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: Paul Engel who wrote (67147)	7/31/1999 6:59:00 PM
From: Dan3	Read Replies (1) \| Respond to of 1573711

Re: Did Intel file for an INJUNCTION against VIA - to stop them from shipping some product ? Don't have the full details, just summaries. Look the link at: theregister.co.uk As usual, the Register is focusing more on clever wording than details (why I love those guys - some of their posts are brilliantly funny) but the jist appears to be: don't ship anything faster than 100MHZ or else! Remember that, while there have been fewer appearances of rambus than K7, the few reports I've seen say that rambus "800" is faster than PC100, but slower than PC133 - and a lot more expensive. But it has to be cheaper and easier to build a motherboard with 16 memory traces than 64. You can see why intel would try to keep PC133 from seeing the light of day. Samsung, either on its own, or with a little encouragement has posted an analysis that shows PC133 with higher latency than PC100 - footnoting the detail that they had mis-matched the memory bus. A modern CPU with a good cache spends close to 99% of its time in cache when running regular business applications so the hit that rambus takes due its latency isn't such a big deal on a 500 MHZ processor running Office 2000. But on intel's big money makers, multiprocessor servers with a lot of ram, what's usually running are either a couple of hundred processes or big databases with all of the indexes and joins cached in ram - here cache miss rates can go way up and a little latency can go a long way towards killing performance. As a possible example say your total latency with rambus is 70 ns and it's 50 ns with CAS 2 PC133. At 500MHZ on Office you execute 99 instructions, then lose either 25 or 35 depending on the memory. 248 compared to 268 ns to complete each 99 instructions - Not such a big deal, less than a 10% difference. But on a server running a 1GHZ coppermine or athlon you execute 50 instructions then lose 50 or 70 depending on the memory so now you're comparing 200 ns with 240 to complete one hundred intstructions. (2x50 to get the 100 instructions plus two cache misses instead of one) that's a 20% increase on lucrative high profile equipment, and I think that may be what's bugging intel.