Click here to print this article.

Re-Printed From SLCentral

Intel Pentium 4 1.7GHz
Author: David Pitlyuk | Paul Mazzucco
Date Posted: August 3rd, 2001


The processor world is currently in an on-going battle between the powerhouse Intel, and its once inconceivable rival, AMD. Each company has a new generation of processors, Intel with the Pentium 4, and AMD with its Palamino.

The first thing we need to understand is Intel's competition, as well as their other processors that are available. Currently the fastest processor (In terms of clock speed) from Intel is the 1.8GHz Pentium 4. AMD currently has the 1.4GHz Athlon.

After reading that, you should have two questions in your head. One, if the 1.8 GHz chip is available, why are we reviewing the 1.7GHz chip? The answer to that question is quite simple, by the time we could finally get Intel to send us a Pentium 4 to review they only had a 1.7GHz chip available, even though we did ask for a 1.8. Not only that, but if you have ever read some of the other Pentium 4 reviews with pictures of the chip, you may have noticed that they say "Intel Confidential". This means that the processor is an engineering sample and the multiplier lock has been taken off, which isn't like that on a retail chip. Without a multiplier lock, it would have allowed us to overclock our sample, but unfortunately Intel didn't send us an engineering sample. Actually the D850GB motherboard, which Intel was nice enough to include didn't seem to have any options to change the FSB or voltage settings anyway. We hope that by the time this review is posted, Intel can set aside a 1.8GHz chip for us to take a look at, if that does happen, be sure to see a review on it soon afterwards. Also as a quick side note, according to Intel's roadmap, their 2GHz Pentium 4 is slated to be released sometime in Q4, most likely in September to be more specific.

The second question you should be thinking is how come AMD has a chip that runs only 1.4GHz, while Intel has one at 1.8GHz…Wouldn't that make Intel a clear winner? Not necessarily, today clock speed doesn't guarantee a performance winner as there are a lot more aspects to what is faster. We will talk about some of the hard numbers and comparisons between the processors later in the review when we start getting into the benchmarks.

Now that I have given you a bit of an introduction to the review and the background of where the processor world is currently at, lets start talking about the technology that goes behind the Pentium 4. Since this is our first actual processor review, and specifically of the Pentium 4, we have published a separate in-depth article today on the technology itself. The article is written by Paul Mazzucco, our processor/technology editor. He is also helping me out with a couple aspects of this review. You can check out the article by clicking here. In case you really aren't interested in going in-depth with the Pentium 4, the next page is a brief overview of what the technology behind the chip offers.

Technology Overview

Hyper Pipelined - One of the secrets to a high-performing CPU is to pipeline the execution of instructions. While this increases latencies, and has other negative side-effects, it greatly increases the throughput of a processor, by allowing it to execute issue instructions every cycle. The Pentium 4 is, by Intel, considered to have "Hyper pipelined technology." The Pentium III and Athlon, by comparison are merely "Super pipelined." All this means is that the Pentium 4 breaks the execution of the instructions into smaller pieces, so as to allow the processor to reach higher clockspeeds.

Better Branch Prediction - The longer the pipeline, the more it hurts when dealing with branches in code, because the penalty for guessing wrong is greater. On the Pentium III, the Branch Prediction Unit had an efficiency of about 90-91%. Because the Pentium 4 has a pipeline length that is twice that of the Pentium III (even more if there is a Trace Cache miss), the penalty for guessing wrong in branches is greater than on the Pentium III. To help combat this, Intel improved the Branch Prediction Unit of the Pentium 4 over that of the Pentium III.

Rapid Execution Engine - The "Arithmetic Logic Units" (ALU) run at twice the base clock speed of the processor. This means that a 1.7ghz Pentium 4, which has two double-pumped ALUs, running at 3.4ghz! The most basic, and most used instructions, add, subtract (which is mathematically addition), Logical AND, and Logical OR, can be completed in one ALU cycle, which means that they complete in the time of one-half the base frequency.

Trace Cache - The Trace Cache, as implemented in the Pentium 4, has a storage capacity of approximately 12,000 micro-ops (RISC-ish instructions that are simpler to execute, and of uniform length, unlike x86 instructions). This has the benefit of lowering the branch penalty that the Pentium 4 would otherwise have, due to the fact that in the case of a Trace Cache hit, the instruction doesn't have to be decoded all over again. The other major benefit is that code is stored contiguously in the order that it was executed.

