Update Math

by Steve Ciarcia

uying a PC has always been confusing. I guess we can credit the combination of massive marketing efforts and bloated operating systems for conditioning us to think that we have to upgrade our PC every 12-18 months. We've had at least seven generations from the venerable old PC, to the '286, '386, '486, PI, PII, PIII, and soon the P4. That's not counting competing brands and same-generation clock speed increases.

Don't get me wrong, for CPU-speed fanatics who design video processing software, a 5% increase in processing speed is worth getting excited about. They're also willing to pay virtually any price to get it. If you are a CPU-junkie (I'm a car fanatic so I'm not throwing rocks), forget everything I say here and enjoy the rest of the magazine. For the rest of us, however, it might be worth a little price-performance calculation.

Last week we needed to buy three more desktop systems for the office. They weren't for server applications or anything strenuous, they were simple office desktops tied to the in-house LAN for doing word processing, spreadsheets, and a little web surfing (DSL through the LAN). Unfortunately for the staff, I asked what we needed and got involved rather than just saying go buy them.

There was a time when increased performance was a necessity just to keep up with software feature-creep. I remember my first Windows machine. It was a '486DX-25 running Windows 3.0. I had upgraded from '386SX-16 running DOS. According to the Dhrystone MIPS comparison of these machines, the '486 was 9.3 times faster. I had a few different '486 versions before I got my first Pentium, a Gateway P90. The P90 was benchmarked as 5.6 times faster than the '486DX-25 (52 times faster than the '386).

I can definitely say that the P90 was faster, but certainly not 50 times faster on the latest versions of the same software. Word ran a couple times faster. Of course, it was moving a lot more megabytes of feature-bloat now. And strangely, it also seemed there was now a hard drive access with virtually every keystroke. As with most of that generation, it had to run faster just to stay even.

It took buying my first PII before I started seeing some real horsepower. As for the PCs I've had since, I only remember gradual improvements. The truth of the matter is that CPUs today might contain millions of transistors but, most of the time (and for most of us) they are executing wait loops. The average PC's "useable performance" peaked a while ago. For the mainstream office/small business user of today there is little benefit to endless increases in CPU power. The bottlenecks in PC performance these days are virtually never with CPU speed. Ninety percent of the time we sit in front of our monitor, waiting for something to happen.

Even with high-bandwidth Internet connections, I doubt a 1-GHz PIII loads a webpage any faster than one running at 500 MHz. The time taken for opening and closing documents probably won't change either. It typically isn't the CPU's fault. Disk drive accesses and reads are enormously slow when compared to a CPU's capability for processing that data. Although SDRAM and large caches provide the structure for a fast system, inefficient software with too much downward compatibility and operating system resource hogs (like Windows) can negate it.

Deciding what system to buy for the office involved a little re-education. Basically, if you aren't ready for an overkill PIII box (or soon the P4), the only Intel-specific alternative these days is the Celeron. Back when I bought an early PII machine I remember checking out the Celeron. The media described the Celeron as having zero L2 cache and it being a "brain-dead PII."

If that was the history, why was I now seeing so many high-speed Celeron machines offered next to the PIIIs? Well, apparently Intel got the message and they put the 128-MB L2 cache back in and increased its clock speed too. As for being brain-dead-not anymore.

A quick search on "processor benchmarks" provided a little more education about today's market. I'm always looking for a cost-effective solution. It's even better if it turns out to be a bargain. The Intel CPUmark 99 benchmark data gave a value of 45.1 for a 700-MHz Celeron and 63.0 for a Pentium III. That says the PIII is only 40% faster than a Celeron for the same clock speed.

Before I get tons of mail about my crummy math and that I've overlooked things like branch prediction architectures, RISC versus CISC, and an in-depth discussion about integrated caches, this isn't scientific. I consider this to be a minimum resolution comparison at best. There's also the fact that most Celeron systems are 600-700 MHz and PIIIs are being packaged increasingly at 800 MHz and faster with larger hard drives and lots of extras in the box.

Nonetheless, for a simple Office 97 application where the CPU spends most of its time waiting for the user, the choice seemed clearer than I originally thought. Where I live, an 800-MHz PIII system sells for well over $1000. At the same place, a 600-MHz Celeron system is less than $500. Considering the expense of some of the other things I've bought in life, people who know me might laugh that I'm going to such extreme consideration over a few bucks. I simply smile back. It's not the savings. It's the thrill of the hunt!