Mention of the Pentium 4 is commonly met with disgust from a certain portion of techies, the Pentium III was a far superior CPU they will tell you. Tech reviews from the time seem to affirm this and the early Pentium 4’s looked to be outperformed by both AMD’s Athlon XP chips and the older Intel Pentium 3 but how bad really was the early Pentium 4?
Its actually not that uncommon for a new CPU architecture to be outdone by the last models of the previous generation. There are several examples of this in the history of CPU development. For instance the AMD 386DX-40 was faster then the early 486’s and late 486 chips like the AMD 133mhz 5×86 (a 486 in all but name) easily outshines the original Pentiums. Why then is there such distaste for the first generation of “Willamette” core Pentium 4 CPUs? When Intel developed the Pentium 4 they used a new architecture called NetBurst which differed from the P6 architecture of the Pentium II and III before it. Many felt Intel would of done better to continue to evolve the P6 architecture and that NetBurst was good for gaining higher mhz numbers (which consumers paid attention to) but not giving better performance relative to those numbers. Once the Northwood cores hit the market in 2ghz+ speeds performance of the Pentium 4 became quite good but I believe most of the hate for the Pentium 4 is centered around those early Socket 423 Willamette core P4’s and these are what I wanted to test. Were the Willamette Pentiums 4s as bad as they say?
Before we start the article proper though I want to make a point very clear. This article is ONLY focusing on the Willamette based Pentium 4 as it operated with high end era correct parts within the end of 2000 and into 2001. It is not meant to represent the overall Pentium 4 line such as later Northwood, Prescott ect… chip revisions. It is also not meant to represent the full potential of the 1.5ghz Willamette CPU, as in using overpowered GPU’s and drivers from much later time periods as not to bottleneck the CPU. One example of this would be taking the machine below and installing a Geforce 6 or 7, a GPU released far after the 1.5ghz Willamette and then seeing how it performs. Also please keep in mind software optimized for the Pentium 4 and thus taking advantage of SSE2 instructions was not widely available in 2000/2001 which is the time period we are looking at for this build.
I frequently read about how the Pentium III easily stomped the Willamette based Pentium 4 CPUs in performance and I wanted to test this myself. To this end I decided to build the ultimate year 2000 Pentium 4 machine. The Pentium 4 did come out in late 2000 but probably wasn’t readily available to consumers until 2001 but I thought it would be fun create a year 2000 specific machine using the best parts that money could buy at the time. I’d like to think maybe it could of been a very expensive high end Christmas present assembled in December of 2000.
For a case I just went with a beige white case that I felt was very representative of the time. I kept things pretty simple with bay drives and limited my build to a pretty standard 1.44mb floppy drive as well as a DVD drive manufactured in 1999. At this point CD-ROM drives would of still been very common but I went with DVD since that would of been the high end and they were widely available in 2000. I believe this drive is a x12 speed but x16 speed drives were available. For a hard drive I’m using a ATA-100 40gb Quantum Fireball AS drive from early 2000.
Motherboard – Obviously the star of this build is going to be the socket 423 motherboard. Socket 423 was the original socket for the Pentium 4 and was a very short lived socket type only being in production a very short time before the Pentium 4 moved on to socket 478. Because of this, socket 423 boards tend to be pretty hard to find these days and can command a high price on the internet. Socket 423 supported the 1.3ghz to 2.0ghz Willamette based Pentium 4s. Intel quickly realized this socket was not adequate for higher clocked CPU’s so it was ditched fairly quickly for socket 478.
The motherboard I choose was the Legend QDI Plantinix 4x board largely on the bases of its availability to me to purchase and its unconventional look with RAM placement. The QDI Plantinix motherboard features RIMM slots for RDRAM (will get to that shortly) as well as an AGP x4 slot and an Intel 850 chipset.
1 ) Usually I mark the PSU connector as an afterthought but socket 423 is actually a bit picky about its power supplies. Socket 423 boards along with a standard 20 pin and 4 pin ATX +12 volt connector requires a supply with a 6 pin AUX power connector. It kind of looks like one half of an old AT power connector and plugs in next to the 20 pin connector.