Quad Pumped FSB (100mhz, 400megatransfers) - The Pentium 4 has bandwidth like no other x86 CPU. A major reason for the massive bandwidth is due to the fact that the Pentium 4 has a quad pumped front side bus. Though it's not as good as a real 400mhz FSB (the latencies aren't as low as a theoretical 400mhz FSB), a quad pumped FSB provides 3.2Gigabytes per second of bandwidth. When paired with 2 channels of PC800, main memory matches perfectly with the bandwidth of FSB, unlike the Pentium 3 and the i840 chipset, which has the same theoretical main memory bandwidth, but is bottlenecked by a much slower, 1Gigabyte per second front side bus.

Advanced Transfer Cache - As stated above Intel's Pentium 4 is a bandwidth hog. As such, Intel opted to beef up the L2 cache even beyond that of the remarkable "Advanced Transfer Cache" used in the Pentium III (256Kb, 256-bit wide, 8-way associative, non-and blocking - for more info, see ). The Pentium III's Advanced Transfer cache was 256bits wide, but could only transfer data every other cycle. The Advanced Transfer Cache of the Pentium 4 transfers data every cycle, thus providing twice the bandwidth per clock cycle over the Pentium III, resulting in a whopping 54Gigabytes per second of bandwidth to the core when at 1.7ghz. Even if a Pentium III managed to get up to 1.7ghz (which would be unbelievably difficult), it would only manage about 27Gigabytes per second of bandwidth from the L2 cache.

Hardware Prefetcher - Part of the reason why the Pentium 4 needs such huge amounts of main memory bandwidth is due to the fact that it implements a hardware-based prefetcher. The Pentium III introduced integer instructions that allowed a programmer to call instructions into a cache before it is actually needed. The only downside of this method is that it requires software support, which isn't always easy to come by, and it also can introduce slight code-bloat. The hardware prefetcher requires no such software support, and automatically fetches some instructions before they are needed. This uses bandwidth, and if there isn't enough, can actually cause main-memory bandwidth contention. The massive main memory bandwidth afforded by the Quad Pumped FSB allows the hardware prefetcher to basically "trade" excess bandwidth for lower average latencies.

iSSE2 - Similar in concept to MMX, 3dNow! and iSSE, iSSE2 is a SIMD technology. This means that if the same operation needs to be performed on multiple data elements, with SIMD, only one instruction is needed. The difference here is that iSSE2 provides double-precision floating point and 64-bit integer SIMD, using 8 new, 128-bit wide registers specifically for the technology. It, like its predecessor, is also fully IEEE compliant (meaning you get the results that one should expect according to a universal standard), while 3dNow! is not fully IEEE compliant (meaning the results from using 3dNow floating point operations can be ever-so slightly off).

Thermal protection - The Pentium 4 has an internal thermal diode which will throttle back the CPU if it gets exceedingly hot (such as if the fan dies on the heatsink). This protects the CPU from overheating as a Thunderbird Athlon would. Also, if the temperature exceeds 135 degrees Celsius, the CPU will shut itself down, so that it won't burn out. The Palomino core has a thermal diode as well, so future Athlons shouldn't have the same growing pains that the Thunderbird.

Should You Buy Now?

When the Pentium 4 was released it wasn't cheap, at least compared to the fastest Athlon. Not only that but the Pentium 4 can only run with Rambus memory right now. The problem with that is that Rambus is high priced, especially when compared to DDR, which is priced about the same as SDRAM. Well that was when the Pentium 4 came out in February, now, about 6 months later, prices have come down on both the Pentium 4 as well as Rambus. Not only that, but Northwood P4 should come out in September. Northwood will feature a 0.13 micron compared to Willamette's current 0.18 micron, and will also have the ability to configure your system with SDRAM or DDR memory.

Our patient

Let's take a look at current pricing for comparable processors from Intel and AMD. Taking a quick glance out our weekly cpu/memory watch you can see that the a retail 1.7GHz Pentium 4 goes for about $330 while a 1.4GHz Athlon retails for about $174. As for a memory comparison, you can get 256MB of PC800 non-ECC Rambus for $135, while 256MB of PC2100 DDR SDRAM will only cost you $42. So as a price comparison on an Athlon you would be spending $216, while a Pentium 4 would cost you $465, almost twice as much.

