Hypercooling for Hyperthreading

Got an email that got me thinking a bit.

Someone wrote me complaining that a particular website used the absolute maximum wattage figure for Athlons, while only using the Thermal Design Power figures for the PIV.

This actually is an apples and oranges comparison. That’s because the Athlon doesn’t have a throttling-down mechanism (as opposed to shutoff) mechanism while the PIV does.

Under normal circumstances, the PIV will not reach the absolute maximum power consumption because it will start automatically throttling down before it gets to that point.

So, under normal circumstances, if you had to use a figure for a general audience, the TDP will probably be a better figure to use.

Of course, that leaves out the throttling-down factor.

Getting Worse

So long as the PIV stays at .13 micron, increasing speed will make the chip hotter and hotter, and make throttling-down more likely, which kind of defeats the whole purpose of overclocking.

Hyperthreading is only going to make matters worse. If the CPU is made to do more work, it will use more power and thus get hotter on average than a non-hyperthreaded CPU.

Intel’s datasheets already hint at this. If you look at the datasheets (page 64 of the document), you’ll see that the 3.06 chews up about six more watts than the scaling of wattage due to frequency increase would lead you to expect.

The maximum case temperature of the processor also drops, down to 69C. That’s probably a good indicator of the point where throttling down begins to occur.

Up to now, while throttling-down has been seen sporadically in PIVs, they haven’t been commonplace. They may become more commonplace next year; especially if you’re trying to get close to 4GHz with a multiple VID Northwood with just air cooling.

The Hypothesis

Next spring, when we see people trying to overclock hyperthreaded, multiple VID PIVs using air cooling, they may well find their CPU throttling down under very heavy load; especially at speeds exceeding 3.5GHz.

Should that happen, my prediction is that those using more advanced cooling techniques won’t run into this problem at the same speeds.

Should both prove to be true, the only way to keep the CPU from reaching throttle-down conditions is to keep the CPU below throttle-down temperatures. That means water, maybe more.

In other words, people may have to go to more advanced cooling not just to try to reach higher speeds, but to prevent throttle-down at those higher speeds.

Describing Tomorrow’s Problems Today

I know a lot of people who do this are allergic to theory. They just want to look at concrete results.

This is the Og approach to overclocking.

Og and Zog see a saber-toothed tiger for the first time. Zog says to Og, “That thing is big and has big teeth. Maybe we should watch it for a while first.”

Og says, “Og is brave. Zog is gay,” charges the tiger, and becomes Stone Age fast food. This confirms Zog’s hypothesis, but doesn’t do a thing for Og.

The problem with concrete results is that they tell you “what.” They don’t tell you “why,” and you need to at least have some suspicions about the “why” before you can effectively find out out what to do about the “what.”

If you don’t, then you’re working blind, and I think you’ll find that working blind is rarely kind.

Theory is not fact. It just gives you a heads-up on where to look first.

Mind you, I don’t know for sure that this is going to happen. I’m just saying that there’s a decent possibility it could happen. Certainly don’t buy a Prometia just because of this article.

When the time comes, there will be plenty of people playing pioneer. If this throttle-down wall doesn’t exist, no harm done outside of a little delay.

But if this throttle-down wall does exist, they’ll run into it. Some will no doubt say “WTF” a lot and replace half their system blindly looking for a solution.

This gives you a headstart on what to look for, and what the proof (or disproof) of it will be. If the air guys hit the wall, and the water guys aren’t (or hit the wall at a higher level), that’s the problem, and you know what the solution will be.

If the solution is unacceptable, holding back a bit to see if this wall exists or not may enable you to avoid the whole extra cost and disappointment of a system that won’t do what you expected it to do.

A little thought and patience can go a long way.

Do you want to be Og or Zog?

Ed