Cirruslvr,
As my cofinancier, I'm writing this post in your honor ;^)
The key points of Dirk's presentation:
1. Paul wasn't there.
2. Large volume shipments begin this month at 600 MHz.
3. The design methodology was very conservative. They did not push the design very hard to get to 600 MHz. They "went to great lengths to learn from the K6 experience." There is "a lot of leg room" in the design.
4. The 3DNow! implementation "pretty dramatically" increases performance relative to K6-2.
5. The point-to-point connections on the bus allow for much higher speeds than Intel CPU's. 400 MHz is easily obtainable.
6. No third party chipsets until the end of the year. Lots of support after then (VIA, SIS, etc.)
7. 22 million transistors.
8. 0.18u in Q4
Now the fun stuff...
specInt K7 w/half speed L2 vs. 550 MHz PIII Xeon w/full speed L2
At 550 MHz, K7 is 5% faster. At 600 MHz, K7 is 15% faster.
specFP K7 w/half speed L2 vs. 550 MHz PIII Xeon w/full speed L2
At 550 MHz, K7 is 38% faster. At 600 MHz, K7 is 43% faster.
Get a load of this...
3D Winbench K7 w/half speed L2 vs. 550 MHz PIII
Fully optimized SSE and 3DNow!
At 550 MHz, K7 is 40% faster. At 600 MHz, K7 is 50% faster.
In summary:
1. K7 runs at a higher clock rate than 0.25u PIII
2. K7 scores much higher on benchmarks than PIII.
3. K7 appears to be quite manufacturable.
Now the bad news.
Dirk refused to discuss systems delivery dates, or talk much about MP.
I spoke with some Intel buddies, who had spoken with someone from AMD, who said that MP was barely tested prior to tapeout. Dirk did say that MP systems were up and running in the lab, however.
Scumbria
|