This connector is necessary to properly power the socket 423 board and CPU. It’s not a very common connector to find on power supplies but I had the most luck searching under “Pentium 4 PSU” in eBay searches. These power supplies will all be fairly old at this point like mine and I had to go through 2 supplies to get a working unit. I failed to find an adapter to add an AUX connection to a more modern supply but they may exist.
2 ) CPU – Socket 423 supported all the Willamette CPUs which were produced in speeds of 1.3ghz to 2ghz. In 2000 only two Pentium 4s were available, the 1.4ghz and 1.5ghz version. The slowest 1.3ghz CPU interestingly was not released until very early January 2001. Originally I used a 1.3ghz CPU as seen in the image below just to see how slow the slowest P4 actually was but later decided to upgrade to the 1.5ghz model after discovering the performance difference in games was only 1-3 FPS in most cases and to make a truer year 2000 machine.
Except for the 200mhz speed difference and indication of clock speed on the chip the 1.3ghz CPU shown above and the 1.5ghz CPU installed on my motherboard look exactly the same. All Willamette P4s have 256kb of full speed L2 cache on die which is the same as most of the Pentium III CPUs (exceptions being some of the Tualatin models and the early slot 1 Pentium III’s which had 512kb of half speed L2 cache). Later the L2 cache would be bumped up to 512kb on the Northwoods and even higher on later model P4 chips.
Socket 423 CPUs are rather large coming in a little bigger then a Coppermine Pentium III and much bigger then the later Northwood P4 as seen below
(CPU’s from left to right, Coppermine Pentium III, Willamette Pentium 4, Northwood Pentium 4)
3 ) RAM – Most socket 423 boards also used an unusual and (in the realm of consumer PC motherboards) short lived RAM type known as RDRAM or Rambus DRAM. RDRAM was expected to be the next PC memory standard replacing SDRAM but was eventually beaten by DDR memory. In the days of the early Pentium 4 though Intel had licensed the use of RDRAM with its chipsets so RIMM slots (Rambus in-line Memory Module) showed up on many socket 423 boards. Also to note is that even though RDRAM is primarily associated with early Pentium 4 motherboards it can also be found on a few Pentium III PC’s such as the Dell Dimension XPS B733r.
RDRAM was both very expensive and ran quite hot. It was so pricey that Intel had to subsidies it and include it with certain motherboards. It also required a heat spreader be attached due to the heat it produced. It was found that the relatively small performance increase in some areas did not justify the cost and RDRAM was quickly replaced in the coming years by DDR which in general was faster, cheaper and ran cooler.
The Legend QDI Plantinix 4x board supports up to 2GB of RDRAM but since I’m going to be running Windows 98se I’m only going to a max of 512mb via four PC800 128mb sticks.
The Legend QDI Plantinix 4x has one of the most unusual RAM slot orientations I’ve seen in a post 1980’s motherboard with two slots being in a typical close side by side configuration close to the CPU and another slot spaced further away with a fourth slot being completely perpendicular to the other three. I’m not sure why they went with this setup other then maybe as a space saving feature. This board is also very picky about RAM as well as placement and sizes. RDRAM requires it be installed in sets of two and any unused slots require a “CRIMM” or dummy RAM be inserted as a terminator.
4 ) CNR Slot – CNR or Communications and Networking Riser was another Pentium 4 era slot that was quickly phased out of personnel computers. It was primarily intended for networking and audio cards.
5) IDE – Two ATA-100 IDE connectors for connecting IDE devices (two devices each) like hard drives and CD/DVD ROM drives.
6) Floppy – standard floppy drive connector.
Now for my sound and video card selections for the top of the line in 2000.
Sound – For a sound card I went with the Creative Sound Blaster Live! which was sort of the default sound card of the time.
The card I have installed in the Value version manufactured in 2000 but other then lacking an extended I/O connector and having color coded ports as opposed to gold is identical to the regular version. The Live! cards also supported EAX (Environmental Audio Extensions) 1.0 and 2.0 which many games of the time supported. An Aureal based card would of been another sound option but I feel the Live! cards are a little more compatible feature wise and more representative of 2000.
Video – Finally we have the graphics card. This was actually a very easy decision as the top dog of 2000 was easily the Geforce 2 Ultra.
