Without Pentium 4/Netburst, how long would it have taken Intel to get to 'Core'

[kent]

Let's say the PIV/Netburst didn't exist and let's say the Core Solo (Pentium III/-M based) was called the Pentium IV.

How long from the release of the last Tutulians until the first part that we today would call "First Generation Core"?

Imagine an Athlon 64 against an Intel Core in say 2003, LOL.

 

[Erasmus354]

The netburst architecture wasn't a full on failure. It was very competitive with Athlon, and Intel learned a lot of valuable engineering lessons. They were forced to come up with HyperThreading to make full use of Netburst's longer pipeline, technology which we still find in Intel processors today. Just because Netburst couldn't scale to where they initially wanted it to go doesn't mean it should have been skipped or that we would be better off today if it had been skipped.

What if instead of the 4004 we just went right to Haswell lololol..... there is a natural progression to, well, progress, and hypothetically skipping steps in that progress is just dumb.

 

[BinarySynapse]

The netburst architecture wasn't a full on failure. It was very competitive with Athlon, and Intel learned a lot of valuable engineering lessons. They were forced to come up with HyperThreading to make full use of Netburst's longer pipeline, technology which we still find in Intel processors today. Just because Netburst couldn't scale to where they initially wanted it to go doesn't mean it should have been skipped or that we would be better off today if it had been skipped.

What if instead of the 4004 we just went right to Haswell lololol..... there is a natural progression to, well, progress, and hypothetically skipping steps in that progress is just dumb.

Netburst was a deviation from that progression though. The only way it could compete is by ramping up the clock speeds and in turn power consumption. It is an example of a company that decided to listen to its marketing department rather than its engineers.

So yeah, it should have been skipped. And I do believe we'd be better off today since AMD wouldn't have become complacent with the performance lead they got only because of Netburst's existence, which in turn has allowed Intel to stagnate it's product line with just enough gain each generation to say they have something faster than what they sold you last year.

 

[GotNoRice]

I think the first Pentium 4 was pretty shitty. Socket 423, required Rambus RAM, wasn't really all that much faster than a top end P3, especially an overclocked P3.

But in Jan of 2002 the Northwood P4's came out. The Socket 478 P4's that weren't using Rambus anymore were the first "good" P4's. The Athlon64 didn't come out until over a year later. Back in 2002 if you were rocking an overclocked Northwood P4 you were running a top-end system.

I do believe we'd be better off today since AMD wouldn't have become complacent with the performance lead they got only because of Netburst's existence, which in turn has allowed Intel to stagnate it's product line with just enough gain each generation to say they have something faster than what they sold you last year.

You really believe that the only thing stopping AMD from releasing a processor faster than Intel is them having become complacent due to stuff that happened 10 years ago?

 

[pxc]

I think Transmeta was a large driver towards the "Core" strategy. While Intel had low power versions of chips since the 486 era, it took a push from the smaller company to make the efficient processor strategy mainstream.

I guess the question is also whether Intel would have continued improvements to the i386 without Netburst. It seems doubtful that was considered since no parallel products were in serious development. Historically Intel made radical departures when major new processors were released. It would have been very odd if Intel had stuck with a P3 derivative.

There were all kinds of problems with Netburst, but P3 didn't seem viable as a base to compete against Athlon XP and K8. It's hard to tell if Intel could have made Core without competition in performance and low power.

 

[GotNoRice]

I guess the question is also whether Intel would have continued improvements to the i386 without Netburst. It seems doubtful that was considered since no parallel products were in serious development. Historically Intel made radical departures when major new processors were released. It would have been very odd if Intel had stuck with a P3 derivative.

You seem to be omitting anything having to do with the Pentium-M, which was the result of their "continuing improvements" to P6 architecture and evolved into the Core Duo (predecessor to the Core 2 Duo). It existed concurrently ("parallel product") with Netburst. The Pentium-M was a direct successor to the Pentium 3 in the mobile space. There was no gap in development.

 

[BinarySynapse]

I think the first Pentium 4 was pretty shitty. Socket 423, required Rambus RAM, wasn't really all that much faster than a top end P3, especially an overclocked P3.

But in Jan of 2002 the Northwood P4's came out. The Socket 478 P4's that weren't using Rambus anymore were the first "good" P4's. The Athlon64 didn't come out until over a year later. Back in 2002 if you were rocking an overclocked Northwood P4 you were running a top-end system.



You really believe that the only thing stopping AMD from releasing a processor faster than Intel is them having become complacent due to stuff that happened 10 years ago?

Well, incompetence and lack of direction from the upper levels really... but I believe that was rooted in complacency. Core 2 caught them with their pants down and they've been unable to catch up since. They should have seen the writing on the wall when Pentium-M evolved into Core Duo, and Intel announced their plans to scrap Netburst on the desktop after Prescott's miserable failure to reach 4.0GHz.

 

[Gman1979]

The socket 423 Pentium 4s were just outright a joke when tied to Rambus memory. The later 478 Pentium 4s got a bit better once they weren't shackled to Rambus but the platform still had some issues. Intel's marketing team knew normal people would just see a Mhz number and think higher was always better. The problem was the process technology wasn't up to the task at the time to make the most of the design.

