Was Prescott really that bad?

[pantsaregood]

Looking around, it seems to be viewed as the point where Intel finally lost to AMD, but that's not really true.

Athlon 64 beat Prescott to the market by about six months, so until then it was competing against (and beating) Northwood. Prescott didn't outperform Northwood, but it did bring some things to the table that get overlooked - SSE3 and Intel 64 were implemented first on Prescott.

As for scalability, it trashed Northwood. Prescott could easily break 4.0 GHz with proper cooling, and I had a Celeron 345 (3.06 GHz) that would run happily at 4.6 GHz with some garbage air cooler. Prescotts overclocked fantastically if you could get around thermal issues.

Netburst was awful as a whole, but to say it was "good" until Prescott is a bit dishonest. They ran hot, but they were monsters at overclocking if cooled properly.

 

[BigDaveX]

Athlon 64 beat Prescott to the market by about six months, so until then it was competing against (and beating) Northwood. Prescott didn't outperform Northwood, but it did bring some things to the table that get overlooked - SSE3 and Intel 64 were implemented first on Prescott.

SSE3 never really provided much of a performance boost; SSE2 and SSE4.1 were the more successful versions in that regard. And while it may have been the first Intel chip to support x64, firstly it was a hackjob version that could actually cause performance degradation, and secondly 64-bit OSes didn't become widespread on the desktop until after Core 2 had hit the market, so it wasn't particularly useful at the time anyway.

As for scalability, it trashed Northwood. Prescott could easily break 4.0 GHz with proper cooling, and I had a Celeron 345 (3.06 GHz) that would run happily at 4.6 GHz with some garbage air cooler. Prescotts overclocked fantastically if you could get around thermal issues.

Which would have been nice if these overclocked Prescotts weren't generally slower than stock Athlon 64s. In terms of overclocking, Cedar Mill and Presler were actually more useful in that department, since you could buy a CPU and a 975X motherboard, enjoy six months of insane clockspeeds, and then swap in a Core 2 when it arrived.

Netburst was awful as a whole, but to say it was "good" until Prescott is a bit dishonest. They ran hot, but they were monsters at overclocking if cooled properly.

True, it is dishonest... because Willamette also sucked, and while Northwood was a lot better, it did kinda get lucky in that AMD's non-SOI 130nm process was pretty mediocre, which caused the Athlon XP to slam into a clockspeed ceiling pretty early on.

Now, if there is an argument to be made in favour of Prescott (aside from the core's issues being exacerbated by Intel's 90nm process being a dud) it's that it was actually a pretty decent encoding CPU, since the Athlon 64 was only single-threaded and had slower SSE2 performance. Even so, that advantage evaporated the minute that the Athlon 64 X2 arrived on the scene.

 

[daxzy]

Intel didn't lose to AMD, even in that time frame. Intel had a very competitive chip in their hand, but for internal political reasons, kept with the Netburst line. Only after the unmitigated disaster that was Prescott, were those execs forced to resign. When Core 2 Duo finally came around to the desktop, only the uninformed were surprised what Intel could actually do.

I, and a few others, didn't upgrade to Prescott, but upgraded to CT-479/Dothan, that blew everything away. Only when the X2's arrived (and the fact that Merom wasn't pin compatible with the desktop lineup), did AMD take a sizeable lead. Ignoring the synthetics (3dMark), Dothan blew everything away at half the power usage.

http://www.tomshardware.com/reviews/dothan-netburst,1041-12.html

 

[witeken]

In my mind, Netburst, much as it's maligned, brought some very good things internally for Intel design teams. First, unbelievable circuit expertise (the FP logic was running at 8GHz in Prescott stock!). Next, the trace cache which you can see reimplemented in Sandy and Ivy Bridge. Also, SMT. Building a validation team that could validate the beast pre- and post-silicon. The power-perf thinking i.e. frequency through power savings. Finally, the development of tools and project management required to do that kind of extreme design. All of these learning continue to this day and it's a very large contributor to why in client and server CPUs Intel can sustain the roadmap we have.

https://www.reddit.com/r/IAmA/comments/15iaet/iama_cpu_architect_and_designer_at_intel_ama/

 

[BigDaveX]

For a four-year old design with a much faster FSB and an L3 cache slapped on, the P4EE's actually putting up quite a valiant struggle in that chart, especially seeing how UT2004 was a benchmark that Intel always did poorly on even when it was only competing against the Athlon XP. The sight of all the Prescott models floundering - especially the 600-series ones, which should have had even better cache performance than Gallatin - makes me wonder just how it went so spectacularly wrong.

