My friend told me that earlier iterations of the Pentium 4 were not as fast as the Pentium III, but he didn't have sources.

Is it true that the Pentium III was faster than its successor Pentium 4?

  • 7
    You might be interested in reading a review of the first Pentium 4 systems vs Pentium III and the AMD Thunderbird anandtech.com/show/661
    – Bert
    May 23 '20 at 14:27
  • 11
    Note that Pentium 4, unlike previous "Pentium-numbers", is written with an Arabic 4, not Roman IV. See e.g. a photo of the chip. Compare this with a photo of Pentium III, where the number is (as is usual for Pentium III) stylized as "!!!".
    – Ruslan
    May 23 '20 at 18:13

Clock to clock Tualatin Pentium 3 was a lot faster (i.e. a P3 got more done in each clock cycle than a P4).

High-end Pentium 3 is available at 1.4 GHz while low-end Pentium 4 is 1.5 GHz so if you compare these extremes P3 is the clear winner.

But as time went on, Northwood Pentium 4 1.8 GHz plus DDR 266 memory became common quickly while Intel constrained Tualatin P3 in the Celeron form with SDR SDRAM so in the field most P4 computers were faster than most P3 computers. High-frequency Pentium 3 was never widely available and its 133MHz FSB speed starts to show its limitations when facing DDR and RAMBUS. But early adopters of Socket 423 1.5GHz Pentium 4 and SDR SDRAM clearly made a mistake: high cost, bad performance, excessive power consumption and heat, and no viable upgrade path.

Ironically P3 continued to haunt P4 through its entire lifetime: around the middle, Intel revived P3 in the form of Pentium M for mobile with good results, and so much so that, in the end, Intel brought it to the desktop in the form of Core Solo / Core Duo to end the P4 era entirely. In fact, even before that, motherboard OEMs had made boards with mobile CPUs that beat both the desktop offerings from AMD and Intel when properly overclocked.

  • 4
    As far as I recall, the original Core (Solo|Duo) was the 32-bit Pentium M architecture. Core 2 was 64-bit. May 23 '20 at 12:40
  • 4
    @iBug: Conroe / Merom (Core 2) are the first gen of the "Core microarchitecture": 64-bit, 4-wide superscalar up from 3 uops / clock in 32-bit Core 1 (P6 uarch). But Pentium-M / Core 1 have some microarchitectural changes vs. Pentium III, like maybe a longer pipeline. As Agner Fog's microarch guide says about P-M / Core 1: Several minor modifications have been made, but the overall functioning is almost identical to the PPro pipeline. May 23 '20 at 16:14
  • 4
    It's not exactly "ironic" that P3 haunted P4; P4 was simply not a good design with several design ideas that were widely recognized in hindsight to have been mistakes and not work as well as hoped (e.g. trace cache). When transistor shrinks hit the power wall, so P4 couldn't scale to 5GHz or above as they'd expected, the only thing it had going for it (high clocks) stopped going. Modern Microprocessors A 90-Minute Guide! explains some of this: P4 was the epitome of a "speed demon" design with lower IPC than the previous generation P3. May 23 '20 at 16:20
  • 6
    @iBug: And BTW, Intel's naming schemes were pretty confusing: "Intel Core" was/is the marketing name for their desktop chips starting with Core Solo/Duo (based on an "enhanced Pentium-M" uarch), continuing to modern "Core i7" and so on. But the "Core microarchitecture" is the name for the microarchitecture inside Core 2 Duo/Quad (starting with Merom/Conroe), Intel's first 64-bit CPUs in the P6 family of uarches. (There were some 64-bit P4 CPUs). May 23 '20 at 18:22
  • 8
    @BlueRaja-DannyPflughoeft Because different CPU architectures get different amounts of work done per clock cycle. Clock-for-clock, Pentium 3 was vastly faster than P4. Many instructions on the P4 took more clock cycles to execute than the same instructions on the P3. The most egregious case of this was in bit-shift instructions; from at least the 486 there was a specialized circuit called a barrel shifter that could do arbitrary shifts all at once, while the P4 had to shift one bit position at a time. May 23 '20 at 23:06

The very earliest, slowest-clocked, Pentium 4 chips were slower than the fastest Pentium IIIs of the time. The PIII was available at 1GHz when the Pentium 4 was introduced, and a 1.5GHz P4 was about the same practical speed: it varied a bit, according to what you were running.

However, the P4 was also available in 1.3GHz and 1.4GHz clock-speeds, and those were definitely slower than a 1GHz PIII.

To get the best out of a P4, you had to compile specifically for it, using the SSE2 floating-point instructions, and a compiler that generated P4-optimized code sequences. This would not make a 1.5GHz P4 as fast as a hypothetical 1.5GHz PIII, but it did improve performance. In the case of the software I was working on, it added about 25% to throughput.

Source: Benchmarking new P4 chips at the time, and having to port the product I was working on to Intel's C/C++ compiler to use SSE2.

