Excuses, Excuses

Prescott: What It Should Have Been And What It Is

Prescott is primarily a shrunken-down Northwood with extra cache and tweaks.

The most notable result of a process shrink is that it reduces power consumption of the CPU, a lot.

Let’s look at the previous two process shrinks for Intel.

First, let’s see what happened to the PIII 550 when it got shrunk from 250nm to 180 nm.

All wattage figure from thermal design power figures from Intel’s datasheets for the PIII (page 51), the Willamette PIV (page 79), the Northwood PIV (page 69), and the Prescott PIV (page 62).

Power Reduction: 250 to 180nm Process Shrink

CPU Speed
250nm PIII
180nm PIII
550MHz
30.8
14.5

Quite a drop, isn’t it? About 50%.

Yes, the Katmai PIII had 512K cache, while the Coppermine PIII had 256K, but the Katmai cache only used 1.37 watts, so the power savings due to less cache is minimal.

Next, let’s see what happened to the PIV 2GHz when it was shrunk from 180nm to 130nm

Power Reduction: 180 to 130nm Process Shrink

CPU Speed
180nm PIV
130nm PIV
550MHz
75.3
52.4

Not quite as much a percentage drop, but still 30% (and it did include a doubling of cache).

Finally, let’s see what should have happened to Prescott, and compare it with what did happen with Prescott.

Let’s be conservative and assume the same level of power reduction for Prescott than occurred with Northwood. We should have seen power consumption figures like:

Expected Power Reduction: 180 to 130nm Process Shrink

CPU Speed
130nm PIV
90nm PIV
2800MHz
69.7
48.6
3200MHz
82.0
57.6
3400MHz
89.0**
62.3

**Wattage figures for Northwood 3400MHz included in latest revision of Intel’s datasheet

Instead, we got this:

Power INCREASE: 130 to 90nm Process Shrink

CPU Speed
130nm PIV
90nm PIV
2800MHz
69.7
89.0
3200MHz
82.0
103.0
3400MHz
89.0
103.0

Let’s show this another way, comparing what the wattage figure reasonably should have been vs. what they actually are:

Expected Vs. Actual

CPU Speed
Should Have Been
Is
2800MHz
48.6
89.0
3200MHz
57.6
103.0
3400MHz
62.3
103.0

Do you see why there is something terribly wrong with the Prescott process?

The power consumption figures are terrible. From these, you have to assume at best a 130W 4GHz Prescott. Any overclocker adding just 10% extra voltage is looking at 160 watts. That’s just for 4GHz. Mr. Freezes should have no problem exceeding 200 watts.

Can’t Overclock, Either…

Ed

Can’t Overclock Much, Either

Please note the overclocking effort worth noting at this point was the attempt made at VR-Zone (several places simply refused to disclose their overclocking, and I doubt it’s because the results were just too good). Don’t just look at the CPU-Z results, the author says he can’t get the system really stable above 3.8GHz with high-end air.

Unless you already have a Northwood system with water cooling or more; it will be extremely difficult for an overclocker to justify buying one of these things. They just can’t be cranked up enough using high-end air.

Missing The Trees For the Forest

The issue with Prescott is not how well it does clock-for-clock against Northwood. Even if everything had gone right (which it certainly didn’t), one could not have expected much more than a 5-7% improvement over Northwood.

The issue with Prescott is not even the extension of the pipeline. Going to 31-stages is a justifiable tradeoff provided you can ramp the beejeezus out of the new design, but right now, it can’t ramp because of the power/heat.

The issue with Prescott is the power problem. The truth turned out to be even worse than the rumors. Prescott chews up far too much power and puts out far too much heat for what it does. That’s what makes this processor a dog. Dog as in canine. Fido. Bow wow. Woof woof.

The power problem is a killer. Intel can scrape through the Prescott generation with this, barely, but it simply can’t go on from there with what they have in place now. 31-stage pipelines do you no good if the processor melts before you can take full advantage of it. You can’t put out a 5 or 6GHz CPU for the average Joe with these levels of power consumption.

And mind you, this is the best Intel could do after a year’s delay.

Socket T? All it looks like it will do is keep your mobo MOSFETs from melting while keeping the hot dog juiced.

We’ve noted before what the probable problem with Prescott is, and that it won’t be fixed until Tejas.

So Intel knows it has to fix this. Michael Dell isn’t going to buy Intel chips that need a Prometeia just to run stock.

Kissing Up…

Ed

Kissing Up To Intel

No doubt you’ve seen other reviews. Didn’t you get the impression from at least a few of them that they felt it their job to defend the chip? To make excuses for it? To spew masses of positive trivia to the point of rupture while ignoring the negatives as much as possible? To essentially say, “Look at all the wonderful things Intel did to make this almost as good as a Northwood” and finally conclude that “Yes, this processor is a dog, but it’s such a beautiful Irish Setter!”

Incredibly, there are huge “reviews” out there that do not say a single word about the power/heat problem. This is work worthy of the Hell Tourism Board, I bet they don’t mention temps, either.:)

Why?

I mean really, do such people work for Intel, or you, the reader?

Which is it?

I don’t see how people can maintain credibility when they do that. I don’t see how people don’t or can’t see it.

2004: A New Competition Between AMD and Intel

It seems like AMD and Intel made an agreement. Intel went to AMD and said, “Just for a change of pace, instead of trying to outdo each other, let’s see who can f— up more this year.”

AMD agreed, figuring it had a lock on that, then Intel blindsided them with Prescott. AMD said, “This is going to be pretty hard for even us to beat. Hmmmm. Until we can screw up our 90nm process, why don’t we just price the s— out of our Hammers so nobody will buy them, and obsolete the ones we have out there?”

I know they’re not doing that, but what’s the competition going to be for most of 2004?

An overheated CPU versus an overpriced one.

You want to go away until 2005? Could I join you? Please?

🙂

Ed

Be the first to comment

Leave a Reply