(Well, okay, it's probably the 50% increase in pipeline length over Northwood, but still...)

 

[krumme]

https://www.reddit.com/r/IAmA/comments/15iaet/iama_cpu_architect_and_designer_at_intel_ama/

Thanx. The thread is a good read.

 

[wingman04]

The Prescott has 31 stage pipeline compared to skylake 14 (16 with fetch/retire) and 31 stage pipeline allows the faster clock speeds however to keep the pipes full it's not achievable so gain in performance is null.

List of Intel CPU microarchitectures Link: https://en.wikipedia.org/wiki/List_of_Intel_CPU_microarchitectures

Instruction pipelining LINK: https://en.wikipedia.org/wiki/Instruction_pipelining

 

[bbhaag]

It's not that the P3 and P4 architectures were bad but AMD managed to acquire NexGen for a cool $850m in 1995 which then led to the K6 and K7 cpus and well the rest is history. I often wonder what would have happened if NexGen didn't sellout and continued to develop cpus. I guess we will never know.

 

[majord]

AMD should have seen the writing on the wall the moment Pentium M was released. This was the beginning of there undoing IMO. And no this didn't mean looking into a crystal ball.. AMD knew full well how competitive it was in late 2002 / early 2003 , and should have ultimately taken a more aggressive approach with K10.

Anyway, this thread is about Prescott, so not sure how PM is relevant. Personally I don't understand why people are still clutching at straws to defend this architecure! it was bad, and you can't just look at it in isolation. For intel, with all its resources, and market share to persist with Prescott, and get so far down the line with Tejas was an epic failure.

 

[Arachnotronic]

It's not that the P3 and P4 architectures were bad but AMD managed to acquire NexGen for a cool $850m in 1995 which then led to the K6 and K7 cpus and well the rest is history. I often wonder what would have happened if NexGen didn't sellout and continued to develop cpus. I guess we will never know.

yup...

SUNNYVALE, Calif. — Advanced Micro Devices Inc. said it will buy Nexgen Inc. for about $857 million in stock, in an effort to keep pace in the computer chip industry with market leader Intel Corp.

Nexgen shareholders will receive 0.8 share of AMD for each common share they hold. The boards of both companies have already approved the transactions, and so have holders of 37% of Nexgen shares, the companies said. Those holders include Compaq Computer Corp., Italy's Olivetti Group and Japan's ASCII Corp.

In making the acquisition, AMD is admitting it couldn't develop a microprocessor that could compete against the Pentium by Intel, which is expected to ship 40 million units this year, analysts said. Nexgen, based in Milpitas, Calif., is already shipping a Pentium-class product and earlier this month announced the design for the next generation of chips beyond Pentium-class.

"We never felt that we could compete against Intel alone," said W.J Sanders, chairman of AMD. "This enhances our overall ability to compete. What the customers want is a credible alternative."

 

[ninaholic37]

P4 hit a wall, no 10GHz CPU for 2005, maybe not until Pentium 18 with graphene or indium gallium arsenide or new elements from Jupiter at this rate

 

[Insert_Nickname]

Yes. They were that bad. [/thread]

SSE3 never really provided much of a performance boost; SSE2 and SSE4.1 were the more successful versions in that regard. And while it may have been the first Intel chip to support x64, firstly it was a hackjob version that could actually cause performance degradation, and secondly 64-bit OSes didn't become widespread on the desktop until after Core 2 had hit the market, so it wasn't particularly useful at the time anyway.