The AGP x4 Geforce 2 Ultra was pretty much unanimously sighted as the most powerful graphics card of 2000 and in some circumstances even proved faster then the initial Geforce 3 card that preceded it. The Ultra came with 32mb of video memory and was clocked higher then the base model GF2 GTS and GF2 Pro beating out both of the competing Voodoo 5500 and Radeon DDR cards.
One thing to note is that finding a Geforce 2 Ultra can sometimes be a challenge since they can be sought after but also because certain other Geforce cards can almost look identical.
So with that out of the way lets look at some benchmarks to see if the Willamette Pentium 4 really does drag this beast of a year 2000 machine down. To compare I wanted to use a fairly contemporary machine so I choose my Dell Dimension 4100 from 2001 as the competition. This PC will be running the exact same video and sound card as the Pentium 4 PC. The CPU I have installed is a 1ghz Coppermine Pentium III released in 2000 and also using the old P6 CPU architecture.
Both machines are also using the same video drivers (version 45.23) as well as Power Strip 2.78 in order to disable Vsync. Driver version 45.23 are later drivers from around 2003 but unfortunately drivers from 2000 were giving me direct x errors and to be honest I didn’t feel like dealing with it. differences should be minor. Both machines are also running Windows 98se and using 512MB of RAM. I am also using “Optimal default” BIOS settings on both machines and have double checked to make sure L1 and L2 cache is enabled.
*UPDATE 2/7/18*
As I discovered my previous tests skewed the results in the Pentium 3’s favor I have redone all tests. Previous tests were done WITHOUT updated Chipset drivers (from 2000/2001) which greatly effected performance of the Pentium 4 therefore I have decided to rewrite the majority of the following section with the updated data. Chipset drivers ver. 3.20.1008 were installed on both the Pentium 4 and 3 boards for all of the following tests.
First test is with SisSoft 99 to benchmark both the SDRAM in the Pentium 3 machine and the RDRAM in the Pentium 4 PC.
Pentium 3 SDRAM – CPU = 296, FPU = 328
Pentium 4 RDRAM – CPU = 993, FPU = 219
Now SisSoft 99 is a little confusing on how it benchmarks the memory and it seperates it into a CPU and FPU score but the RDRAM clearly has a bandwidth advantage here as its “CPU” score is about triple that of the SDRAM in the P3
The next test I wanted to run was just a few synthetic benchmarks just to get a feel of the two systems. The benchmarks I ran were 3DMark 2000 and 20001se. All tests were run at a resolution of 1024 x 768 with 32 bit color. This was a high but reasonable resolution for the time.
Initial Benchmarks put the Pentium 4 ahead with a slight lead in 3D Mark 2000 and with a slightly larger lead in 3D Mark 2001se. Not looking good for the Pentium 3 so far.
Lastly I wanted to run some gaming benchmarks as I feel they give a better idea of performance so I ran seven different game benchmarks multiple times on each machine to get a average. I ran all these at 1024 x 768 with 32 bit color. I also used 3d acceleration where I could since these tests were of the systems as a whole and not just the CPU. With this in mind I also performed all benchmarks with sound enabled where the option existed. All results were rounded to the nearest whole number.
In three of the seven games the Pentium 4 beats the Pentium 3 if only just barely in some cases. In the more demanding Comanche 4 and Serious Sam: First Encounter we basically have a tie due to margin of error. Both of the older Quake games also results in a dead even tie.
When run at a lower resolution of 640 x 480 the Pentium 4 did pull ahead more so in several of the tests.
Keep in mind results may vary depending on the motherboard used. Now in truth the Willamette based Pentium 4’s may not deserve the hate they tend to get. In my experience the chips have been fairly stable and do run basic tasks just fine. They also when paired with a powerful graphics card of the time like the Geforce 2 tend to run games of the early 2000’s just fine on moderate settings or even high settings and resolutions depending on the game. The Pentium III on the other hand just tends to do tasks a little worse. Though results may be quite improved with Tualatin based Pentium III’s with double the L2 cache and 400mhz more clock speed then my 1ghz Coppermine CPU.
One important thing that is worth noting about the early Pentium 4’s is that software in the early 2000’s had yet to take advantage of the P4’s SSE2 instructions and thus in the early days of the 2000’s software was not optimized for the Pentium 4 CPU. There is also the matter of controversy of the time around Intel fudging the benchmark results to favor the Pentium 4 over the Athlon and Pentium III which eventually led to a class action lawsuit though these admittedly underhanded doings did not seem to involve game benchmarks. In the end though there is a reason Intel quickly abandoned socket 423.
