Was pentium 4 an over engineered CPU?

[Prime2515102]

My first pc back in 2005 was powered by intel p4 3.00ghz CPU. I don't know if I'm exxegareting but I remember it was so smooth and quick CPU even for today's standards. The system load speed, browser responsivity was so good that I've never seen anything like this back then. Was this CPU mistakenly built far too powerful than anything known or over engineered or I'm mistaken or the windows xp was very light for pentium 4?

Except in floating point, my AthlonXP "Barton" mobile chip (that fit in a desktop socket) running at 2.4Ghz beat the pants off a Pentium 4 @ 3.4Ghz.

They made the pipeline too long (took too much time to recover in the case of a misprediction or whatnot) causing the performance to be completely non-indicative of the (at the time) clock speed. They were inefficient, and the engineering was dictated by the marketing department. It was a shameful fiasco that the general non-enthusiast public took up the tailpipe.

I think everyone learned from Intel's lesson.

 

[QuietBob]

It's not just Athlon 64, the little guy Sempron also run miles around it. I got Northwood 2GHz that clocked at 3GHz that run slower than Sempron 2500+ s754 which only clocked at 1.4GHz

I'll second that.

A few years back I tested that same s754 Sempron 2500+ (Palermo) 1.4 GHz against an LGA775 Celeron D 326 (Prescott) 2.53 GHz in Win7. These happen to be the slowest 64-bit enabled desktop SKUs from their respective manufacturers, that launched about a year apart in 2004/2005. Both were tested with identical RAM and a very similarly performing HDD.

Though the two were mostly matched in CPU benchmarks (the Celeron often winning synthetic tests), the Sempron config felt more responsive in use. Everything from booting, navigating the OS, to launching apps and basic browsing was noticeably faster on the AMD rig. When paired with the same GPU, the Sempron could even play VP9 YouTube videos up to 480p, while the Celeron couldn't handle 144p.

 

[Vya Domus]

P4/NetBurst was one of the last architectures built on top of the assumption that clock speeds will go up and power consumption will go down a the same rates as it did before and then the unimaginable happened : they didn't.

 

[freeagent]

The super extendo pipeline killed it, amongst other things. They ripped pretty good when you wound them up.

 

[mclaren85]

P4/NetBurst was one of the last architectures built on top of the assumption that clock speeds will go up and power consumption will go down a the same rates as it did before and then the unimaginable happened : they didn't.

At least they should have predicted it theoretically weren't they?

Btw, almost everyone mentioned "pipeline" thing I don't know what that is.

 

[Princess Garnet]

I couldn't tell you how it was or wasn't "engineered", but Netburst (the Pentium 4 architecture that all revisions used) is often considered one of Intel's lowest points, if not the lowest point, if that tells you anything.

To understand what happened, the late 1990s/early 2000s were when frequency was growing perhaps faster than ever. Much of the advancement at that time was therefore just coming from "brute force" due to clock speeds going up (adding, and then localizing, cache was the other that happened in the mid-1990s). From what I understand, a shorter pipeline is harder to stabilize at higher frequencies (?), but regardless, Intel made a gamble here; they made the pipeline much longer compared to the Pentium III with the idea that the performance impacts would be offset if the frequency scaled up enough. This is why the initial Pentium 4 was quite a large frequency jump over what came before (especially if you ignore Taulatin which was uncommon) and Intel anticipated that Netburst might take things to 10 GHz. It didn't. On top of this, the later Prescott variant (which was different enough it could have been called Pentium 5) extended the pipeline even further. Therefore, clock for clock, Pentium 4 was sometimes a regression compared to Pentium III, and later Pentium 4s were perhaps a further regression.

I had a Pentium III as well as a few different Pentium 4s I used (with an Athlon 64 mixed in, but time with it was short) in those years.

