SI
SI
discoversearch

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.
Technology Stocks : Advanced Micro Devices - Moderated (AMD) -- Ignore unavailable to you. Want to Upgrade?


To: kash johal who wrote (4676)8/11/2000 10:16:09 PM
From: Joe NYCRespond to of 275872
 
Kash,

Just to put your message in perspective, what is the IPC number for Thunderbird or Mustang? It seems that you are assuming 1.

Joe



To: kash johal who wrote (4676)8/11/2000 11:52:15 PM
From: ScumbriaRead Replies (2) | Respond to of 275872
 
Kash,

Merced uses EPIC so IPC is much higher than current x86 approach. Max is around 8 : min say 2x AVG of 4x for typical due to compiler difficulties etc. Thats maybe equivalent to 3.2Ghz 64 bit clocks.

There are a number of flaws in your assumptions here.

1. Your guesses about EPIC IPC values are probably too high.

2. Most x86 instructions are the equivalent of three or four EPIC instructions, so you need to divide EPIC IPC by about 3.0 to get an apples to apples comparison.

3. The code density of EPIC programs is terrible, causing poor cache performance.

Merced will run desktop software slower than fast x86 CPU's. The only value which Merced may offer is for server applications.

Scumbria



To: kash johal who wrote (4676)8/12/2000 12:14:47 AM
From: semiconengRead Replies (2) | Respond to of 275872
 
I think you raise a good point.

It is very unclear to most everyone what the relative performance will be.

If we assume that Merced runs at 800Mhz tops with todays Intel 0.18 process it gives us a figure of merit. Merced uses EPIC so IPC is much higher than current x86 approach. Max is around 8 : min say 2x AVG of 4x for typical due to compiler difficulties etc. Thats maybe equivalent to 3.2Ghz 64 bit clocks.


This is all Device Design stuff, I'm Process, so if you say it's equivalent to 3.2GHz 64 bit clocks, I'm willing to take your word for that. Good enough for me.

If we assume a hammercore - modified Mustang tops out at 1.2Ghz with AMDs current 0.18 process. In a broad brush look -- with dual CMP that gives us 2.4Ghz 64 bit clocks.

Now I have to say that THIS, seems like a BIG assumption, and the width of the "brush stroke" you're painting could cover a football field. I also couldn't help noticing your comparison of Dual CMP Hammer, to single Chip Itanium.

Now obviously both CPU's will scale as process/design improves.

The key is that Sledgehammer may well be 20-30% slower.

I can't see it being 2-3x slower.


Maybe I'm not reading this right, but it still seems like you're comparison is based on Dual CMP Hammer and Single Itanium. Am I missing something?

However it will be much cheaper. It will also add the legacy s/w support.

Kash, cumon. You know just as well as I do, that in the Big Server market that AMD wants for this chip, cost is one of the last considerations. You'd have a point for desktops, but we both know that's not what AMD wants for this chip, don't we?

It certainly has potential to be a winner in displacing the current Xeon 2/4/8 CPU space and blowing it out of the water.

Let me tell ya something baby, As soon as McKinley shows it's face (right around the time of Hammer intro), XEON is DEAD. Itanium will be the low end server processor, Willamette's second generation will be high end Desktop, and CuMine will replace the then obsolete Celeron. And you know it.

I think AMDs plan has some merit.

The S/W is the driving factor as it is usually the most expensive part of the equation when coupled with its attendant support issues.

regards,

Kash


Maybe the plan has some merit. I just think it's going to be too little, too late. I don't think you really believe that bullcrap that Mike Magee is spewing, about SledgeHammer samples, now do you? AMD's own Web Page has them "announcing product" at THE END of 2001:

http://www.amd.com/news/prodpr/20145.html
AMD's x86-64 technology will first be supported in the family of processors codenamed "Hammer," planned to be announced at the end of 2001


And you're right, it's all in the software baby. I could cut and paste all the web links to the announcements about IA-64 support, but I'm sure you've seen plenty. And we both know that Sun isn't about to throw down UltraSparc for no chip, that's more than a year away, don't we?

I have no doubt that AMD will do fabulously with Gamers that are wanting to transition from 32 bit to 64 bit, but in my opinion, I honestly believe that when system administrators are sitting there thinking "Should I buy intel's second generation IA-64 McKinley, or go with this band-aid new SledgeHammer product", I just don't believe they'll pick AMD.

I could be wrong, but I don't think so.

SemiconEng