Celeron overclocked to 8.1GHz

Haven't gone that fast since about 1990. Perhaps you mean GHz...

Outside of the UK, some countries use the full stop instead of the comma to denote 1000's
Therefore UK = 1,000 = one thousand.
In Spain I regularly saw 1.000 = one thousand.
Knowing this, the writer was not incorrect to put it - be it intentional or not.

i overclocked my 386 sx-16 to over q9750 speeds using a diesel generator for power and a polarbear turd for cooling!

Correct. But if that was the case, and they were using continental notation - then the '(1,352.7MHz effective)' bit wouldn't make much sense, would it?

To play Quake II Expert mod, obviously!

intel got a lot of criticism for netburst. about how the the architecture was projected to scale to 10ghz. maybe they werent so far off the mark. team italy once pushed a pentium 4 631 past 8ghz too. makes you wonder how far some of these parts would go on a high k 45nm or even 32nm process

It always a heat problem. They designed the chip to hit super high clocks and it will, but the heat and power considerations were grossly underestimated.

It's a real 'shocker' that a balanced approach of clocks/pipeline depth/cache/latency proves to be more effective in the real world....

Intel never said I'd need a liquid nitrogen cooling system in my home to let the CPU hit 8 or 10 GHz. Think back: high stock clocked P4s dissipated something like 150W heat - clearly there was *not* a lot of headroom left if one cooled them with air. You do not propose the market would have accepted LN CPU cooling systems, do you?

willie eckerslike....

You are officially my new random inquirer hero.

party onward.

i propose that a 32nm high k pentium 4 or celeron would be able to reach tremendous speeds without using LN. intel once said probably a decade ago now that they project netburst architecture would scale over X period of time given process advancements. well considering these are old skool 65nm parts and that intel's 32nm process has been sampling already for some time then it would be interesting to see the old netburst on the newest generation process. just to see if intel were in fact, correct.

i suspect that it would scale. it will obviously never happen. netburst was retired because of low k leakage problems, in effect the process had simply not moved on enough for them to keep bumping clocks. the mhz wars are long over and the biz has moved on to other philosophies. also being blessed with core 2 i cant say im upset. but i would still be interested if intel were correct about netburst a long time ago

Hmm, so CPU-Z measures 8 GHz. Weren't there double-pumped ALU's in that REE? Does this mean that cooked celery was running some AND's and XOR's at 16 GHz?
Impressed.

I compleetly agree to the author of previous post. It really would be very interesting to see what could 32nm Netburst core deliver.

The press allways takes the mainstream opinion and the distributes it. So in Northwood times everyone wrote how succesfull architecture Netburst, but after Prescot when Intel decided to move away from Netburst, everyone suddenly wrote how flawed the Netburst was and how it was never going to work, and how they saw it all before evereyone else..

The turth is Netburst was very interesting and brave architecture. Yes it costed Intel a few percent of market share at bad times, and that is why it was discontiniued, but you have to admit that creators of Netburst surely had balls (no offence intended to ladies involved in project) - cause that is what it takes to create a 31 stage pipeline, place trace cache instead of L1 cache,move most of the FPU instructions to ALU, and announce it is the SSE age now..

And thet weren't compleetly wrong you know... With all those GPU's arround traditional FPU speed does not matter so much anymore.

Netburst is very interesting architecture indeed.. There are still some areas where Netburst will outperform both Core 2 and K10

They simply were caught in a bad time - probably Intel 90nm process was not doing so well and Prescot team took part of the blame..

Who knows how far the team was from solving Netburst problems when Intel management decided to cancel future of Netburst.. Sure Netburst was a failed project, but it probably was not a failed architecture..

I guess we will never know..

2Author of "Double-pumping at 8 GHz ?":

Not all of ALU instructions were double pumped.. I think AND and XOR was not among the ones double pumped.. I once did some quite extensive testing on this. ADD instruction was doulbe pumped for sure.

And one more thing - both ALU's were not identical - not both of them could do all instructions and one of them had to do some FPU.

Also the ALU's were changed during revisions. Test results for Willamete and Norhtwood were quite scalable, but Prescot was a different storry.

I have to look into my results again, but I seem to remember to conclude they removed doulbe pumped ALU's in Prescott

After a quick perusal of the wiki Pentium lists, the highest TDP on any P4 was 115W (pres-hot, of course). The PDXEs however, did have a rediculous 130W TDP.

Some Agena-BEs have a 140W TDP but I haven't compared the two for heat/power.

Hey, what's 20W between friends?

Why 'doughboy'? I'm neither a Yank soldier nor doughy, though I am pretty pale, never could get a tan only a burn.

The Cedar Mill variant actually solved one of the clock-scaling problems of the Netburst design. The ALU and AGU's where remade with "domino logic" instead of "low voltage swing" architecture. That should mean a LOWER clock frequency but actually it became HIGHER because its TDP is only some 85W. So its no longer quite so power constrained. Of course the 65nm process also helped.
The Core2 also uses "domino logic" but the pipeline is 14 stages instead of 31.

"INTEL'S NETBURST might have become a retrospective embarrassment"

Weren't you the guys who coined the term "Netbust"? Now you are calling the P4 architecture an embarrassment because it was that good??? You certainly are a fickle lot. Your penchant for criticizing Intel any way possible, borders on excessive compulsive.

The P4 was originally designed for high clock speeds because of it's large pipeline, at least that was the idea at the time, or has that escaped your keen technical eye?

No matter, I seems whatever so called 'failures' you editorialize, things do have a (retrospective) way of coming around in Intel's favor. Besides, not to worry, you'll have "Itanic" to beat up on as fall back position. All's not lost (yet) fella's.

SPARKS

I prefer my celery raw.

Any benches on this to report, just out of interest?

does it play Crysis?

(sorry)

To the double-pumped comments above. Actually, the execution core ran at 2x core frequency in the Prescott/Cedarmill designs. Technically, there is logic running at 16 GHz.

1. Italy 8179.89 mhz ThuG OC Team Italy Pentium 4 631 @ 8179.89 »
2. Belgium 8116.19 mhz Blind Celeron LGA775 352 @ 8116.19 »
3. Russian Federation 8113.22 mhz Nordling Celeron LGA775 347 @ 8113 »
4. Russian Federation 8004 mhz DeDaL Celeron LGA775 347 @ 8004 »