Other games tested were GTA III, Halo and Far Cry. GTA III ran at a pretty consistent 30+ FPS when resolution was brought down to 800 x 600 x 32. Far Cry ran more or less fine at 1024 x 768 on medium settings at around 30 FPS with dips in the mid to low 20’s when several enemies were on screen or there was a large explosion. Turning the Resolution down to 800 x 600 x 32 gave better results with FPS hitting as high as the 90’s at points and rarely dipping into the mid 20’s when a lot was going on. Halo seemed to be playable at 1024 x 768 with medium settings but suffered lower FPS rates but at 800 x 600 with textures and particle effects set on high it was perfectly playable. Quake III even at 1024 x 768 x 32 pulled a consistent 90+ FPS. A more powerful graphics card will obviously improve results but is out of the scope of a “year 2000 build”.
Despite the results favoring the Pentium 4 you were likely better off either sticking to your Pentium III or going the AMD route with the Athlon XP especially considering the high price of the early P4s. I also would of felt pretty burned when Intel dropped socket 423 for socket 478 after only a year give or take leaving socket 423 owners with not much of an upgrade path. I don’t feel the Pentium 4 really became competitive until the later Northwood cores started clocking over 2ghz and especially with the 2.5ghz hyperthreading models. There IS a noticeable performance boost over the 1ghz P3 but it’s not quite as much as I would expect from 500mhz more clock speed and a brand new CPU architecture and I think this is what let most people down. It’s not that the Pentium 4 was slower overall then a similarly equipped Pentium 3 build (if your PC was properly optimized) but that it wasn’t that much faster, at least not enough to justify the costs. I’d certainly be interested in how much difference a 2ghz Willamette makes against the Pentium III or even an early Northwood against a 1ghz Pentium III but that’s a matter for another article.
ancientelectronics 
18 Comments
Nice article. It’s quite surprising how much slower the P4 is in so many of the tests, it certainly wouldn’t have justified the high price people paid for the early P4 systems. You’d think with faster memory and a higher clock rate it should easily beat the P3, I wonder where the bottleneck in the early P4s was. Is it perhaps that the software was better optimised for the P3?
From my understanding it comes down to the NetBurst architecture of the P4 as well as the RDRAM not really being any better then SDRAM. NetBurst was able to hit higher clock speeds then the old P6 architecture but it seems it just wasn’t as fast clock for clock that’s why we didn’t see the p4 surpass the P3 until it started hitting 2+ghz and then having its L2 cache upped. In the future Id like to see how a 2ghz Willamette and 2ghz Northwood stack up.
I absolutely doubt that this test was done using optimal settings on the motherboard and/or BIOS. You people can easily look up a number of system benchmarks from 2000/2001 and find out that the Pentium 4 was doing considerably *better* than the 1Ghz Pentium III in the overwhelming majority of tests. Your speculations about something you have clearly done very little research on isn’t going to change the fact that any new cutting-edge processor architecture ever introduced was poorly supported software-wise out of the gate. Even someone like you should know about the lack of applications accounting for the SSE2 instruction set early in the processor’s life. That’s the only explanation for why the P4 underperformed in places it shouldn’t. It’s highly doubtful a 0.5Ghz increase in speed was the only thing that made the P4 a force to be reckoned with.
The answer is: the P4 had no bottlenecks, but the people coding x86 applications certainly had a few.
why do you feel you have to express your criticism in a somewhat dickish manner? I report what the tests tell me. I preformed them several times and checked everything over. This was also done with BIOS optimal default settings on both machines. This test was done from the perspective of the average user, not someone that would take the machine home and tweak all the BIOS settings. I confirmed all cache settings were enabled but other then that, optimal default settings on both machines. Perhaps the motherboard itself is to blame as its not one of the better performing brands perhaps? I also did mention at the end of the article SSE2 was not implemented on these early games and that is a reason some of these results could be as low as they are, perhaps you missed that. That said I wasn’t testing the Pentium 4 as if it was 2002 I was testing it as if it was 2000.
