Unrelated to AMD (well, it could be ...)
<If you look at the 21364(EV7 gets little press) and compare what we know of Merced... A very good analysis of Merced bandwidth was done in one of the Usenet groups. The conclusion was Merced does 2.1 to 2.6 Gigabytes/second CPU <-> Main Memory. These are 21264 bandwidth numbers. Outstanding numbers for 1998/1999. However, the 21364 does 6 Gig/sec memory bandwidth:>
I won't comment on the Merced numbers. I know exactly what they are, however, and I know exactly what the McKinley numbers will be as well.
I also know about the archtecture of the 21364 from Microprocessor Report, 10/26/98. Looks great; seems like a very ambitious architecture. (It even involves RDRAM, one of my arguments against the rumored demise of Rambus.) The time frame for 21364 is shipments in late 2000 or early 2001, or so says MPR.
That means that the 21364 will be released in the timeframe between Merced (mid 2000) and McKinley (late 2001), assuming the schedules don't change. So as MPR said, it will be hard to see what is positioned against what.
<Merced is no where near 21364 from what little we know about Merced. At least we have a good feel for Merced memory bandwidth, 21364 bandwidth shows up in McKinley.>
I will tell you this, however. Much of Merced's performance was targeted toward a late 1999 release, before the Merced delay. So it doesn't surprise me when you say that Merced is no where near 21364. The thing is, when will the 21364 be released? And considering that the 21264 was itself delayed over a year, will the 21364 be released on-time?
Tenchusatsu |