Regardless, you're looking at Windows XP on CPUs that are much, much higher than what it needed, so the fact that it felt fast isn't a sign it was over-engineered. Service Pack 2 in particular did make Windows XP feel a bit more sluggish and need more memory, but any later Pentium 4 system probably would have absorbed that well.

 

[Vya Domus]

At least they should have predicted it theoretically weren't they?

Frequency limits were expected to be reached at some point as there are clear limits on how fast signals can travel but the power consumption thing really did caught the entire industry by surprise, nobody expected power scaling to become so bad after the ~3Ghz mark using early 2000 era nodes.

 

[stickleback123]

I remember well the billy big balls talk of 10GHz within a couple of years at the launch of the P4. They let the marketing department design set the specs for that CPU!

 

[Dr. Dro]

Honestly Pentium D on 775 was a lot more fun!

I still kinda wanted to try my hand at overclocking a Cedar Mill. Those were pretty wild, 65 nm P4's. Unfortunately my S775 boards are all G41 trash that die at 340 FSB, most I managed was 3.5 on a QX9650 because unlocked multiplier

 

[Darmok N Jalad]

I had forgotten about Willy until mentioned above, which was also a lousy launch since it was getting bested by Pentium III still. And then there was the whole Rambus issue. Intel really wanted that to succeed, but thankfully DDR showed up and saved us from those dummy DIMMs that were necessary for empty slots. I believe it was SiS that had a DDR P4 chipset that was superior to the Rambus ones. Intel finally abandoned Rambus for DDR in the consumer space, but Rambus hung around in the server/workstation space for years.

Also, don’t forget about Itanic. Intel had big dreams of that replacing x86 around that time too. Thankfully that one started its slow trip to the morgue when AMD launched 64bit Opteron and the hypertransport bus.

 

[ShrimpBrime]

I still kinda wanted to try my hand at overclocking a Cedar Mill. Those were pretty wild, 65 nm P4's. Unfortunately my S775 boards are all G41 trash that die at 340 FSB, most I managed was 3.5 on a QX9650 because unlocked multiplier

I didn't get into OC old Intel until much later. I just stuck with AMD mostly through those years. The nice thing about that time line, is everyone had worked out what boards and processors OCed the best. So that took the trial and error away from me. Thankfully I should add.

 

[LabRat 891]

No, and yes.

As others have gone over, the Pentium 4 (netburst) was kinda ass. (I had a Northwood 3.0)

That said, there's some Socket 478, many 604 , and most 775 CPUs that will 'run' modern Windows x64 and modern software with Modern security requirements.
Whereas all of the 'superior' S754, S940, and S939 AMD64 chips, cannot.

So, while it has 0 to do with performance:
Yes. The Pentium 4 was over engineered.

 

[oxrufiioxo]

Pentium 4 is what prompted me to ditch Intel for amd i don't remember the specific models but the p4 was hot garbage.

 

[Shihab]

For me it was Westmere or Nehalem... of course thats because its what I had at the time (for a very long time as it were). YMMV.

+1 for Nehalem.
Although, objectively, I think the entire stretch from Core 2 to Sandy/Ivy was a great run for Intel.

Sad how things stagnated since Haswell. Haswell itself was decent, but other than AVX2, it was hardly noteworthy (and I'm partial to it for the same reason you are to Nehalem).

 

[R-T-B]

+1 for Nehalem.
Although, objectively, I think the entire stretch from Core 2 to Sandy/Ivy was a great run for Intel.

Sad how things stagnated since Haswell. Haswell itself was decent, but other than AVX2, it was hardly noteworthy (and I'm partial to it for the same reason you are to Nehalem).

Yeah my next upgrade was Skylake, which wasn't really notable other than I got one of the rare 6th gen chips that could do 4.5Ghz (or was it 5GHz? been too long lol) allcore, but yeah after haswell it really stagnated bad.

 

[68Olds]

Btw, almost everyone mentioned "pipeline" thing I don't know what that is.

