CPU of Tomorrow, Just Not the CPU of Today
Written: Nov 28 '00 (Updated Apr 14 '01)
|
Product Rating:
|
|
|
Pros: SSE2 instructions, high RAM throughput, high clock speeds
Cons: Poor speed-to-performance ratio, expensive, reliance on SSE2
The Bottom Line: Even with an all-new architecture and insanely high clock speeds, there is virtually no reason to buy a Pentium 4 based system at this time.
|
|
|
| vicwang's Full Review: Intel Pentium® 4, 1.8 GHz (BX80532PC1800D) Pr... |
After years of waiting, the Intel Pentium 4 (P4) was finally released last week, representing the most dramatic departure from the original Pentium Pro architecture since its inception in 1995. Oddly enough though, the P4's release was accompanied by hardly a whimper from the Intel marketing machine, at least compared to the all-out marketing blitzes that typically come with major CPU releases (Pentium MMX, PII, PIII). It almost doesn't "feel" right that the Pentium 4's release was downright tame by comparison. Even the Pentium MMX, a chip that in retrospect was about as "revolutionary" as a Pentium III 700 is to a Pentium III 650, seemed to have several times the marketing budget (and several times the hype) of the P4. And for anyone following the trials and tribulations of Intel over the past year, the irony that Intel would just now begin a major TV advertising campaign for the Pentium III (over a year after its release), instead of the P4, is downright palpable.
It seems almost a tacit admission of what many industry insiders have known for quite some time: even though the Pentium 4 represents an all-new architecture (the first in over five years from Intel), and is designed to reach insanely high clock speeds (currently up to 1.5 ghz), there is virtually no reason to buy a P4 based system at this time. Even Intel has avoided marketing the P4 as a "mainstream CPU," while stating that the Pentium III will continue to comprise a bulk of Intel's sales through 2002.
Admittedly, this notion that "the average user has no need for this chip" is not a new one when it comes to the latest and greatest CPU. Every time a faster CPU comes out from AMD or Intel, industry pundits are right there ready to downplay its significance by saying, "it's not going to make a difference for most users," or "nobody really needs this chip right now," etc. But the Pentium 4 takes this mantra to an extreme. Not only is it the most expensive mainstream CPU available today, but in most standard benchmarks and applications it can only sometimes match, and rarely exceed, the performance of the fastest Athlons from AMD (which happen to be significantly cheaper). Surprisingly enough, even the 1 ghz Pentium III (a chip released almost nine months ago) can outperform the latest Pentium 4's at many common tasks, while the P4 only truly shines in the relatively few tasks that currently exploit its advantages.
All in all, the P4 is a highly unusual CPU, which makes a conventional "strengths vs. weaknesses" breakdown difficult if not impossible. It could be said that many of its strengths are actually "weaknesses in disguise," as well as vice versa (or they may turn out to be, depending on future developments). The following sections will discuss each of the key aspects of the P4, while showing how they may prove to be strengths or weaknesses in the long run. Note that if there are any terms you're unfamiliar with, I've included a glossary with definitions of all major technical terms (don't worry, there really aren't many) as the first comment on this review.
Hyperpipelining: "Speed, Not Performance"
Most people consider "speed" to be the same as "performance" when it comes to processors, and as a general rule that has been true in the past. But there's actually a tradeoff that needs to be made when designing a processor: very high performance relative to its MHZ rating, but a difficulty in reaching high MHZ speeds (e.g. Apple G4's, which are currently only available up to 500 mhz); or, in the case of the Pentium 4, a CPU that can reach incredibly high clock speeds but is relatively underpowered clock-for-clock. Neither is necessarily better, but it's simply a decision made when the "blueprints" of a new CPU are first laid out. In the long run, it should really make little difference to the end user, since "real-world" performance should be similar either way.
Of course, it's a brilliant decision from a marketing standpoint. With the ever-escalating "MHZ War" between Intel and AMD (a war which, ironically, Intel lost resoundingly the last time around with the Pentium III vs. the Athlon), the general public has become conditioned into thinking that MHZ is everything, no matter how meaningless it can be (not to mention the "bragging rights" that come with owning the "fastest" CPU!). If the average user walks into a store and sees a 1.5 ghz Pentium 4 system and a 1.2 ghz Athlon system side-by-side, which one do you think he'll pay more money for?
In the case of the Pentium 4, it achieves such a high speed/performance ratio through something called "hyperpipelining". Every time a CPU needs to perform a specific instruction, that instruction needs to go through many "stages" before it's complete (the most common analogy is a factory's assembly line). Since it takes time for an instruction to make its way through all the steps, and CPU's these days can process instructions far faster than they can actually receive them, they use something called "branch prediction" where the CPU begins carrying out instructions before it knows what's going to happen. That's why the length of the "assembly line" can be a liability: any time it predicts wrong, that instruction needs to go all the way to the beginning and start over. And unless it can predict correctly on a consistent basis, that means a lot of wasted processing power.