There's another issue as well, a Pentium 4 requires a new power supply and enclosure. The power supply needs to be ATX12V which adds a 12V power connector that goes into a Pentium 4 motherboard. Also, Pentium 4 motherboards are a bit wider, and use the extended ATX format, which is why it requires an enclosure that supports it. Not only that but there are 4 new places to put stand-offs in order to mount the heatsink to core VIA Intel's new retention mechanism.

Sounds like I am trying to bash the Pentium 4, doesn't it? Not true, these are just the numbers. In the two processors we just compared, the Pentium 4 has a 300MHz advantage, and technically an extra 134MHz FSB (4x100 compared to 2x133). Does this really mean anything though? It's all about the numbers that the processor can achieve.

The Intel D850GB

Before answering whether or not now is the right time to buy, we will need to take a look at the benchmarks and see what kind of performance we are talking about. Remember that the Pentium 4 is known to boost its performance substantially for software that is optimized for it. There isn't much out there currently, but give it some time and there will be, so that is definitely something to take into consideration as well.

On to the benchmarks...

Benchmarks Introduction

Before presenting the benchmarks, there is one issue that needs to be discussed. We had some issues getting some comparable processors in the labs, as mentioned before, this is's first processor review. In our benchmarks we only used the P4 1.7GHz and a 1.1GHz AMD Athlon (200MHz FSB). We will be forced to recommend you a few other sites benchmarks for some comparison, but make sure to keep our numbers on the Pentium 4 in mind. We will also be referring to various sites benchmarks as well. In our next processor review there should be no problems in terms of having a few comparable processors available.

Here are a few benchmarks that should be helpful to see other processors numbers. Note that the system configurations may be different and different benchmarks may have been run:

Another issue that came up had to do with 3D performance in Windows 2000. The tests that were run on any 3D performance benchmark (Eg: Quake III, Aquakmark, 3DMark) all scored very low numbers. In fact, Aquamark kept on freezing on us. Since the issue showed in just 3D performance, the first thought that comes into mind is that it is an issue with the video card, which happened to be a Visiontek GeForce 3.

We contacted Intel right away and sent over the numbers. Intel's performance engineers went over them and stated that they were getting the same sort of results over there as well, and that it was mainly because of Windows 2000's lack of gaming support. They were using a GeForce 2 in their tests with an older driver version, and said that the GeForce 2 would perform better than the GeForce 3. I thought this was a bit of an odd conclusion, considering the fact that we looked on some other sites benchmarks for 3D performance in Win2k, and their numbers seemed fine. Just to make sure, we threw a GeForce 2 into the system, and still had the same results, so we're not sure if it was a video card driver issue, a motherboard issue, or what, but we decided not to post 3D performance results in Windows 2000 until we found out.

The Test

Processor(s) Intel Pentium 4 1.7GHz AMD Athlon 1.1GHz (200)
Motherboard(s) Intel D850GB MSI KT7 Turbo (MS-6330)
Memory 256MB PC800 Samsung RDRAM 256MB PC133 Mushkin SDRAM
Hard Drive IBM Deskstar 30GB 75GXP 7200RPM ATA100
Video Card Visiontek GeForce 3 64MB DDR
CD-ROM(s) Sony CDU-5211 52X Plextor PlexWriter 12/10/32A
Ethernet Netgear FA-311 Linksys LNE100TX
Operating System(s) Windows Millenium Edition
Windows 2000 Professional Service Pack 2
Video Drivers NVIDIA Detonator 3 v12.41
VIA 4-in-1 4.32V for all VIA boards
  Benchmarking Applications
Productivity BAPCo SYSmark 2001
Winstone 2001
OfficeBench 2001
Gaming Quake III Arena v1.29f
3DMark 2000
3DMark 2001
Aquamark v22

Benchmarking System Setup

Each system had a formatted hard drive with clean installs of Windows 2000 Proffesional SP2 and Windows Millenium Edition in a dual boot format. All necessary updates to the operating system were made VIA Windows Update. Each benchmark was installed, run, and uninstalled. After each uninstallation the drive was defragmented, and the next benchmarking program was installed.

For Quake III Arena benchmarks, we used Q3Bench to automate things.

