A Crisis In Counting

Initial benchmarking of Intel’s first dual-cores are in.

In the pre-game analysis, we had been saying that dual-cores were likely to greatly advantage only the relative handful of applications designed to be multithreaded, plus offer “a smoother ride” when a lot of programs were up and running at the same time.

Most places reviewing the system focused on aspect one, and found pretty much what we expected.

Out of the early posters, only one place tried to test the second part of the prediction, and that’s where the big surprise emerged; dual cores did far better when you did more than one thing at a time. Now. Today.

Getting multiple tasks done faster is the main advantage of dual-cores now for the average computer user, and it will probably stay that way, even in the long-run.

This creates a crisis in counting. Current benchmarking will proclaim higher-speed single cores to be better than duallies, but for many, maybe most, reality will say the opposite.

Single-minded = Simple-minded?

The reality is different people use computers in different ways, and even the same person uses them in much different ways at different times. If you are fighting for your life in a typical single-threaded game, you don’t care that you can’t download a lot of emails at the same time. But you will after the game is over.

The CPU companies are presenting us with an either/or situation: You either get better single-app performance, or you get much better multitasking. The best choice for you depending on how much you do in either mode. There is no single right answer. If all you do is play games, high-speed single may well be the way to go, but the opposite will probably be true for your teenage female cousin who thinks Doom III is getting three pimples before a date.

Anyone who says that one is better than the other for everybody is just . . . well, stupid.

Unfortunately, there are a lot of stupid people out there in geek land, not generally stupid, but stupid in certain ways.

One very common mental flaw out there (and one hardly restricted to geeks) is “if you can’t easily and exactly measure it, it doesn’t exist,” and that’s just what dual-cores don’t give you. It is easy to measure single-program performance. It is not so easy to measure the performance of a mix of programs, and any measurement is bound to be somewhat approximate because people will have different mixes.

I’ve observed over the course of time that when the measurements get tough, the measurements get going . . . out of use.

Nor is it a revelation that people who are used to doing things a certain way tend to keep doing them long after reality suggests a change, especially one which requires more work.

None of this is meant to fault anyone who did this first set of reviews; this is a new phenomenon. I just fear that the bulk of people will just keep doing things the simple old way when the simple old way has been greatly compromised.

There are also biases and prejudices to be considered. Many may well rebel at the notion that something running at a slower speed can actually be faster in some way which doesn’t easily reveal itself in numerical form. Those who want to believe that will no doubt find enough benchmarking factoids to back the bias (which admittedly is accurate sometimes).

Sometimes. Not all the time.

For the next few years (at least until the PC world goes completely dual-core), a fair and accurate comparison between single and dual-cores is going to require both single-program and multitasking testing.

Dual-cores are going to require two-dimensional reviews and benchmarking. One size won’t fit all.

Let’s see if we get it.

Overclocking: Having Your Cake And Eating It Too?

Even the best, most accurate measurements in the world will still leave most purchasers with an unpleasant choice: Do I want a better system for this program, or do I want a better system for all programs.

Most would prefer not to have to permanently choose between the two, but rather switch between the two depending on what they’re doing.

Here’s where overclocking can come to the rescue and let people have their cake and eat it, too.

Dual-cores will present the biggest power and heat challenges we’ve ever faced, but provided the mobos can supply the increased juice and (reasonably cheap) cooling can get rid of the heat, people will be chewing some cake.

Do not be at all surprised to find out in the next couple years that being able to shut down one CPU while running the other full-blast proves to be rather . . . easy. CPU architects have been talking about that more than a little, and no doubt are planning the necessary circuitry to do it for at least some chips.

Don’t be all too surprised to find that each core of a dually can be made to run at different speeds, either. That would obviously complicate multi-threading, but there could easily be CPU modes that preclude it.

We don’t know exactly what the future will hold, but it’s almost certainly the case that overclocking will be a much different beast two years from now than it is today.

Ed