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.

Politics : Formerly About Advanced Micro Devices -- Ignore unavailable to you. Want to Upgrade?

To: steve harris who wrote (79797)	11/13/1999 10:48:00 AM
From: Bill Jackson	Read Replies (1) \| Respond to of 1586221

Steve, What a delightful bonus for AMD!!! I can hear the wailing and gnashing of teeth from Santa Clara. I bet the InHell high command is drawing up the 'who do we shoot next' list. What a fate, the largest chip maker in the world brought down by an infestation of silicon lice. Bill

To: steve harris who wrote (79797)	11/13/1999 10:57:00 AM
From: Dan3	Respond to of 1586221

Re: So much for iNhEL's Credibility....... Hi Bill, I don't think Intel has presented any false information, but they have certainly been misleading. The only quote I've been able to find is "We plan to ship millions of Coppermine (chips) in the fourth quarter". Sure sounds like they should be everywhere, but maybe not. First of all, those 450MHZ mobile parts may not sound like they should be part of the discussion, but they are still coppermines, so if 500K of those parts get shipped by December 31, we're down to 1.5 million desktop parts. There are a lot of reports that Intel is facing more mindshare problems in Europe and Asia than in the United States, so they would be sure to send a decent share to those areas, so call that another 500K parts. If the ramp is steadily increasing, then more parts will be shipped in December than in November, and more in November than in October, so of the total 1 million US Desktop parts to be available by the end of the year, only 2 or 3 hundred thousand would have been shipped by now. If 700 and higher parts are 20% of the binsplits, then there would have been 40 to 60 K of those parts shipped by now that would have to be split between Dell, Gateway, IBM, Compaq, HP, general distribution, etc. That volume of parts would be just about invisible - remember what 250K of Athlons looked like? And none of the Athlons were going direct from Dell, Gateway, etc. to business desktops. So Intel will probably ship "millions of coppermines in the 4th quarter" - but that phrase is pretty misleading - especially since anything shipped after December 20 won't see the light of day until next year. Despite shipping millions of coppermines in the 4th quarter, a visit to some Chicago area Best Buy and Circuit City stores last week showed both stores with Athlons from 500 to 700MHZ in stock, and nothing from Intel over 600MHZ from anybody. Others have reported similar experiences. The high end machines were generally Athlons (though I saw one HP Pentium III that was loaded with CD writer, etc). There is still the possibility that there are still some manufacturing wrinkles to iron out. I know nothing about the FAB process, but I've seen the results over the years. Two of the more frustrating problems I've seen involved cache tag ram. First on an Everex step 25 (386, 25 MHZ, had CACHE!, it was the fastest thing on the planet at the time :-)that ran SCO UNIX. It would boot OK, run for a while, then panic (like a blue screen if you're not used to UNIX). Took a while to track down the problem to the cache tag. Similar thing happened with an early Multiprocessor Pentium machine - a real headache. We saw that AMD had a tough time when it added 256K L2 cache to the K6, it may be that the same thing is happening to Intel. If you have a better understanding of these issues, can you or someone else step in here? As a rough start, as I understand it, a cache controller is basically a very large ALU structure (with similar, very high, performance requirements) that can be harder to test than other ALU structures, since many of the possible "mistakes" that its logic can make (not correctly reflecting memory updates) often don't cause errors (since failure to correctly update some area of main memory won't matter unless that area of cache is reused before the next access). An 8 way cache running at 700MHZ sounds tough to build - doesn't that mean 8 logic compares must be made for every memory access? With 3 integer units, it could require up to 24 cache logic operations per cycle (more if floating point operations are also going on). That's a lot of work to do in a few nanoseconds - but then I still think the 8080 was a fabulous device, so maybe I'm way off base and just being goggle eyed about modern CPU technology. Regards, Dan