I plan to eventually do more tests using different cards and settings so please if you have any suggestions on benchmarks to run or settings to double check please let me know.
Products are not released into a vacuuum, they need to perform well with existing applications whether optimised for the new architecture or not. The games tested were ones that people would have reasonably wanted to run at high FPS at the time. The benchmarks used fit the same description. The new design failed to deliver even equal performance with higher clock rates and power consumption on the same applications. Something most users would see as inferior, regardless of performance on software not yet released at the time.
Of course eventually the software and hardware caught up and the better performance finally emerged, but the P4 still had issues that put people off. Especially after the release of the AMD 64bit CPUs. The P4’s were exceptionally expensive, especially with expensive memory and mainboards while Rambus RAM was a thing. They were power hungry, which only got significantly worse as they got faster, which lead to noisy machines due to the required heat dissipation. They also required significantly more servicing to remain reliable. Every summer I would see quite a number of various P4 based machines requiring cleaning and new thermal paste just to continue operation, other architectures typically suffered this less.
The P4 architecture was not further developed after it was replaced by the core architecture which had it’s roots in the Pentium 3 and Pentium M (a variant of the Pentium 3). If it was so good why did they abandon it in favour of developing a older design? It turns out it wasn’t so cutting edge after all.
Justin, I’m not here to give you a hard time. I understand that assembling older components and making it run optimally can be a tricky ordeal. My SSE2 statement was not specifically aimed at you, but simply a general side note for anyone eager to jump head-first to conclusions.
Point taken but I just want to clarify this article was only meant to focus on the Willamette core chips and only focusing on the last month of 2000 and the year 2001. It wasn’t about Northwoods or Prescotts or running a P4 in 2003 or 04 when more software using SSE2 was available. And again it was meant from the point of view of an average user who got the PC home, hooked it up and did not play with the BIOS settings. As for the benchmarks results of the period favoring the P4 what do you even know what to trust with all the controversy surrounding Intel supposedly doctoring benchmarks to favor the P4 early on. I don’t have it out for Intel or the P4 and I was honestly surprised my P4 did so poorly against the P3 hence why I made sure I double checked everything and ran the tests so many times. I believe I even made a point to say that the Willamette did acceptable for most games of the times and earlier and said I felt the later Northwoods were pretty decent CPU’s for their time. If I had multiple socket 423 boards to run the tests on I would but this is the only one I have so I have to trust the results to some degree. If you think the results are inaccurate even for running “optimal default settings” please let me know what I could change or check and I’d be happy to do it and rerun the benchmarks. For the record I also played a few other games on this P4 such as Halo and GTA III and both ran very poorly at moderate to high settings. Now both of these games I believe were released after 2000, I think Halo may of been 2003? but you would expect a high end PC from only 2 or so years prior to at least run it half decently but this didn’t seem to be the case. Again I was surprised and truly expected the P4 board to do better but if you have a possible explanation or theory of what’s holding this particular setup back (within the parameters of this being a year 2000 build) please let me know.
@sparcie It was cutting-edge for the period of time when new. Can you really blame the component if the software written during this age did not account for the higher memory bandwidth and SSE2 features? There is a reason why Quake3 loved the P4 and UT99 did not. The other was built on top of an ageing framework/engine which didn’t exploit either one of the P4’s technical strengths. This kind is issue has had a long history and given how many architectural rules were changed with NetBurst it’s not exactly a surprise then that it was going to face some obstacles never intended for it to quickly breeze past. The problem is that you’re still hellbent on the belief that the PIII was somehow able to outdo with the P4 when in reality there is too much data out there gathered through history to prove otherwise. Even the early Tualatins falls short in that regard, and only in the least favorable conditions did the P4 fall behind.
Power consumption has always been a problem with newly introduced high-end parts. I guess you decided to not pay attention to any of the occasions when the P4 received a refresh cool enough to be overclockable. It took over 6 years for the P4 to be replaced, which was a record compared to any of Intel’s previous designs so I wouldn’t be so hasty as to call it “abandoned”.