BAPCo SYSmark 2001

On MadOnions page, they had this to say about SYSmark 2001: "…SYSmark2001 accurately reflects the way users actually work! Finally there will be a resource that truly measures application performance in the same way that the real world does." While this may not be quite true, they certainly are a lot closer than running tiny integer loops in the L1 cache and calling it a "CPU test" * cough* drystones * cough*. Below is a summary of their overview, which can be found here (

SYSmark 2001 has "real-world" usage built into it. It uses scripts and timing delays to perform the tasks similar to how a real person would, performing tasks a real person would do.

Content Creation:

These first three have work saved in web format:
MS Word2k - Opens a document, adds some graphics, text, backgrounds, and prints.
MS Excel2k - Sorts data, and creates charts.
MS PowerPoint2k - Makes a presentation, formats, previews the slides as they go along, and adds a background.
MS Access2k - Loads the database and reorganizes material, processes information, and opens/prints the results.
MS Outlook2k - Searches for text in a message, archives old emails, and sends/receives email
*Netscape 6.0 - Opens a web-file, does a word search, views the source code, then views a web-based power point presentation.
Dragons NaturallySpeaking Preferredv.5 - Transcribes a *.wav file in the background.
WinZip 8.0 - compresses some videos in the background.
McAfee VirusScan 5.13 - Scanning files in the background.

* While this is a personal opinion, and while I certainly support competition, there is no good reason to use Netscape 6.0 instead of MS Internet Explorer (5.0 or greater). While Netscape is still prevalent, it has nowhere near the installed user base as Internet Explorer, and it is well known that Internet Explorer is faster to load, and faster to use. I see no good reason why Netscape 6.0 was chosen over the MS version, as they used MS products for everything else in this part of the test. In my humble opinion, it would make sense to include Internet Explorer into this group (perhaps in addition to Netscape 6.0), simply because it is the more-used browser.

Internet Content Creation:

Adobe Photoshop 6.0 - Opens a large picture, does sample filters, tests resizing/orientation, adds other effects and resaves as a jpeg, and prints.
Adobe Premier 6.0 - Makes a video, adding effects and is then compressed.
Macromedia Dreamweaver 4 - Creates web pages with flash animation and pictures
Macromedia Flas0h 5 - The user creates a movie file.
MS Media Encoder 7 - Takes a video and encodes in the background.

An overall score is given, where the baseline machine gets a score of 100, on a machine set up as follows:

Motherboard: 815EEA Intel motherboard with P09 Bios
CPU: Intel® /Pentium® III processor
Core Frequency: 800/133 MHz
Memory: 128 PC 133 CL2 SDRAM
Video/Resolution: Creative Labs GeForce Annihilator 2 with 32 MB DDR onboard memory, 1024x768 resolution, 16 bpp color, 75 Hz monitor resolution.
Video Driver: nVidia v6.31
Disk: 30 GB ATA 100 IBM DPTA-373420.
Operating System: Windows 2000 (Service Pack 1)

As a "Power user" myself, some of the descriptions above sound realistic for someone who knows how to get the most out of their machine in terms of multi-tasking to get as much done in as short a time as possible. However, the average corporate user does little more than type up some reports, browse the web, or check email. The majority of PC users do not do many of these tasks simultaneously (some, they wouldn't do at all). However, as people become more computer literate, the described workloads become more realistic.

Despite this, there are users who fit the profile given above. In such cases, the aforementioned usage stresses the memory subsystem more heavily than running only one program at a time, due to the number of applications running in the background, and due to the user switching between programs. As an example, doing such causes the L1 and L2 cache to change their contents much more frequently than running one applications at a time, stressing both memory latencies, and bandwidth (as well as, of course, the CPU itself).

Internet Content Creation

Higher is better

Higher is better

We can see here that Windows 2000 performs better on both processors, this generally occurs in most of the benchmarks that we ran. In Win2k there's about a 35% improvement in over the 1.1GHz Athlon. Make sure to take in mind that the Athlon is using a 200MHz FSB rather than a 266MHz FSB. Not only that but the memory is PC133 SDRAM rather than DDR, which can improve performance by sometimes as much as 15%.

On other sites benchmarks there is about a 10% speed difference between the P4 and a 1.4GHz Athlon. That's a pretty big difference, and Intel is usually known to do pretty good with multimedia type applications over business type applications, where AMD usually does better.