Simply put, it is the number of steps between the CPU's process in and process out. A s478 northwood had a 20 step pipeline and the s478 prescott has a 31 step pipeline. So all thing being equal, a northwood 'felt smoother' at 3.2GHz vs a prescott at 3.2GHz. Although the prescott smoothed out that long pipeline when it was overclocked to 3.8GHz or higher.

 

[oxrufiioxo]

Yeah my next upgrade was Skylake, which wasn't really notable other than I got one of the rare 6th gen chips that could do 4.5Ghz (or was it 5GHz? been too long lol) allcore, but yeah after haswell it really stagnated bad.

Probably 5ghz... Sandybridge could already do 4.3-4.5 pretty easily from memory.

 

[R-T-B]

Probably 5ghz... Sandybridge could already do 4.3-4.5 pretty easily from memory.

Yeah, that sounds right. I remember being pleased that both the cache and core could reach a nice rounded off number.

 

[Dr. Dro]

I didn't get into OC old Intel until much later. I just stuck with AMD mostly through those years. The nice thing about that time line, is everyone had worked out what boards and processors OCed the best. So that took the trial and error away from me. Thankfully I should add.

You should try a Cedar Mill sometime. They clock like crazy, CPUs cost next to nothing, so getting a whole tray and binning should make for a fun weekend

 

[Frag_Maniac]

My first pc back in 2005 was powered by intel p4 3.00ghz CPU. I don't know if I'm exxegareting but I remember it was so smooth and quick CPU even for today's standards. The system load speed, browser responsivity was so good that I've never seen anything like this back then. Was this CPU mistakenly built far too powerful than anything known or over engineered or I'm mistaken or the windows xp was very light for pentium 4?

There's no such thing as too much engineering in PC components as long as it translates to good performance and isn't overpriced. I had a P4 in one of my older rigs and ran Crysis at 1200x900 on high textures, and med shaders, by using a few visual enhancing cvars I found on Tweakguides. The thing was an absolute beast for it's time.

 

[ShrimpBrime]

You should try a Cedar Mill sometime. They clock like crazy, CPUs cost next to nothing, so getting a whole tray and binning should make for a fun weekend

View attachment 387664

Oh one of these days I'll get around to it no doubt. Trying to get some experience with any and all platforms possible. Which, is not as easy as it sounds. There's a lot of gear out there!

 

[Endymio]

Athlon - 23.06.1999. Athlon XP was released on 14.05.2001;

Pentium 4 - 20.11.2000. Northwood version release date- 23.06.2003.

What are you babbling about?

Learn to read. He was comparing Northwood to a Core 2 Duo.

 

[Aken Bosch]

Learn to read. He was comparing Northwood to a Core 2 Duo.

My bad, reading comprehension fail .

 

[azrael]

I remember PresHOT lol.

Back in the day I had a Northwood celeron 2.4 that oc'd well, but a later Athlon Xp Barton core was considerably faster.

The very first Pentium CPUs (60 and 66 MHz) back in the early 90s got nicknamed the "M & M chips". Apparently, they could run so hot as to desolder the socket from the motherboard. May very well be an urban legend, though. My first PC sported the Pentium 90 MHz, a generation later. And yes, it did suffer from the infamous FDIV bug, and yes, I got it replaced.

 

[Darmok N Jalad]

The very first Pentium CPUs (60 and 66 MHz) back in the early 90s got nicknamed the "M & M chips". Apparently, they could run so hot as to desolder the socket from the motherboard. May very well be an urban legend, though. My first PC sported the Pentium 90 MHz, a generation later. And yes, it did suffer from the infamous FDIV bug, and yes, I got it replaced.

I believe some of the issue there was the lack of active cooling. The 486 didn’t need more than a small aluminum heatsink without a fan. That wasn’t enough for the Pentium, and really when the aftermarket cooling market started to take off. I remember having a Golden Orb cooler on my K6-2.