Actually yes I can blame an architecture for not running then current software even as well as what were current chips. So did most of the tech reviewers of the time. Intel doesn’t live in the dark, they know how their chips are being used and what for. They likely knew that early P4s were under performing on what was current software well before they even released them. This is likely the reason they discontinued P3 chips sooner. Clock rate and memory bandwidth are misleading figures, as most instructions take more than a clock cycle to complete, and can use varying amounts of memory bandwidth depending on many factors. It’s likely that the instructions shared by both architectures took significantly more cycles on the early P4 (also negating any memory speed benefit it might have had), otherwise it wouldn’t have had worse results in equal tests. And lets face it they are significantly worse.
Of course the deficiency didn’t last, as software providers updated and Intel made the netburst faster. But that doesn’t change the initial poor performance.
If there’s so much data that shows the supremacy of the early P4 over the late P3 kindly share it instead of making unverifiable claims, Justin did some proper testing, on a level play field as much as possible. So I think his results are fair for the time period of the hardware and software he aimed for. I’m not sure why Quake 3 would like the P4, it was released before it by at least a year, in fact at a similar time to UT99.
The power problems got worse and worse as the P4 got faster (even after the refresher, they just kept pushing the limits) to the point where Prescott cores always were noisy beasts. Sure the late ones were fast, but they were also unsuitable for many applications including media centre PCs. I used to work servicing them and building them. As I said the P4 has a poor service history in terms of reliability, they regularly needed maintenance because of their high power consumption in a way other chips never needed.
Part of the reason the P4 held on a bit longer was because they abandoned (wikipedia’s wording) the netburst architecture in the next generation. This was unusual as most other chips have a successor based on their design. It took them that much longer to develop the Core 2 architecture because they had much more R&D to do on the P3 to increase it’s performance as it is a much larger leap in terms of clock speed and features. There are no chips which have developed the netburst architecture since the end of the P4.
@sparcie
(1) The reason why some of these results were not in P4’s favor is twofold and fairly easy to comprehend: Firstly, most of the applications were not as bandwidth-dependent under the hood (Quake 3 ran very fast on a P4 because of this, as addressed by both Tom’s Hardware and Anandtech – among others – circa 2000/2001) and, secondly, next to no software accounting for P4-specific instructions/functions.
(2) You ask me to provide the data, but it’s easily available under your fingertips. For example, try looking for *1.5Ghz Pentium 4 tom’s hardware* or * […] anandtech*, but you could also type in *Tualatin [insert tech website]* or perhaps one referring to the later P4 models in 2001 for an overview between between the different clocked samples (*Pentium 4 1.8Ghz [tech site here]*) and how they fare against the competition. These were not some nobodies in the world of tech websites in its day and I doubt any of them had any reason to glorify Intel as they were indeed not shy to point out the worst-case results for the P4.
All I know is that Justin’s tests are held back somewhere – not sure where – as it’s frankly absurd to assume the P4 (even at 1.5Ghz) would be performing this horrendously in GTA3 and Halo. Ask yourself this: why would GTA Vice City’s system requirements ask for a PIII as minimum and an “Intel Pentium IV or Athlon XP processor” (i.e. ANY model of P4) as recommended if the people Q&A testing this game thought the P4 was a slower performer? It doesn’t stack up, at all!
(3) Sticking to the later P4 models was indeed an option, but why bother? I never once said that I adore NetBurst or that the P4 never got worse over time (by late 2003 you would be insane not to grab an Athlon 64). The thing is, even with a 2.0Ghz P4 from 2001 you would be surprised how far it will get you before you have to replace it with something faster (even TES4: Oblivion and Hitman: Blood Money run fine). It was a great way to future-proof yourself back in the day and by sticking to the die shrunk Northwood of early 2002 you had an even cooler/colder part on top of that. What was ironic back in ’01 was that the Athlon Thunderbird was the one suffering the most from heat-dissipation, not the P4. AMD was certainly no stranger to that limitation either.
(4) NetBurst was an attempt to try something new and it ended up becoming a failure in the long run. That much is true. Within its first 2 years, however, the P4 itself was far from a failing product. RDRAM ended up costing Intel dearly, but it was a long-running business deal that they could not realistically cut themselves off from as Rambus were contracted to make memory for a few years. It was a bad gamble, even Intel admitted that, but SDRAM-supported motherboards already arrived by 2001 for the P4.
