RealTechNews.com - Independent Tech

Friday, October 15, 2004

Intel Nixes Plans for 4-Gigahertz Pentium
Uh, oh, looks like Intel is unable to release a 4 Gigahertz version of its Pentium 4 CPU. The company nixed plans entirely, after anouncing an end of the year delay in July. This is another bad moment for a bad year at Intel.

From AP: "The latest change in Intel's roadmap follows a string of adjustments and missteps by the Santa Clara-based company. In May, it canceled work on a successor chip to Prescott so that it could focus on more promising technologies. In June, a manufacturing glitch forced a small recall of chip sets, which handle communications between the processor and the rest of the system. Then, in July, the company said design problems would delay the release of mobile computer chip set dubbed Alviso until next year. And it said better-than-expected performance in manufacturing of Pentium 4s resulted in an inventory buildup. "

Bill? Thoughts? --Alice

BILL SPEAKS: Here I am! You know, must be that coastal time difference because you beat me to the punch! I talked to these guys no more than 45 days ago for a piece I did in the most recent Maximum PC and they were totally on track for the 4GHz processor. But I think they woke up. (Couldn't have been anything I said.) They were already having a delivery problem with the 3.6GHz CPU by then and probably anticipated something similar arising with the 4GHz model. Duh! But the truth is, Intel doesn't really need 4GHz right now. If you shrink the die size, you pick up performance because the data endpoints are closer together. If you increase the cache, you also pick up performance, at least "in the meantime." Odds are that they really will get appreciably more from those two moves than they would have with a 400MHz speed bump --and they should be able to do that without worrying about RFI induced crosstalk scrambling the CPUs ops.

All in all, I think Intel finally discovered that the road on the map leading strictly to clock speed gains was just a hair on the page and have blown it off. For those who will now compare Intel's epiphany to AMD's modus operandi, please note, Intel is working in comparatively stratospheric clock rates --something that AMD's model numbers mimic but don't come even close to approaching. That's not a negative for AMD, just a fact. Currently, unless you get into the $deep$ end of the Intel gene pool, AMD's processors are super gaming cpus. But if you're thinking that this move is an attempt to catch up to AMD, you're being naive. If you overclock a smaller die CPU, the effect is enhanced because of the shorter data paths. AMD will now have to hustle a bit more to keep up with Intel. Effectively, AMD is about to be forced into spending more on R&D than they would have if just faced with a clock speed increase.

Now, when Intel gets to dual core, if it also incorporates HT, the world may well be its oyster. --Bill

# Permalink Posted at 8:40 AM 5 comments Email this Link

Comments on this Item:

Bill,

So what you left out is that AMD hasn't degraded its roadmap in the past two years, and Opteron's core was always intended to be at least dual. AMD hasn't done the paper-launching that Intel has either, nor have they recalled processors/chipsets.

WRM

# posted by Anonymous 5:26 PM

Didn't necessarily leave it out, but I'm trying to be fair. AMD has a much more conservative road map, letting Intel do the primary groundbreaking.

Okay, I can hear you already, "Intel didn't ground break 32/64 bit technology!" Yes,that's right. But that's sheer marketing for the desktop crowd. By the time there's a real 64-bit OS for desktops and 64-bit apps, I can almost guarantee you there won't be a 32/64. For now, it sounds real good.

# posted by Bill 5:53 PM

Considering how long 16/32 lasted, I do not doubt a 32/64 co-existing in a world of the cheap. Not too mention a world of the poor :p Incidentally, claiming Intel is breaking new ground constantly is patently wrong. That's like saying it had the first slot based processor. Many things in the current line of chips came from purchasing patents or reverse engineering, even in the mobile lines. Now will AMD have a tougher time considering that Intel has finally pulled their head out of the clockspeed ***? Yes, but remember that Intel didn't do it by choice, they had to cancel chips due to thermal considerations. Heck have they even fixed the whole 1+1=3 debacle, or the pi=3 foolishness yet? Incidentally Bill, you might like to consider that a Sempron 3100+ (A budget chip with less cache and lower clockspeed) beats an Intel Prescott 3.2 for Doom 3 Framerates, without overclocking. Cache is not a panacea, as proven by the Extreme Editions being beaten by normal Clawhammer 64's with about 1/4 the cache in the same game. Right now P4's are only good for two things: Music and Video. That long pipeline gives the P4 considerable power over the stream that AMD currently cannot match. Still given the difference in raw power, and thermal considerations, the A64 is the chip on the market right now. Hyperthreading? Pah! I've got a P4-3N too, and I'll tell you what, HT doesn't really improve the user experience.

# posted by Anonymous 6:12 AM

Bill,

Bottom line: Intel has been getting beaten/matched ever since the introduction of the "Classic" Athlon. If anyone is hustling it's Intel. Itanium is not working out, the P4 has hit the wall, and there is nothing on the roadmap to replace either of them. A total R&D "pipeline" purge will cost Intel about five years to get a "new" architecture to market. In the meantime they plan to pull the PIII(M) off the bench?

AMD's conservative roadmap still has a dual core processor on it, Intel still does not.

Maybe your Buds over there can shed some light on how Intel's hiring and firing, and forced early retirement practices played into this, but from the looks of it they dumped too many of the "old heads" prematurely.

WRM

# posted by Anonymous 9:22 AM

Smaller die size and "shorter end to end paths" for data within a die dont usually mean a faster CPU, this is because of the way that a synchronous circuit (which most of a CPU is) works. Shorter end to end paths will only yield results if it means that they can push the clock speeds higher (due to having to worry less about propagation delays).
Smaller die size also means lower power consumption and less heat (which is definitely a good thing for the P4). And this can sometimes allow the clock to be pushed higher.
The current P4 architecture was pretty much designed to scale well with clock speed, yielding almost a 1:1 ratio between clock speed increase and performance increase. So this is actually quite a blow to Intel, especially as all signs point to their next architecture being at least a few months away.

-Perros-

# posted by Perros 12:36 PM

Post a Comment