Office Productivity

Higher is better

Higher is better

Here we can see what I mean by AMD processors usually doing better in office type applications. The gap between the two processors is about 12%, over a previous gap of 35%, that should prove a lot.

Taking a look at some other sites benchmarks for other processors scores, the Athlon 1.4GHz tears the Pentium 4 apart. In fact, even lower speed Athlons like the 1.2GHz come very close to the Pentium 4 1.7GHz.


Higher is better

Higher is better

Now we know that each processor has it's strong points in certain types of applications, but overall the Pentium 4 came out on top, but not by much. Actually, taking a look at some other sites scores and averaging them out, the Pentium 4 only edged out by less than 5% in some cases. In the introduction, I mentioned clock speed isn't everything anymore, here's the proof.

Winstone 2001

Winstone is a rather "old-school" benchmark, but it does use real applications. Business Winstone 2001 is comprised of the following applications:

  1. Norton Antivirus 2000 from Symantec
  2. WinZip 7.0
  3. Microsoft FrontPage 2000
  4. Lotus Notes R5
  5. Microsoft Access 2000
  6. Microsoft Excel 2000
  7. Microsoft PowerPoint 2000
  8. Microsoft Project 98
  9. Microsoft Word 2000
  10. Netscape Communicator 4.73

Winstone has a base machine was configured as follows:

Benchmark Ziff Davis Media's Business Winstone 2001 Version 1.0
1024x768 using 24-bit color
Cirrus Logic(R) 5465 with 4096 KB RDRAM
Driver version MGA 64.sys, MGA 62.dll
800 X 600 16 bits/pixel
Refresh rate 70 Hz
Seagate Medalist ST 3654 0A hard drive
Hard disk controller Intel 82371AB/EB PCI Busmaster IDE
512Kb L2 cache
Intel Pentium 2 Processor at 333mhz
Operating system, Windows system Windows NT 4.0 Service Pack 6, Build 1381

A score of 10.0 refers to the base machine described above. However, the tests done by Winstone are not as indicative of the performance that a CPU, let alone a machine, as other tests. This is due to the fact that Winstone does not benchmark in the same, multitasking way as other benchmarks we have included (despite many similar programs that are run). This causes the scores to be more forgiving to less powerful memory subsystems, as they are not stressed nearly so much. Though, to be fair to Winstone, many users run only a couple programs at a time, and don't switch between windows as often as a power user would. However, as Winstone is a relatively common benchmark elsewhere, we opted to include it as well.

Content Creation

Higher is better

Higher is better

Here's another look at a content creation benchmark, this time from eTestingLabs. We can see here that in WinME there is a 6% performance increase with the P4, but in Win2k there is a 10% increase. 10% is where you start seeing a noticeable difference, and note that this is only a 1.1GHz Athlon we are comparing to.


Higher is better

Higher is better

We see the same sort of thing we saw in the content creation benchmark, but the Pentium 4 has just a bit more of an edge over the Athlon. When looking at a few other sites benchmarks we can see that the Athlon does start to close in on the Pentium 4 with higher clock speeds.

OfficeBench 2001

It should be noted that OfficeBench2001 runs only in MS Windows 2000 - it does not run on any DOS based MS operating system. This benchmark has some similarities with SYSmark2001, at least, insofar as they both run concurrent workloads to better simulate reality, rather than a linear script where one program is run after another. The benchmark consists of a relatively short script, which uses programs such as Internet Explorer, Power Point, and other programs to do content creation, such as creating complex tables, etc.

The baseline load simply runs the script. This script can be adjusted so as to include section delays, which help to simulate pauses that a user would use, such as switching to a different application, pausing, scrolling down, pausing, etc. However, very intensively used computers may have other programs running concurrently, such as acting as an exchange server, database accesses, and streaming video. The load-levels represent a differing number of instances of the programs running at once. For example, load-level one has one instance of the aforementioned tasks running while the base script runs, while a load-level of two has two instances of each of the task while the base script runs, etc.

The greater the number of tasks running at once, the longer it will take to complete the base-script (which is what the benchmark times), in part due to the increased number of calculations required because more instances of the various programs running, and also in part due to the context switches involved with the CPU rotating through the various active threads. Another impact is due to L1, L2, and main memory bus contention due to the great number of loads and the types of applications being run. See below to see the results from the varying load-levels.

