA, serverworks, etc. have been cranking out such SMP chipsets for years. But putting even two chips on such a bus results in contention for the memory bus. It's a simple bus that doesn't scale well for multiprocessor applications, but it does make it easy to add additional processors, however fast any incremental benefit may fade. Xeon is also limited to caching 12 locations for any given offset. As long as you are running a small number of processes, that's plenty, but as you scale up the size of a server and increase the number of users and/or the complexity of the applications being run, Athlon's ability to cache 18 offsets per page will keep its performance from degrading under loads that flatten Xeon. A Xeon with 4 meg of on-die L2 cache would still be limited to caching 12 locations with the same LSBs. Even a Duron with only 64K of L2 can cache 18 locations with the same LSB. As a server is scaled up and expected to run more processes, the PIII, P4, or any Xeon will begin thrashing its cache long before any Athlon or Duron. So the Athlon architecture, with its 2-way + 16-way cache puts less of a load on its memory bus to begin with than Xeon with its 4-way + 8-way cache. Add to this the point to point design of Athlon's memory bus compared to Xeon's simple shared bus design, and it would be quite amazing if Athlons didn't substantially outperform Xeon in multiuser database and other server applications. Xeon is architected like a network hub, while Athlon is configured like a network switch. The heavier the load, and the larger the number of "users" (processes in this case) the more beneficial the switch architecture is compared to the hub architecture. There is, of course, much more to selling multiuser systems than just performance, but the same thing that will save Intel from AMD (for a while) will also make it very hard for Itanium to make much headway for quite a while. Eventually, Intel will have to offer up something better than a simple PC architecture with extra processors tacked on, especially when the competition has a real server platform available. If they don't, AMD will take that market away from them, just like Intel's better performing processors have finally been making marketshare gains against SUN in the workstation market.

To: Tenchusatsu who wrote (140387)	7/30/2001 11:25:09 AM
From: pgerassi	Read Replies (2) \| Respond to of 186894

Dear Tench: There is a case of where a dishwashing detergent helps remove stains for washing clothes (My mother recommends Dawn Dishwashing Detergent for stain removal). Similarly, it is one of Athlon's strengths to be very good at a wide range of tasks. It is a mark of a well rounded processor. P3 Xeon is no longer a computation powerhouse. P4 Xeon is definitely not a well rounded processor. It displays schizoid tendencies, on some tasks is does quite well, on others extremely poorly. Quad P3 Xeon is not a very powerful at computation, yet so many others seem to think it is. Database serving does not require great amounts of computation but, do require lots of disk both in speed and in quantity. It also requires lots of I/O. Due to much time to maximize the platform's traits in this area, it is currently quite good at this. There are signs that Athlon is better at it but, acceptance will take time. It is interesting that both Mary and Tony think that most systems are built for computational loads and not database loads. Database loads are much more typical for servers. But, they keep pushing the computational configurations. Well, Athlon has proven itself on computational loads and is beginning on database loads at the small and medium levels. High end database loads will come in time. Another curious fact is that 4 ways composed of two pairs of Xeons on two separate busses are much faster than four Xeons on a single bus. It shows both in database loads and web serving loads. This shows that the bus concept does not work well going from 2 to 4. Athlon gets more from adding a second CPU than Xeon does when running on a single bus. It appears that AMD and DEC were justified in their belief that P2P links were better than busses to connect CPUs to chipsets. Pete