Aceshardware's CeBIT 2002 Coverage.

[AGodspeed]

http://www.aceshardware.com/read.jsp?id=45000314

AMD Coverage:

"When will we see ClawHammer?". Jan Gütter said that he can not recall a time when anybody in the CPU industry ever got working A0 silicon (ClawHammer), and surely not A0 silicon that booted a complex operating system such as 64-bit Linux or 32-bit Windows XP. Mark grinned and added, "It is very, very likely that we'll launch the Clawhammer in Q4 of 2002." Mark de Frere put up an even bigger grin when I asked about the SledgeHammer's timeline and commented: "It is going to blown you away." Well, Mark told us that we will see quad-processor SledgeHammer server motherboards in H1 2003, or most likely June 2003!

What about software? AMD has not confirmed this officially, but the slightly vague comments of AMD's PR made it clear to me that Microsoft is working pretty hard on both a x86-64 version of Windows XP and a compiler for x86-64.

Linux support is, of course, very good and we have heard a few performance estimations. As you know, AMD estimates that ClawHammer will run 20 to 25 percent faster than the current Athlon XP at the same clockrate in 32-bit applications. In other words, IPC seems to be 20-25% better.

Intel Coverage:

...Intel showed us an "ice cooled" 4.1 GHz Pentium 4 as well as an air-cooled 3 GHz Pentium 4. The 3 GHz Northwood Pentium 4 will see the light around Christmas 2002.

While HyperTransport is AMD's crown jewel, HyperThreading is Intel's crown jewel. We still have to wade through the enormous amount of technical white papers, but Intel sees a great future for this technology. In the second half of 2003, Intel will launch Prescott, the first 0.09-micron (90nm) CPU. According to our industry sources, Intel's transition to 90 nm might be 4-6 months ahead of AMD's. Hans-Jürgen Werner emphasized that this process technology superiority allows Intel to produce more complex processors with more cache.

Intel really wants to bring HyperThreading technology to the home desktop, as they believe it will add more value than 64-bit adressing. Intel speculates that the current desktop PC might be a sort of home server in the future, with many thin home appliances connected to it. In other words multi-threading will become more and more important as each of these home appliances (MP3 music player, MPEG4 movie box.... ) will run different tasks and threads on the central PC. Therefore, Prescott, which will probably feature 1 MB of L2-cache (not confirmed), will bring HyperThreading to the home desktop market.

 

[KenAF]

Other information from the same page on Aceshardware's CeBIT coverage:

<< Both AMD's Thoroughbred and Barton have a 256 KB L2-cache and a 266 MHz FSB, so no change from the current Athlon XP.
Athlon will not be killed off quickly (like what happened to the K6) when the Hammer family appears. Throughout 2003, the Athlon and Hammer will co-exist and the Athlon will probably live on even longer.
Each Hammer with a different clockspeed has a different Mem/CPU-speed divider.
SledgeHammer L2 cache is up to 1 MB in size. >>

From Johan's report here and on the forums, it seems the Hammer will have a hardwired memory/cpu divider in the processor. So if it is somehow possible to overclock (still doubtful, IMO), doing so will also allow use of DDR faster than PC2100 or PC2700.

<< Intel's transition to 90 nm might be 4-6 months ahead of AMD's. Hans-J&uuml;rgen Werner emphasized that this process technology superiority allows Intel to produce more complex processors with more cache. >>

I told you so. Analysts on financial boards are saying Intel will have full production of 90nm P4's by April 2003, with processor release in May or June. AMD's first 90nm processor is the 4400+, for release by end of 2003.

<< Therefore, Prescott, which will probably feature 1 MB of L2-cache (not confirmed), will bring HyperThreading to the home desktop market. >>

The 1Mb cache doesn't explain the "more than 100 million transistors" that Intel has quoted for Prescott. A "Northwood" P4 modified with 1Mb L2 cache would have less than 56+2*13=82 million transistors. There will be more than 1Mb cache...or Intel will be heavily reworking parts of the P4.

 

[AGodspeed]

Lol, notice the word might in that sentence. Intel might be 4-6 months ahead of AMD's .09u transition. Of course I wouldn't doubt it, Intel completely dominates AMD in every way in terms of Fab capability and time to market ability, but it's still premature for a "I told you so."

Besides, what will happen when UMC starts pushing out 7-8 Million+ Thoroughbreds out the door by next year? Surely transitioning will be easier.

 

[zephyrprime]

How can we be getting .09 so soon after we just got .13?

 

[KenAF]

> How can we be getting .09 so soon after we just got .13?

Just think about it....

Intel released its first .18um P3 processor on October 25, 1999. Intel shipped its first .13um processor (Tualatin) on August 26, 2001. That's a .18um life of 22 months. The move from .18um to .13um required completely new equipment and fabs.

Intel's first .13um processor (Tualatin) was released on August 26, 2001. Analysts predict the first 0.09um "Prescott" P4 will be released in May or June 2003. That's a .13um life of 21 months, or one month difference.

Moreover, .13um has the advantage that much of the equipment can be reused for .09um production.