OfficeBench 2001 is a new benchmark that was recently introduced to our world through a Pentium 4 1.8GHz review on AnandTech. After running the test ourselves, we were a bit perplexed. After checking AnandTech's scores as a comparison, we noticed that our times were significantly lower(better) than Anand's, especially at load level 2. For example, on load level 2, Anand finished in 86.13 seconds with the 1.8GHz Pentium 4, we finished in 60.36 seconds with a 1.7GHz Pentium 4. Both of our systems are configured almost identically, although Anand's system had a different motherboard, but our scores were much better with the Athlon as well.

After realizing this, I got on the phone with Randall Kennedy, creator of OfficeBench, to see if he had any ideas. We couldn't come up with anything, so I e-mailed Anand to see if how he tested or if he had an idea why the results were so off, but we never recieved a response.

We won't be able to refer you to any other sites for other CPU's benchmarks because we are unsure as to whether or not Anands scores are accurate (as we talked to Randall, and he too was stumped), and no other sites have use OfficeBench yet.

Baseline Load Level

Lower is better

With no load we can see here that the Pentium 4 completed the test 25% faster than the Athlon, but considering that the Pentium 4 is clocked 35% faster and has faster memory, is that really a big feat? This is where we really wish we could have had more CPU's in the lab to compare with.

Load Level 1

Lower is better

We notice here that the Athlon starts to move towards the Pentium 4, and looks to be handling the multitasking a bit better. Lets see if this trend continues with the load level turned up another notch.

Load Level 2

Lower is better

Looks like that trend didn't continue as the Pentium 4 takes a huge leap back around 32% faster. This is where P4's massive memory bandwidth comes into play. When load levels are this high, the processors speed gets somewhat taken out of the picture, and performance boils down to which platform can get the data to the CPU the fastest.

Load Level 3

Lower is better

The performance gap seems to stay pretty stable here, but with load level 3, it's testing a load that none of our systems would probably incur. How many of you will be running 3 exchange servers, accessing 3 databases, streaming 3 videos, and using Microsoft Office at once? Thought so :).

Quake III Arena

Quake 3 has been one of the staple benchmarks of review sites since the games inception. While it is getting a bit on the old side, there are other games which use its engine, such as American Magees "Alice," and Star Trek: Elite Forces. While benchmarking all the games would be best, sometimes triage must be performed with respect to a reviewer's time. So we've stuck with the tried and true Quake 3 to be representative of OpenGL performance, and the other great bonus is that it's popular, and many scores to compare with elsewhere.

Quake 3 also has the distinction of having the ability to change visual complexity by increasing the number of polygons that are rendered in a scene. It also has multithreaded support, though it is somewhat limited. While we didn't test the former simply because no hardly anyone (at least that I know of) uses it, the latter due to lack of a dual P4 platform, this bandwidth intensive game is also greatly affected by the speed of the processor, especially in the lower resolutions in the same way that the 3dmark tests were.


Higher is better


Higher is better


Higher is better

Quake III Arena is one of the tests that we usually see Intel perform really well in. As you can see from the charts, the Pentium 4 performs almost twice as fast at the lower resolutions. Note, that the lower resolutions are more important as they tend to use more processing power, rather than the video card. The higher the resolution, the more dependant on the video card it is. Other sites tests aren't really setup nearly the same way ours was, but you can see that the Pentium 4 really just pulls away here.


3dmark2000 is a MS DirectX 7 based benchmark. The main goal is to get a rough estimate as to how a DirectX 7 based game will run. This benchmark obviously takes the video card into consideration, however the lower resolution scores can be used for a relatively good estimate of a CPU and it's memory subsystems' performance. The greater the resolution, and the higher the color depth, the more the video card becomes the bottleneck, as it has more data to process to display on the screen. However, as the game engine used in 3dmark2000 is DX7 based, it does not make use of the pixel and vertex shaders that the GeForce3 has, so there is again less reliance on the video card.

3dmark2001, on the other hand, is based on the Max Payne game engine, (which we shall be reviewing shortly) which has a higher polygon count, and does make use of the pixel and vertex shaders on the GeForce3. The scores between 3dmark 2000 and 2001 can't be compared, however.

The most important scores here are in the lower resolutions for the reasons stated above. However, the other resolutions are included to show how even far faster processors make much smaller differences at the highest resolutions (where the video card becomes the limiting factor). We used a GeForce3, which is generally the fastest video card around to limit the impact the video card has on the benchmarks.

