IA-64 vs. x86-64 and other stuff...
First off, if you believe those 32-bit performance numbers allegedly from the Itanium are representative, I've got some beachfront property in Colorado I want to sell you...
Note this little blurb from JC's, which was conveniently buried at the bottom of one of his updates:
"Crap, that reminds me ... remember that article at tweakers.net about the Merced/Itanium's performace in x86 apps? A reader popped me a message informing me that the beta version of Whistler (in particular, what was used in the article) is most likely in fact using true software emulation and not the internal x86 translation capabilities of the Merced. Sorry about sitting on this, but I was waiting for a response from the emailed letting me talk about this (I don't remember getting one, but I've decided that I'm saying vague enough stuff here). Anyway, I just wanted to note this, because the slow performance may simply not be as much of an issue as indicated."
In other words, it looks like they turned a Merced into a Crusoe and then made it try and run x86-32. Not exactly cricket.
Aside from that, I'll grant everyone here that there is always risk (as opposed to RISC) in moving to a new architecture. However, there are limits to x86/x87. Intel knows that and is moving on to something new rather than letting themselves stagnate. AMD has always been a niche player, and that's what their x86-64 is. It will never be able to compete with IA-64 in raw performance because x86-64 is simply an evolutionary rather than revolutionary architecture.
What really blows my mind is the double standard I so frequently see when it comes to AMD and Intel. AMD's DDR chipsets with the 266MHz bus have been an unmitigated disaster... but if you read Ace's this morning, it's only a "minor disappointment." If the 400MHz bus for the P4 only ran at 300MHz, the IT press would be calling for Intel's head and plastering it on page 1 daily, "Intel fails again!" When Intel takes a chance and tries to develop a new high-performance architecture, they're a greedy monopolistic entity trying to gobble up more of the pie... but when AMD tries to fill a niche with an evolutionary change in an existing architecture they are a visionary destined to topple the mighty giant...
All this hyperbole makes a great Saturday morning cartoon plot... but this is the real world. AMD has recently been able to profit because Intel stumbled just as Intel has frequently been able to profit because AMD stumbled.
AMD may be in mid-stumble at this point. Back in September of last year AMD promised a speed grade every 6 weeks culminating in the 1.5GHz K7 in January of this year. It's February now, and their top speed grade is 1.2GHz. If you look at the entries on pricewatch, there are 39 entries for the "limited production" P4 1.5GHz and only 36 entries for the K7 1.2GHz. The parallels go down through the speed grades... Does AMD have a production problem at 1.2GHz??? Where are the 1.3, 1.4, and 1.5GHz speed grades which are now LATE??? ...and don't tell me "they don't need to release a higher speed grade." Given that the ASP of the 1.2GHz part is a pathetic $273, AMD could use the extra margin. Given the relatively weak performance of the P4 on non-optimized code, AMD has a window now to really pick up some sales if they released those higher speed grades. Waiting only allows more software to be released with SSE/SSE2 optimization which reduces the performance of the K7 clock-for-clock when compared to the P4...
Intel fouled up by going too conservative and not leading 0.18 micron production off with the P4... They also failed to properly forecast the cost of RDRAM, which led to high system prices for limited benefit. Now that the P4 is out which actually uses the bandwidth, RDRAM is more attractive, but it's still relatively expensive (though not nearly as bad as it once was). This left the door open for AMD for three full quarters. AMD took advantage by slowly ramping up the K7 and gaining mindshare. Now Intel seems to be getting things back on track... and AMD seems to be running into several problems... especially with DDR and 1.3+GHz K7's... Intel is also ~6 months ahead of AMD with 0.13 micron which will allow Intel to introduce much faster P4's and maybe even P3's (and bring the P4 to the mobile market). It will be interesting to see how the next 3 quarters play out... |