  • 2
    Clock speed is not CPU speed. Do you have the instructions-per-second numbers for the two generations? May 23 '20 at 12:33
  • 4
    @another-dave No, just experience of the practical throughout. May 23 '20 at 12:34
  • 1
    Clock speed is not CPU speed, neither of which have anything to do software written for specific hardware classified by a name, none of which tell you anything about practical throughput. +1
    – Mazura
    May 23 '20 at 17:07

The answer is sort of. Intel also went with a different memory subsystem, RDRAM, which had a big influence on system performance.

I'll be basing my answer on the AnandTech review of the Pentium III vs Pentium 4 vs AMD Thunderbird.

Let's start out by looking at performance in a popular game of the time, Quake III Arena. You can see the Pentium 4 is the clear performance leader, but notice that the chips run at 1.4 and 1.5GHz, compared to just 1GHz of the Pentium III. The Pentium 4 is around 25% faster.

enter image description here

Next we'll look at another game, Expendable, and you'll notice the Pentium 4 is now the slowest in the test! This is due to the way this application uses memory.

enter image description here

SSE2 instructions were also introduced which gave the Pentium 4 a performance boost, in certain tasks. If you had an application that made heavy use of these, such as media encoding, certain photoshop filters, you could see a big jump in performance.

All in all, the new Pentium 4 systems had the ability to be much faster than the Pentium III in certain tasks, but in some cases it performed similar, or even worse. And clock for clock, the CPU itself was generally a worse performer.

  • It should be noted, that with higher resolution this pretty much levels out.
    – Raffzahn
    May 23 '20 at 14:53
  • 4
    @Raffzahn: Only because then you are limited by your graphics card.
    – Michael
    May 24 '20 at 7:09

Is it true that the Pentium III was faster than its successor Pentium IV?

Yes and no. As usual it all depends on the time to look at, but also what is defined as 'fast'.

MHz vs. Throughput vs. Bang per Buck.

In the MHz game (*1) the Netburst (Pentium 4) design clearly outran the P6 (Pentium II/III; *2). In throughput it was much like with every new design. The fastest models of the previous design were faster than the lower end models of the actual one - especially when the 'old' design continued development (*3).

When the first Pentium 4 model (Wilamette), was introduced in late 2000 as 1.3 and 1.4 MHz, the fastest Pentium III was a Coppermine with 1.100/1.133 GHz (100/133 MHz FSB). While it featured a higher MHz number (good for advertisement) its performance was significantly lower - at best some two-thirds of the Pentium III.

But less than 10 months later, in summer 2001 Willamette-based Pentium 4s reached 2 GHz, outrunning all Pentium IIIs at the time. In January 2002 the new Northwood core debuted at 1.6 - 2.26 GHz, cranking up to 3+ GHz until summer of 2002. Those were regions even the improved Pentium III (Tualatin) with 1.1 ... 1.4 GHz could not compete with.

In addition there's the difference in floating point. The Pentium 4 not only introduced SSE2, but more importantly redid the whole floating point architecture resulting in a much higher performance ... except, 2000s games weren't as floating point hungry as today. This is best seen on 3D games of the time - the ones that could still operate without 3D acceleration on the graphics card. So again, YMMV.

Now, what the Pentium III really could compete on was performance per money. The continued Pentium III line was set at a considerably lower price. So if one really wanted an Intel CPU (*4) the Pentium III series would offer a better price/performance ratio.

So, depending on the viewpoint, but always comparing the top models at each time:

  • The Pentium 4 was slower MHz-for-MHz comparison
  • The Pentium 4 delivered less punch per Euro
  • The Pentium 4 was absolutely faster than the Pentium III

Bottom Line: It's rather unfair to claim the PIII being faster without adding the moment in time and criteria to be used.

*1 - The Netburst architecture was in part selected as follow up because it promised high MHz numbers, which would look good in sales. At the end of the 90s Intel was in the desktop market and more importantly in public view outclassed by AMD CPUs (K6/K7). This image was mostly communicated by Megahertz numbers. After all, Intel did run behind not only at throughput per MHz but as well on absolute MHz numbers.

The other reason was to go the way AMD has shown - turning the classic P6 architecture into a outer shell for a RISC core. After all, AMD did show since K5/K6 times how much better this worked out. So the Pentium 4 was an all-in game. And it worked ... well, in part, as they could only close the gap but not overtake. AMD upped the bar in addition with x86-64 and multiple cores.

*2 - Pentium II and III are essentially the same design. The renaming was more of a marketing measure to present it as something new, freeing the image of previous less-than-perfect performance :)

*3 - Already the original Pentium was a slow sell. A 60 MHz P5 wasn't really faster than a 486 DX4-100.

*4 - No gamer in his right mind in the early 2000s would buy an Intel CPU. They were a simple no-go. An Athlon XP delivered better performance than a Pentium 4 at a lower price, even more so the later Athlon 64. It wasn't until the Core 2 in the late naughties that Intel became competitive again.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.