3DMark 2000

Higher is better

Higher is better

3DMark 2000 tests the processing power a bit more than 3DMark 2001 mainly because of its DirectX 7 usage over DirectX 8. Especially since we are using a GeForce 3, which supports technologies like pixel shading in DirectX 8. As any 3D performance test, the higher the resolution, the more the video card comes into play over the processor. We can see from the numbers that the Pentium 4 keeps a good 25% performance gap between the Athlon. Again, other sites ran 3DMark 2001, so we can't tell you how close some of the higher end Athlon's come unfortunately.

3DMark 2001

Higher is better

Higher is better

The Pentium 4 here keeps a pretty consistent 20% performance lead over the Athlon. One thing that you should note is that 3DMark 2001 supports SSE2 instructions, giving the Pentium 4 a bit of an edge. It would be interesting to see what the next generation Athlon can do, since it will support SSE instructions as well.


While 3dmark2001 is based off of a very current game engine (Max Payne reviews have already started popping up), Aquamark is a benchmark of an actual game: Aquanox. Aquanox is a DX8 game, and has extremely advanced visual features. This might lead one to believe that it should be totally video-card bound - other sites (Example here) have shown that only a GF3 can make use of its T&L engine for this demo. If this is indicative of the real game, then, on systems without a GF3, the CPU itself will limit the gaming performance. This should lead one to believe that, though the game is indeed very much graphics-card bound, even if the system has a GF3, performance should improve considerably with a faster CPU. Our results below indicate this as well.

Higher is better

Higher is better

Since AquaMark is a DirectX 8 application, it does use a lot of the GeForce's 3 power over the processors, although we left the pixel shader off so that it would not depend on the video card as much. The Pentium keeps a 30% - 35% lead here, but when looking at other sites, the Athlon totally wrecks the Pentium 4. Even a 1.2GHz does better, and the use off DDR memory helps the Athlon substantially. The one thing we do notice is that the numbers overall aren't very gapped throughout a lot of the clock speeds, which just shows that the video card is really doing a lot of the performing.


There are a few points that need to be touched on to put everything into perspective. Intel has a great chip here that performs relatively well. At higher frequencies, the Pentium 4's performance is able to distinguish itself from Intel's previous archtecture, and can easily outperform the lower end Athlons. AMD's 1.4GHz processor performs close as well, sometimes performing better than a 1.7ghz Pentium 4, sometimes slower, as other sites have shown. Note that AMD will release faster clock speeds by the end of 2001 which may close any gap that we see right now. On the other hand, Intel should be moving to 2GHz with the Pentium 4, so we'll have to see what kind of numbers each of those higher-end this generation chip boasts.

The main concern with the Pentium 4 is its price comparison to an almost as fast Athlon. Is it worth twice as much money? At this time, we can honestly say no, but there are a few paths. First of all, even though the Athlon may cost cheaper and perform better, there will a huge percentage of people, whether they are home users or corporate users, that will stick to Intel because of the name. For corporate users, even if you personally believe that the Athlon route is the way to go, your boss may want solely Intel products. In that case, in terms of performance the Pentium 4 is a great buy over an older Pentium III. Looking at it this way, the Pentium 4 is very much exceeding the performance of their highest performing .18 micron Pentium III's. If you must stick with an Intel chip for performance, there's no reason to choose a .18 micron Pentium III over a Pentium 4.

For the rest of you, our best recommendation is to wait until Northwood and Palamino come out. With the Northwood Pentium 4 there are a few things that will make a big difference from the current Pentium 4. The main thing is that, like all new architectures, the comparatively large Pentium 4 is itching for a smaller process. This means that it should scale in frequency much better in .13 compared to the Pentium III than it did on .18, and thus, scale better in terms of absolute performance. This means that the Northwood should perform a lot better. Not only this but there will be no needed for a new enclosure (although you still need a new power supply). Also, it will be socket-478 rather than the current socket-423, which is planned to die out. The Northwood should also be cheaper to make, so the price savings go directly to the users, so prices should be lower.

For the final words, the chip is great, although right now the costs are just too high compared to the Athlon, and it makes sense to wait a little while longer for either Northwood or Palamino.

Rating: 7.5/10 SystemLogistics

Re-Printed From SLCentral