The next great overclocking frontier. –Ed
There’s one game being played out by Intel and AMD by everybody else. Then there’s the game as played by overclockers.
The official roadmaps show AMD getting Palomino up to 1.733Ghz by early next year. They show Intel getting up to 2.4Ghz.
But that’s not what overclockers want to know. Official ratings are for other people. They want to know what they can do with these chips.
Palomino Won’t Overclock Much
AMD isn’t stretching out introducing speed ramps because they’re getting tired of product launches. They aren’t adapting a PR scale just because they like it. They’re doing these things
because they can’t push the product much more at .18 micron. They know it, and they’ve known it a long time. Back in January, Jerry Sanders said not to expect 2GHz until 2002, and the AMD roadmaps show the Palomino falling well short of that plateau just when they need the numbers the most.
Unless extreme measures are used, I don’t think the average overclocker is going to hit 2Ghz with one. My best guesstimate is that most folks will get somewhere between 1.8-1.9GHz.
There will of course be a lucky few with choice chips, and a greater percentage may hit that point next spring, but there’s just not going to be a lot of headroom on these things. You want overclocking room, you want Thoroughbred (AMD’s 0.13 micron desktop) next year.
A couple days ago, Intel got a Northwood up and running (though not doing much) at 3.5GHz.
How important is that?
Well, it’s not “the sky is falling” material. They didn’t have the CPU do too much at 3.5, and supercooled it.
These demos don’t tell you what’s likely to happen any time soon. However, it’s an excellent indicator of what the top-notch Intel processor is likely to do a year or so from now.
For instance, last year a Willamette was demoed at 2GHz. Guess what Intel just delivered.
Intel has also been passing around the word that they expect to get to 3.5Ghz without too much trouble, and 4Ghz eventually with the PIV at .13 micron. Such estimates have been pretty reliable in the past, and given that this is a chip designed to ramp up better than previous Intel generations, seems pretty reasonable to me.
You might say the roadmap doesn’t show such optimism, with just a 2.2GHz and 2.4Ghz processor on tap. This is true, but keep in mind these chips won’t have any AMD competition MHz-wise for at least the next six months.
Matters will heat up once AMD does come out with Thoroughbred, and then I think you’ll see Intel ramping up.
Northwood has all the attributes of being a big jump forward. It’s the first member of a new generation of processors using a new process.
The original PII, the 233, was usually able to reach a speed of 300Mhz or a little more, or roughly an overclock of 33%.
The original Deschute PII, the 333, could reliably get into the range of 400-412Mhz, or about 20-25% above rated speed.
The original PIIIs, the 550/600, could reach 800Mhz with little sweat, at least a 33% overclock.
(I’m not considering Celerons since they’ve usually been PentiumX cores that have only come out after the PentiumX cores have been tweaked a bit, so they’ve tended to O/C by a greater percentage.)
So 33% seems to be historically what we’ve gotten from the first one out of the gate.
The first Northwood PIV we’re going to see is going to be a 2.2Ghz Northwood. Apply the 33% formula, and we get a figure of about 2.9Ghz.
Will The Mobos Cooperate?
Intel’s own plans for Northwood mobos will start off with a familiar 100Mhz quad-pumped RDRAM board, followed by a DDR200 board, again, with an FSB of 100Mhz.
This means the original 2.2GHz Northwoods should be multiplier-locked at 22X100.
It’s probable we’ll see at least 133Mhz quad-pumped RDRAM boards once Rambus has come up with PC1066 RDRAM, but that won’t be until next year. At that point, we’ll likely see two flavors of PIV: 100Mhz and 133Mhz, just like we did with Coppermines.
It has not proven to be a problem to get RDRAM boards running at 133Mhz (especially if you slow down the memory speed in BIOS). DDR memory of course is perfectly comfortable at 133Mhz, and if Intel doesn’t want to acknowledge that, Via and SiS certainly will.
At least at the beginning, you’d probably want a mobo that provides a /4 divisor at speeds lower than 133Mhz just in case the Northwood isn’t quite up to 2.9GHz, but that’s old hat to the mobo wizards of Taipei.
What Does 2.9GHz Really Mean?
We know enough to come up with a fairly rough estimate of how an overclocked Northwood will fare against an overclocked Palomino.
At this point, all we really want to know is whether or not Northwood will become a must buy and an AMD-killer from a performance standpoint or not.
If you simply cannot stand estimates, or cannot think without numbers set in granite, that’s not being smart, that’s being anal-retentive. Go away. 🙂
Right now, a 2Ghz PIV is roughly even overall with an AthlonMP running at 1.4Ghz. Since an AthlonMP looks to be an equivalent of the Palomino, we need not make any adjustments on the AMD side.
Northwood is supposed to have a bigger cache and some other improvements which are supposed to improve performance 5-10%. Since I’m going to make an adjustment to processor speed later on, and since performance doesn’t scale perfectly with MHz speed, I’m going to take high estimates, so let’s take 10%.
Running at 133Mhz rather than 100Mhz will also improve performance. 3% is probably a good overall figure, but since we have a lot of gamers, I’ll take 5% as a more reasonable estimate for this audience.
10% + 5% = 15%. So a Northwood at X speed is the equivalent of a current PIV running at X + 15%.
Let’s go back to our 2.9Ghz estimate. Add 15% to that, and our 2.9Ghz Northwood works out to roughly the equivalent of a current 3.3Ghz PIV.
From our initial statement, a 2Ghz PIV = 1.4Ghz MP. A little algebra, and we see that a PIV MHz is worth only 70% as much as an AthlonMP.
So we need to multiply our 3.3Ghz figure by 70% to come up with an Athlon equivalent. That comes out to 2.3Ghz.
Since we’ve figured the Palomino should be good for 1.8-1.9GHz; 2.3GHz is about 20% higher than that. Since applications don’t scale perfectly, a 20% Ghz difference probably means a real performance difference of somewhere between 10-15%.
In other words, significant as these things go, but not a killer. The Intel advantage will certainly be much less than the Athlon had over the PIII, and roughly what the Athlon had over the early PIVs.
If the 2.9GHz estimate is too low, obviously the numbers get worse. Figure roughly 2.5% for every 100Mhz above 2.9Ghz.
I think it more likely, though, that these chips won’t start reaching close to 3.5GHz (3.3Ghz?) until there’s been one stepping change (probably some time in the spring).
I’ve left out one wild card, and for once, this one is in AMD’s favor: nForce.
If nForce gives Palomino a 5% boost over current mobos, now we’re not talking 10-15%, but rather 5-10%.
This isn’t the greatest news in the world for AMD, but it’s hardly as bad as 2.9Ghz vs. 1.8Ghz would lead dimwits to believe.
If you are loyal to AMD (or just stuck to them due to a recent buy), this is no reason to personally panic.
Northwood pricing is unlikely to be even in the same ballpark as Palomino pricing until spring is ready to spring.
By that time, Thoroughbred should be around the corner. At the least, that should more or less level the playing field.
So it’s likely to be a bad winter, but you won’t freeze to death.