But the benefit of a long assembly line is the extremely high clock speeds such a CPU can reach. Without getting into the technical details, consider that it's easier for a CPU to coordinate many short stages than a few long ones, especially when it's operating at high speeds. To further that end, the Pentium 4 has the longest pipeline of any mainstream CPU, an incredible 20 stages (hence the term "hyperpipelining"), compared to 10 stages in the Athlon and 12 in the PIII.
SSE2: The Wild Card
Just as the Pentium MMX introduced new "MMX" instructions into the Pentium processor and the Pentium III introduced "SSE," the P4 extends those instructions by adding SSE2. The idea behind these instructions is to take common tasks that a CPU must perform (for example, common instruction sequences used during 3D graphics, DVD playback, etc) and "hard-wire" them into the CPU as a single instruction. So what used to take four separate "normal" instructions might only require one SSE2 instruction, making the processor much more efficient at those tasks.
Unfortunately, Intel made another conscious tradeoff when they designed the the SSE2 unit inside the P4, by compromising the performance of its "floating point unit" (AKA "FPU," the part of the processor that usually handles tasks such as 3D graphics). Intel is making a gamble (and a bold one, at that) that SSE/SSE2 instructions will be used extensively by software developers in future applications, since without software being designed for them, those SSE instructions are useless. That's one reason many current applications (which use some SSE, but no SSE2) show such disappointing performance on the P4. Instead of the powerful SSE2 part of the processor being used, its relatively weak floating point unit is used instead.
Fortunately, once SSE2-enabled applications are released, the P4 should begin to shine. Intel is currently claiming up to 200% the performance through SSE2 optimizations. Keep in mind however, that similar claims were made in the past about both 3DNow! and SSE; while technically true (remember the 3DNow! version of Quake II that more than doubled performance), real world performance benefits have typically been 5-20% with optimized software, with only the occasional application where it really makes a big difference. Also, none of the previous extensions (MMX, 3DNow!, SSE) have achieved anywhere near the market penetration that Intel seems to be relying on for SSE2. Not only does Intel need widespread and rapid support for SSE2 for the P4 to remain competitive, but they'd better have at least a few "killer apps" with heavy SSE2 optimizations, or the P4 will look rather anemic compared to its future competitors.
Rambus RDRAM
At this time, Pentium 4 motherboards are only compatible with Rambus RDRAM, and not the conventional SDRAM found in most systems today. This may ironically the Pentium 4's greatest advantage since the 400 mhz front side bus of Pentium 4 motherboards finally can take advantage of RDRAM's strengths, which until now have been virtually nonexistent. The irony lies in the public perception of RDRAM as one of the most despised technologies in history, in large part due to its cost, previously poor performance, and the litigious (some would say bloodsucking) nature of the Rambus Corporation. Yet now that a platform has been released on which it is not only sane, but preferable to use RDRAM, public perception of RDRAM is also at an all-time low.
In any case, the memory performance of the Pentium 4 platform is currently unparallelled, even by the fastest DDR (double-data rate) systems available. Synthetic throughput benchmarks show the P4 outperforming current SDRAM systems by a wide margin, and even outperforming DDR systems respectably.
On the other hand, any discussion of the P4's memory performance should also be tempered with the fact that applications utilizing such vast bandwidth are currently almost nil, as well as the fact that Intel is currently doing everything within their power to facilitate the release of an SDRAM-compatible P4 motherboard (whether from a competitor such as VIA, or one developed in-house). Rumor has it that Intel may even break their "unholy alliance" with Rambus Inc (which means losing out on $100+ million in Rambus stock options) so that the Pentium 4 can support both RDRAM and SDRAM (including DDR). Once this is achieved (most likely in the 2nd half of 2001) expect to see significantly cheaper DDR P4 systems, with the possibility that RDRAM will become a "niche" product shortly thereafter.
Price: Not Exactly a "Bargain"
"Intel Corp. releases its Pentium 4 desktop PC chip to stores Monday -- and at a bargain price."
-John G. Spooner, Recent ZDNet.com article
I'm curious to know how Mr. Spooner comes to this conclusion, since he writes this at the beginning of his recent P4 article at ZDNET.com, without following up with any explanation in the article itself. In fact, the article never even discusses P4 pricing besides providing some numbers (such as those below) and quoting an Intel Executive VP, who says "we're very happy with where (prices) are starting." Hmmm...
In any case, Mr. Spooner is not alone in this sentiment, if the many P4 articles I've read are any indication. It's just that I have a hard time seeing any CPU as a "bargain" when comparable (and in many ways more powerful) CPU's are already available for about half the price. The 1.5 ghz Pentium 4 is "officially" priced at $819, while its cheapest Pricewatch.com listing is $939. This is in stark comparison to the 1.2 ghz Athlon and 1 ghz PIII, which are available for $484/489 respectively. Complete P4 systems are also priced accordingly, ranging from $1,999 to $4,167.