Having tried running an x64 OS on a P4 651, you're absolutely right. It'll run, but the actual performance is abysmal when running x64 code.

Now, if there is an argument to be made in favour of Prescott (aside from the core's issues being exacerbated by Intel's 90nm process being a dud) it's that it was actually a pretty decent encoding CPU, since the Athlon 64 was only single-threaded and had slower SSE2 performance. Even so, that advantage evaporated the minute that the Athlon 64 X2 arrived on the scene.

I'm not sure I subscribe to the idea that Intel's 90nm was such a dud. You only have to look at the Dothan Pentium M's to see what was possible.

http://www.tomshardware.com/reviews/generation-cool,1063.html

Disclaimer: This board provided the foundation for my first (what we'd now know as a) HTPC. It was every bit as impressive as the article makes it out to be.

I, and a few others, didn't upgrade to Prescott, but upgraded to CT-479/Dothan, that blew everything away. Only when the X2's arrived (and the fact that Merom wasn't pin compatible with the desktop lineup), did AMD take a sizeable lead. Ignoring the synthetics (3dMark), Dothan blew everything away at half the power usage.

The Socket 478 to Socket 479 adaptor brings back a few memories.

AMD should have seen the writing on the wall the moment Pentium M was released. This was the beginning of there undoing IMO. And no this didn't mean looking into a crystal ball.. AMD knew full well how competitive it was in late 2002 / early 2003 , and should have ultimately taken a more aggressive approach with K10.

Anyway, this thread is about Prescott, so not sure how PM is relevant. Personally I don't understand why people are still clutching at straws to defend this architecure! it was bad, and you can't just look at it in isolation. For intel, with all its resources, and market share to persist with Prescott, and get so far down the line with Tejas was an epic failure.

Willamette was bad, especially when running legacy x87 code. SSE2 helped somewhat, but wasn't at all common back then. Northwood was okay'ish, once you got the frequency high enough, and didn't mind the additional power draw. Prescott on the other hand didn't have a single redeeming feature at all.

The only thing worse then a P4 Prescott was a Celeron Prescott...

 

[SPBHM]

Prescott had generally an OK performance gain over Northwood, if you consider the higher clock and other improvements, the negative was power draw, Northwood was already weak at it, and Prescott used even more at times.... that's mainly the criticism I remember, the "preshot", still with at times 1.5GHz clock advantage over AMD, you could see "preshot" competing with some K8s here and there, but the Athlon was so much more efficient and Dothan made it clear that Netburst was not the right path.

Celeron Prescott was fairly decent, considering the competitive price; with 512K l2 and good OC potential, it was a lot nicer than Northwood Celeron, Prescott was a bigger win for Celeron than for Pentium in terms of improvements,
Sempron vs Celeron D had some decent fights, I ended up picking the Sempron.

 

[Insert_Nickname]

Celeron Prescott was fairly decent, considering the competitive price; with 512K l2 and good OC potential, it was a lot nicer than Northwood Celeron, Prescott was a bigger win for Celeron than for Pentium in terms of improvements

For the enthusiast willing and able to OC, paired with a decent board and other components, sure. For regular users paired with the usual OEM-crap, definitely not.

512KB L2 wasn't all that much then either. Both P4s and PMs had 1-2MB L2.

 

[Dresdenboy]

It's not that the P3 and P4 architectures were bad but AMD managed to acquire NexGen for a cool $850m in 1995 which then led to the K6 and K7 cpus and well the rest is history. I often wonder what would have happened if NexGen didn't sellout and continued to develop cpus. I guess we will never know.

I don't think, that NexGen had the K7 in preparation. More likely, their engineers, together with AMD's original ones (29K, K5), and those acquired from DEC (Alpha), did the K7 from ground up (1995 - 1999 sounds about right).

 

[poofyhairguy]

Yeah wasn't the DEC gang the superstars back then?

I feel like I keep hearing how amazing the Alpha was for most of my lifetime. I mean the Chinese copied it for a reason right?