It took a lot of Frankenstein-surgeries to adopt P6 for a new age so I’m not even sure one could call Intel Core direct relatives to the Pentium 3. Certainly, P6 was a great foundation (you might as well call me a Pentium Pro fanboy too) and they had no reason to scrap what already worked well by principle of cycle efficiency.
Having read some articles on Toms hardware, they’ve said that the early P4’s “struggled” to out perform the P3. This seems to be the general consensus. The main reason the P4 eventually got better is because it was at the beginning of it’s life and was destined for higher clock rates.
Tom’s hardware review in 2000 compared the early one to a P3 @ 1Ghz (not the fastest one) and the 1.5Ghz P4 still under performed it in a few of their tests. When run at the same clock rate, the P3 was better than the P4. Some tests showed better results for the P4 (such as Quake 3) but it certainly wasn’t a win for the P4 at the time. Given their results, I’d say these test results are reasonable, and that tinkering with the board/settings won’t improve the P4 results much. Later models of the P4 aren’t relevant as that wasn’t the aim of the test, which was to compare the first P4’s with late P3’s. Bare in mind that Tom’s hardware and Justin have not used the fastest P3.
I found all P4 models (except Celerons for obvious reasons) had power consumption related reliability issues. Basically as they all ran hotter, they all ran their fans faster, clogged heatsinks faster and dried up the thermal paste faster. Every summer I’d see numerous P4’s of all varieties require servicing, whilst other chips required far less servicing, newer ones requiring none at all.
That’s not to say the P4’s weren’t good performers, as some certainly were like the northwoods, but getting the performance came with a power and reliability trade off.
System requirements of games are irrelevant, they don’t tell you anything about how well something will work, sometimes the minimum worked great, sometimes it was awful.
Yes the Athlon XP (thunderbird etc) did consume lots of power, but for some reason they didn’t end up in the workshop every summer requiring service. It was also during this period where AMD held the performance advantage, the same Tom’s hardware review showed the contemporary Athlon doing better than Intels chips for most tests.
The fact the P6 architecture required a lot of work to make the Core 2 architecture is actually proof of how flawed netburst was. Intel went back to an much older tech as the basis to begin work, and abandoned the netburst design.
I can see that you’re somehow personally offended that the P4 should under-perform in any way. So I won’t trouble you any further.
@sparcie: I’m not sure where you’re getting the idea that I’m somehow “offended” and in need of not being troubled. Do I sound troubled to you when I clearly acknowledge the strengths of other architectures where applicable (early P6 & AMD64 etc.)? No. I’m trying to offer another perspective compared to the ‘P3 vs. P4’ misconceptions being tossed around liberally and I will have no one brush me aside pretending that I’m looking to end a currently ongoing discussion for being emotionally attached to one line of processors. I ADMITTED in my last response that NetBurst was flawed in the long term. How does this make me the least bit partisan? I don’t have to lap up any wide-sweeping assumptions as gospel just to keep this conversation civil. I do my best to present arguments backed by verified data and excuse me for not having a P4 system currently available of my own to prove any of my sources undeniable. It will change soon.
More importantly, I’m wondering why you would even be defending the Tualatin P3 parts when they were more (obscenely) expensive than even the fastest available P4 models in 2001. Anand Lal Shimpi himself made the observation that the P3 had no future and would be shelved out in favor of DDR SDRAM P4 systems despite its showings in the test bed [https://www.anandtech.com/show/804/12]. This is not to say the P4 wasn’t on the somewhat expensive side either, but given the stable performance it offers in a longer span of time over its P6 and K7 contemporaries (2000/2001 to 2005/2006 with occasional 3D card upgrades) and its advantageous areas where all other parts fall flat I wouldn’t say it’s bad value.
What I was saying regarding system requirements had nothing to do with under- or overestimated minimum performance targets – yes, XIII is a game that I have been able to successfully run and play on a P2 400Mhz while UT2003 struggles on my 733Mhz P3 system (I own a P3 computer; imagine that!) despite its minimums asking for an extra 33Mhz (XIII: 700Mhz, UT2K3: 733Mhz). No, instead I’m not going to repeat myself and ask you once again to consider the GTA:VC as well as Thief3, SC:CT and MP2 cases I already mentioned in my latest reply to Justin. What would any of these companies gain from lying about the P4’s gaming performance in contrast to the P3?