It was no coincidence AMD made a huge leap in technology between the K6-3 cpus and the introduction of their Athlon cpus that caubght Intel off guard. I'm sure there are still a few of us around here that remember when DEC was still an independent company making ALPHA cpus. When Compaq bought DEC, a nice chunk of their engineering teams got snatched up by AMD along with licenses for some really nice technology (AMD double data rate system bus for example). Essentially AMD got an infusion of enterprise level brain power and patents that started a trend of CISC cpus incorporating RISC chip design features. If Intel could have buckled down and pushed out the Tulatin Pentium designs instead of the Pentium 4 they went with it would have been a very tight race.

Some of you may wonder why I just wrote such a long paragraph concerning AMD in a post concerning Intel but I assure you it's relevant. You see, Compaq was later bought by HP. When this happened some more of the ex DEC guys were out of jobs and got work at Intel, where they supplemented Intel's already deep engineering ranks. Intel then gave us the CORE design (the DEC guys probably didn't have enough time to influence it's design much but I'm sure they didn't hurt). When HP signed it's deal with Intel concerning the Itanium cpus, part of the agreement was for the ALPHA cpu line to get killed off by HP with Intel absorbing what was left of the ALPHA engineering teams. Intel hit the jackpot in that deal. Itanium wouldn't have to try to compete with ALPHA and the engineering expertise they gained helped us get the CORE 2 and the designs that came afterwards.

While the Pentium 4 design was sub par for what we normally expect out of Intel, it was an evolutionary step in cpu design all the same. It also provided the kick in the arse Intel needed to spend the cash it needed to the build the superior chip fabs it has today. It also provided us with the first use of simultaneous multithreading, AKA Hyperthreading; a feature that was designed by DEC for its ALPHA cpus .

Not really relevant:
It saddens me a bit when I think of what might have been possible had the ALPHA cpus continued their evolution. Both Intel and AMD cpus continue to benefit from ideas and technology that originated with the ALPHA.

 

[Track Drew]

Imagine an Athlon 64 against an Intel Core in say 2003, LOL.

Core Solo (Yonah) was pretty mush a process shrink of Pentium M (Banias/Dothan) with some tweaks, SSE3, and DDR2. That process shrink was likely accelerated because of the investment Intel had to make to get the P4 onto the smallest fab possible.

Without Pentium 4, what are the chances that Intel would have been much slower to move to 90nm and then 65nm?

Athlon 64 did exist at the same time as an Intel Core in 2003 - it just wasn't named it yet and didn't have desktop parts.

 

[pxc]

You seem to be omitting anything having to do with the Pentium-M

The Pentium M was more of an oddball, near end of the P3 line design with hybrid features borrowed from Netburst.

Banias/Dothan, Yonah and Merom are only loosely related as they have common P3 ancestry, but Yonah and Merom are not really evolved from Banias/Dothan as they are a re-engineering of the base architecture (plus many uarch improcements) with power efficiency in mind.

It's a fair point though that at least in mobile there was a parallel design going on with Pentium M.

But that goes back in a circle. Without Transmeta, the driver behind Yonah, which itself beget Merom, and without competition from AMD, I'm not so sure Intel would have gone back from Netburst. The original Tejas/Cedarmill probably would have gone forward as some kind of Netburst 2, still chasing GHz over IPC (and I'm not implying those would be better than what Intel makes now). Clearly, Netburst was a really bad strategy from the start, and Intel never seems to have expected something like K8 out of AMD. That, as the kids say, changed everything.

 

[bigsnyder]

If memory serves me correctly, the engineering team behind Core was completely different from the Netburst folks. In my opinion too much overlap in the development timeline to have borrowed from each other, but I could easily be wrong.

 

[aphexcoil]

What if instead of the 4004 we just went right to Haswell lololol.....

I would imagine there would still be HardForum threads like these:

"Just delidded my 4004 and I'm only counting 2,299 transistors. Is this not the K version?"

"Why does Intel want us to buy new motherboards for Haswell? My 4004 had 16 pins and now they need 1,150 of them? WTF Intel?"

"Just bought the Haswell CPU and encoded a video in 3 minutes instead of 5,400 centuries. A little disappointed with the performance increase. Think I'll wait until sub-atom processors."

"Is 23 volts too much for my 4004? I'm getting BSOD in Windows. Ran fine on 22 volts."

"Should I purchase the Haswell now or just wait for the 4004-E processors coming this fall?"

 

[Michaelius]

Netburst was a deviation from that progression though. The only way it could compete is by ramping up the clock speeds and in turn power consumption. It is an example of a company that decided to listen to its marketing department rather than its engineers.

So yeah, it should have been skipped. And I do believe we'd be better off today since AMD wouldn't have become complacent with the performance lead they got only because of Netburst's existence, which in turn has allowed Intel to stagnate it's product line with just enough gain each generation to say they have something faster than what they sold you last year.

I remember there was a presentation posted somewhere on [H] where one of Intel designers who made Nehalem told that if Netburst were continued it would be faster than Nehalem by around 10% but it'd cost higher power consumption.

So the desktop priority of power was sacraficed for better power efficiency for growing laptop and server markets.

 

[BinarySynapse]

I remember there was a presentation posted somewhere on [H] where one of Intel designers who made Nehalem told that if Netburst were continued it would be faster than Nehalem by around 10% but it'd cost higher power consumption.

So the desktop priority of power was sacraficed for better power efficiency for growing laptop and server markets.

It would have only been 10% faster because they were planning on having 10GHz clock speeds by 2011. But if they couldn't get power consumption under control beyond 3.8GHz, even after moving to 65nm, I can only imagine what it would have been like at 10Ghz..1000w TDPs anyone?