 

[whm1974]

Yeah wasn't the DEC gang the superstars back then?

I feel like I keep hearing how amazing the Alpha was for most of my lifetime. I mean the Chinese copied it for a reason right?

For a time the Alpha CPUs was the most powerful CPUs on the market.

 

[Thunder 57]

Intel didn't lose to AMD, even in that time frame. Intel had a very competitive chip in their hand, but for internal political reasons, kept with the Netburst line. Only after the unmitigated disaster that was Prescott, were those execs forced to resign. When Core 2 Duo finally came around to the desktop, only the uninformed were surprised what Intel could actually do.

I, and a few others, didn't upgrade to Prescott, but upgraded to CT-479/Dothan, that blew everything away. Only when the X2's arrived (and the fact that Merom wasn't pin compatible with the desktop lineup), did AMD take a sizeable lead. Ignoring the synthetics (3dMark), Dothan blew everything away at half the power usage.

http://www.tomshardware.com/reviews/dothan-netburst,1041-12.html

P-M was clearly better than P-4 while using way less power. It was very competitive with AMD as well but I wouldn't say it "blew away everything". The P-M had a somewhat weaker FPU than it's competition (from AMD and Intel). It was Core 2 Duo that corrected this while completely taking the performance crown back.

http://www.anandtech.com/show/1610/7


The other problem was the price/performance ratio wasn't particularly great. Here is a list of prices of various CPU's from when AMD did their P-M Desktop experiment:

http://www.anandtech.com/show/1610/3

 

[SPBHM]

For the enthusiast willing and able to OC, paired with a decent board and other components, sure. For regular users paired with the usual OEM-crap, definitely not.

512KB L2 wasn't all that much then either. Both P4s and PMs had 1-2MB L2.

lots of cheap boards could OC back then, but sure, OC was not for everyone, for your average OEM PC, but I think even at stock it was OK, Northwood Celeron was a lot worse, the P4s didn't like 128K l2, 512K was a lot less of a problem.

 

[PaulIntellini]

I hope we will someday get some details about the canceled Tejas.

 

[Insert_Nickname]

Northwood Celeron was a lot worse, the P4s didn't like 128K l2

Not those. ANYTHING but those.

 

[DrMrLordX]

Prescott managed to lose benchmarks against Northwood @ the same clockspeed while putting out more heat.

Most Prescotts did not OC past 4.0 GHz, and there were plenty of 3.2c Northwoods that would go just as far.

For the money, the 3.2c and 3.4c (when it could be found) were the best P4s on the market. There was nothing Netburst worth buying until MAYBE Presler/Cedar Mill, and by then it was too late.

 

[bigboxes]

No. They did heat my room.

 

[wingman04]

lots of cheap boards could OC back then, but sure, OC was not for everyone, for your average OEM PC, but I think even at stock it was OK, Northwood Celeron was a lot worse, the P4s didn't like 128K l2, 512K was a lot less of a problem.

Overclocking now is much easier no FSB to deal with limiting the overclock. Back then the Prescott did not overclock better than they do now, also many more Hz with my cheap motherboard GA Z170 HD3, i5 6600k 4.5GHz on air, that is great.

 

[nismotigerwvu]

Intel didn't lose to AMD, even in that time frame. Intel had a very competitive chip in their hand, but for internal political reasons, kept with the Netburst line. Only after the unmitigated disaster that was Prescott, were those execs forced to resign. When Core 2 Duo finally came around to the desktop, only the uninformed were surprised what Intel could actually do.

I, and a few others, didn't upgrade to Prescott, but upgraded to CT-479/Dothan, that blew everything away. Only when the X2's arrived (and the fact that Merom wasn't pin compatible with the desktop lineup), did AMD take a sizeable lead. Ignoring the synthetics (3dMark), Dothan blew everything away at half the power usage.

http://www.tomshardware.com/reviews/dothan-netburst,1041-12.html

Dothan was a pretty solid chip, but blowing everything else away might be a bit of an overstatement. http://www.anandtech.com/show/1610