Why did the Athlon outperform the P4 in some tests? As said by the major tech sites of the day, the FPU in the Athlon was better than the P3 and P4. On the other hand, the performance tests were clearly not tailored to the strengths of the P4. I repeat, the P4 did not work the same way as more conventional CPU designs and as such code had to adapt to its idiosyncrasies. Consider the Anandtech Lightwave rendering test from August 2002 [https://www.anandtech.com/show/977/7]. How does the faster AND the slower P4s compete so well with the best Athlon XP parts? The newer versions of 3D Studio Max and Lightwave accounted for its pros. It’s safe to say the Willamettes would fair quite a bit better against its Thunderbird peers with these added tweaks.
I’m going to take the servicing argument with a grain of salt and to prove I’m not in denial I have my reasons: it’s from the perspective of one individual in one portion of a region within a whole country. The distribution of P4 vs. Athlon machines has to be taken into account. Was the P4 a greater seller than the Athlon? Was the P4 employed in more stress-inducing environments that would justify the occurrence operational obstacles? Were they self-built systems with insufficient or incorrect cooling solutions? You will have to pardon me for having doubts, especially seeing how much trickier it is to find widely available statistics related to this issue.
well, to play the devils advocate a little bit system requirements aren’t always the best metric and they can be inaccurate. Anyways I acquired another socket 423 motherboard and provided it gets here in working order I’m going to rerun the tests on it and see if the results vary.
@justinwl True, system requirements have a tendency to exaggerate things by under- or overestimating how well a game might run on specific processors, but it’s extremely uncommon (if a thing at all) for them to swap the minimum and recommended categories completely. Q&A would have to KNOW that the P4 (even at 1.5Ghz) is a better performing part for them to consider printing these details onto the box. Keep in mind that multiple game companies would have to screw up quite royally for that to occur. Thief: Deadly Shadows (and DE:IW by extension), Splinter Cell: Chaos Theory and Max Payne 2 also state a minimum requirement of PIII and recommends a Pentium IV (or Athlon XP).
On another, but far more interesting note, RandomGaminginHD’s video titled ‘The $2200, 17 Year Old Dell Desktop | A Retrospective’ showcases a 1.3Ghz (!) P4 computer with a GeForce 2 MX 400 capable of playing GTA: SA and Far Cry at a surprisingly good speed. Comparing your test results to other people’s video recordings of Pentium 4 footage it shouldn’t come as a shock then that some component in your latest test may be in need of replacing, provided of course that you’re not able to get similar results in the above mentioned games on your system.
he was playing GTA:SA at 640 x 480 on low and he said it felt more or less like the ps2 version I was playing GTA III at a default graphical settings at 1360 x 768. its quite a difference. Halo was at 1028 x 768 with all settings on high he played Farcry at 800 x 600. Throw in the fact I was also running FRAPS which probably hurt things a small degree. When Halo was set on the lowest settings and to 800 x 600 I was getting 25 – 30 FPS but for all I know Farcry is better optimized or different video drivers may make a difference. If I can find time in the next few days I’m going to retry GTA III (or GTS:SA and Farcry if I can) and see if I can match his results
You should really do a 2001 comparison with a 1.5 ghz Tualatin, a 1.53 ghz “Palomino” Athlon XP and a 1.6 ghz Willamette P4.
It would be a more accurate comparison of the end of life P6 architecture with the bleeding edge P4 stuff.
I’m a little surprised to see anybody defending the 423 based Willamette core P4’s.
Most people building machines around the time, including myself, avoided them because they were clearly a dead end both in terms of socket architecture and memory.
I’m not saying it was a good CPU, just not as horrible as its been said to be, at least in the aspect of being faster then the PIII. The problem was IMO for being a hyped new CPU with higher clocks and new architecture it wasnt much faster and sometimes struggled to be faster then the PIII and for the price they wanted for it wasn’t worth it that was hard to swallow.
In the end I kept this build pretty strickly a “year 2000” build as far as parts and what it went up against but I like your idea so maybe I will put together a “Ultimate year 2001” machine based on this P4 motherboard.
Thanks for this article. I recently got one of these S423 PC’s and wanted an idea of it’s true performance.