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: Saturn V who wrote (89756)	10/8/1999 8:31:00 PM
From: Fred Fahmy	Respond to of 186894

Saturn V, I hadn't seen the article. Thanks. That one is definitely a keeper. Of course most of us here already understand and appreciate these positives, but the author has outlined them very very well. Thanks again, FF

To: Saturn V who wrote (89756)	10/8/1999 10:06:00 PM
From: vince doran	Read Replies (2) \| Respond to of 186894

SpecFP and the mighty clouds of joy - (Those who also read the AMD thread please forgive the cross-posting) For all of us trying to pierce the fog around those stunning CUmine benchmarks, JC's page jc-news.com has a very interesting message from a fellow doing some scientific simulation work. He modified his heavily floating point oriented code, running on an Athlon 500, to use the SSE prefetch instructions and voila! Up to 45% increase in performance on data-sets of similar size to those used by SpecFP. So the mystery remains: what will the actual performance of 733 CUmine look like? If Intel's assertion, as reported by Ten, (and by the way let me add my thanks for your great MPF reporting and other analysis) of ~20% imp over PIIIB holds, then the Athlon will be in trouble on many benchmarks. If, on the other hand, the bulk of the improvement was due to compiler-optimization, then the picture is much brighter for AMD. We have all had a couple days to ponder, what do you think? (I consider the estimated 266 FSB Athlon scores to be fighting today's fire with tomorrow's rain; I am mostly concerned with Athlon 700 performance on Oct. 25.) Vacillating, Vince

To: Saturn V who wrote (89756)	10/10/1999 1:39:00 AM
From: Rob Young	Read Replies (3) \| Respond to of 186894

streetadvisor.com "The Itanium processor is perhaps the most robust processor ever built. With Intel's billions of dollars of research and development funds pouring into its development, it will be virtually impossible for the Alpha processor to maintain any current lead for much longer." Okay.. I'll take a stab at that. Maybe the guy is just getting started or has Sun confused with Compaq's Alpha. The UltraSparc III rolls at year-end but MPF just last week , Sun presented UltraSparc V aka "Millenium". Doesn't it sound impressive? You gotta love Sun's marketing. They really do try. Anyhow, if you look at the numbers that Sun is estimating (gee, how hard can this estimating stuff be? Seems Intel is having an awfully tough time coming clean with Merced aka Inanium) 70 SpecInt95 and 120 SpecFp95 for US V, which is expected to tape 3Q 2000 (you can take that with a grain of salt). Compaq will be delivering those numbers by 3Q next year in 21264 and 21364. Back to the paragraph at the top: "perhaps the most robust processor ever built" That's a silly statement. So it has some fancy error correcting built in at the hardware level. The 21364 will support lockstepping. A must if you are supporting a fault-tolerant OS for the Tandem folks i.e. NSK. You really can't get more robust than that. Will it add to the server cost? Shertainly. Let the marketing types fight that one out. Point is, you can't get more robust than lockstepping. "it will be virtually impossible for the Alpha processor to maintain any current lead for much longer" We go from silly to absurd. Intel has yet to release performance numbers for Itanium/Merced. The pipeline length got a few grins going in various forums but as Intel claims: "initial implementation , blah blah" Okay, fine. Get around to shortening the pipeline and "boy we will really get good performance." So pipeline length and as the IBM fella points out "EPIC isn't high clock friendly" (paraphrasing). So how well does it perform? Compaq has shown/claimed: 21264 39 int 68 fp current numbers at 700 MHz (833 parts soon) 21264 will look much more powerful this time next year: Expect: .18 copper SOI part at 1.5 GHz, 1 GHz DDR-SRAM with several variations by Samsung. On-chip and off-chip L2. Low-end desktop to Workstation to Low-end server. 21364 on-chip network and memory switches, 70 int, 120 fp 3Q 2000. Much higher memory bandwidth, 1.5 MByte L2. High-end server. The part for I[nt]anium to fear greatly. 21364 aka Arana aka EV8. Just discussed at MPF. Tench mentions 100% greater integer than 21264. That must be 21264 at its peak as EV8 is projected to be 200 SpecInt95 , 300 SpecFp95. Now the question might be... What is that analyst talking about when he states: "it will be virtually impossible for the Alpha processor to maintain any current lead for much longer" He really doesn't know what he is talking about as the EV9 design team has been formed and EV10 is in the pipeline. If he even had a clue, he might be able to understand that Alpha should be a factor of 2 faster than Unatanium for at least 3 more years. That is saying something as today Alpha at 39 SpecInt95 is maybe 25-30% better in integer than X86 but by 21464 the gap should widen considerably (i.e. McKinley versus 21464). Not only "he doesn't know what he is talking about." He doesn't have any projected Intel numbers to compare it against and yet Alpha projected numbers are very much available, so how does he make a statement like that? Marketing. Okay. Rob