What Mr. Spooner should have mentioned is that the P4 is not as expensive as many had expected it would be (many reports had predicted $1500+ for the CPU alone), and not as expensive as it could be. Since Intel thought PC manufacturers may be hesitant to produce P4 systems containing high-cost RDRAM, Intel decided to provide "rebates" for each P4 sold. Not rebates to the consumer, mind you, but to Dell, Gateway, HP, or whoever made the PC, to effectively bring the cost of RDRAM to the same level as SDRAM. At least in theory, the savings from those rebates would be passed-on to consumers in the form of cheaper P4 systems.
Of course, it's important to note that RAM upgrades will not benefit from these rebates, which means some expensive upgrades. A single 128 MB chip of PC800 RDRAM currently costs $183 (a full $120 more than conventional RAM) while 256 MB RDRAM chips carry even more of a premium ($319 more than an equal amount of SDRAM, $436 vs $117). Hopefully RDRAM prices will decline sharply once the P4 becomes more widespread and RDRAM ramps up into greater production. However, Rambus (and others) have promised that decline for almost a year now, and while RDRAM prices are currently at an all-time low, so is the RAM market overall. Whether RDRAM will ever truly become "only 15% more expensive than SDRAM" as Rambus Inc promises has yet to be seen.
To Intel's credit, the pricing of the P4 itself is not terribly out-of-line with Intel's past CPU releases, which have historically been in the $900+ range. Of course, it could be said that the CPU market was an entirely different beast back then, back when Intel had a commanding lead in the CPU Wars and could essentially charge whatever it wanted. These days, even such "relatively" low prices are rather hard to justify when considering the P4's performance. That's why I think characterizing the P4 as "a bargain" is making a flawed comparison; sure it's a bargain compared to past Intel CPU's like the 1 ghz PIII (which sold for $1500+ on the open market just five months ago), but the keyword here is "past". Such a comparison is irrelevant when current CPU's truly are "bargains" by comparison.
Conclusion
Although the P4 undoubtedly has incredible potential, it's tough to justify buying one so early in its lifecycle. A 1.5 ghz Pentium 4 system from Gateway currently costs over $600 more than an identically configured 1.2 ghz Athlon system (identical besides the necessary differences in motherboard and RAM), even though the Athlon is currently more powerful in most common applications, as well as high-end applications which require a powerful FPU (e.g. 3D rendering, MPEG encoding). Although it tends to get ignored (or even erroneously contradicted) in many articles I've read, where the P4 comes up short by the widest margin is these high-end applications--applications where most people would assume the P4 to excel. Again, this is likely to change once more of these applications become SSE2-optimized, but that may not happen for some time.
If you're in the market for a new PC, a 1.2 ghz Athlon--or even a 1 ghz PIII--would be the better bargain, and in many ways an even more powerful system. When (and if) SSE2-enabled software allows the P4 to overtake its competition, P4 prices will have dropped significantly, far more powerful P4 systems will be available, and you will have owned a relatively underpowered system for at least several months (and possibly much longer). According to a recent Tom's Hardware Page article, the Pentium 4 currently performs anywhere from "slower than the Pentium III" to "very badly" in all but one of their tests. These tests included MPEG4 encoding, current office applications, Unreal Tournament, Linux kernel compilations, 3D Studio Max, and MDK2. The only test where they felt the P4 performed favorably was Quake III (and in that case it was due to the RDRAM's memory throughput, and not so much the processor itself).
All in all, initial tests have shown that the P4 is neither the revolutionary Godsend many had hoped, nor is it the D.O.A. dud that some have portrayed it to be. If anything, it's a CPU that's simply ahead of its time. It may very well be the "CPU of tomorrow" in the long run; it just isn't the CPU of today.
Thanks for reading,
-vicwang
Update 4-14-01: Pentium 4 price cuts since this article was originally written have dropped its price significantly, into the $400-500 range for the 1.5 ghz version. However, since it still costs at least twice as much as an equivalent Athlon, the conclusions of this review remain the same. When (and if) the upcoming Pentium 4 price cuts move it closer into the Athlon price range, I will update this review accordingly.
UPDATE: It has just been reported that current Pentium 4 motherboards are incompatible with Voodoo 3 and Voodoo 5 graphics cards, since current P4 motherboards do NOT fully conform with the Intel AGP specification. For more details see http://www.theregister.co.uk/content/3/15118.html
Thanks to Maximum PC, Tom's Hardware Page, The Register, IXBT Hardware, Anandtech, Ace's Hardware, JC News, ZDNET, and CNET for much of the information contained in this article. Also thanks to artic00cat, wolf_knight and gr8ful for their input during this article's many revisions.
Recommended:
No
|
|
|
|
Epinions.com ID:
vicwang
|
|
Member: Vic Wang
Location: Texas
Reviews written: 45
Trusted by: 216 members
About Me:
Systems Analyst and all-around computer guru who's always keeping up with the latest technology.
|
|
|
- Add to My Yahoo!: 
- Add to Google Homepage: 
© 1999-2009 Shopping.com, Inc.