Pentium 4 vs. Celeron

[seacue]

In an earlier life I made my living trying to develope techniques to quantify computer performance. Thus I know that no one number can express "performance".

We see multigigahertz Celerons and "slower" Pentium 4s.

I'm so naive that I don't know the architectural differences between the "C" and the "P4".

Knowing these difference one might guess how they would impact performance on a specific task. Can the higher clock rates often seen in Celerons compensate for the architectural difference in some applications?

Any takers? Naples?

SC

[BayAreaLondoner]

seacue - this (http://www.anandtech.com/cpu/showdoc.html?i=1927) might be a useful article to read.
Brief description of differences between a P4 and a Celeron as well as AMD CPUs, together with some ratings.

[seacue]

Thank you BALon.

Perhaps not your reaction, but it appears to me that their data shows the higher clock rates of the Celeron goes a long way toward making up for the architectural advantages of the cheapest P4.

They certainly claim that AMD is selling more performance at this price point.

SC

[Naples Air Center, Inc.]

seacue,

I see BayAreaLondoner reads the same articles I do. The article he referenced is an excellent resource. I have a lot of respect Anand Lal Shimpi and his reviews.

When the PIII was Intel's flagship, I only sold PIIIs (and Cels with the PIII core). When Intel released the PIV, the numbers on the CPU were so dismal, I stayed with the PIII. Intel made some real compromises with the PIV and most of the early PIVs were slower than the PIIIs. When tne first nForce Chipsets came out for the AthlonXP, I switched from the PIII to the nForce/AthlonXP and I have never looked back.

Intel only caught up, if you want to call it that, with the PIV HT. Then Intel comes out with the PIV Extreme Edition, I find it hard to call anything Extreme when an Athlon running at 1Ghz less can keep up with it.

In order for the PIV to work a its full potential, it needs to a lot of memory bandwidth. The PIV based Celeron is a "Crippled Core" PIV. When Intel has a PIV which the die is bad in one of the areas for the on chip memory, they lock off the bad side and just use the other side for the onchip memory. That is a Celeron.

You can imagine how dismal the Celeron's performance is since the design is very memory hungry and it only has half the memory of a PIV.

I hope that was what you were looking for and I was not just ranting on.

Take Care,

Richard

[seacue]

Additional thanks to Naples/Richard.

Since the typical Celeron has a high clock speed, am I to assume that slower PIVs with sick memory are converted into sand, not into Celerons? That would be in recognition of the architectural problems.

Maybe they also use this cripple technique for PIVs which are sick enough that they can't multithread.

There must be a bunch of fuses to select which sections to keep to convert a PIV into a Celeron.

Thanks for the insight.

A bit off-topic, but I wonder about the yield on such complex chips. Dull 7400-series chips approach 100% yield. Yield of fully functional PIVs must be much, MUCH poorer if Intel goes to the trouble of laying them out in a way that they can shed bad sections and make Celerons.

Bob

[Naples Air Center, Inc.]

Bob,

Intel has been doing the same thing since the first PIIIs went to Socket 370 Form Factor. (Might be longer.)

From Anantech:

The new Celeron core is based on a 128KB L2 version of the original Willamette core that the Pentium 4 debuted with in November of 2000. Unlike previous-generation Celerons, the Willamette-128 core is no different architecturally than the Pentium 4's old Willamette core. The cache organization and mapping algorithms are still the same, the only difference is that the Celeron core is only outfitted with a 128KB L2 cache instead of the 256KB cache present on the original Pentium 4.

Only having a 128KB L2 cache increases the Celeron's dependency on a high-speed memory bus.

Here is an article from Tom's Hardware on the latest generation of Celery:

The New Generation Is Here: Celeron 2.0 GHz, with 0.13 µm (http://www20.tomshardware.com/cpu/20021016/index.html)

(You can see a picture of the Core in the article.)

Here is an article also from Tom's Hardware on the last generation of Celery:

Battling Brothers: Celeron vs. Pentium 4 (http://www20.tomshardware.com/cpu/20020903/index.html)

(There is also a picture of the Core in this article.)

Take Care,

Richard

[25F]

Richard,
Intel have been selling "semi-functional" chips for a long time. What came to mind immediately was the 486SX:
http://wombat.doc.ic.ac.uk/foldoc/foldoc.cgi?486SX

Other manufacturers have attempted to do something similar with memory modules: fuse out the defective bits and use what works. It makes a lot of sense.

[Naples Air Center, Inc.]

25F,

The PIII Celery was very overclockable and when the PIII was still in the Katami Core, the Celery was within a few %Peroformance, Mhz for Mhz.

But with the corners Intel cut with their PIV, the difference between the two had drastically widened over the years.

Still cannot beat the AthlonXP 1800+ TBred for band for the buck though. Nothing like spending $51.00 on a CPU that you can overclock an additional 800Mhz. (From 1.53Ghz to 2.33Ghz with proper cooling.)

Take Care,

Richard

P.S. I thought it was a very good bit on ingenuity, finding a use for the low yield processors.

[ORAC]

The Intel core was designed to be scaleable beyond 10Ghz, so it's not at it's best down at the 2Ghz level. They were thinking long term when they designed it.

Think of it in terms of a 12 cylinder car engine working on only 4 cylinders, it wouldn't compare well against a purpose designed 4 cylinder engine, but when it's fully running....

The P5, to be introduced in various speeds starting in early 2005 is known as the Tejas. Speed will start at 4.5Ghz and move up to 9Ghz by the end of the year. Die size is 90nm, Bus speed is 1066Mhz.

The version to be introduced in 2006 is the Nehalem, based on a .65nm die and a 1200Mhz bus, with rumours of a 4Ghz bus version. The starting speed will be 9.6Ghz with 10Ghz+ versions to follow shortly thereafter.

Stick with the AMD for now, it's the best price/design for the current average CPU performance. But I'll be watching with interest to see what AMD can produce to match them.