Polling and Databases

Stuffing The Ballot Box

A few weeks ago, someone wrote me and said that we weren’t a horribly popular website because we had only gotten a handful of votes in some poll, and that some other website
had ninety times what we had.

So, just to see what would happen, and to get a little publicity for the site, I put a little blurb on the top of our page talking about this poll, and asking you to vote for us if you thought us worthy.

Well, we ended up a strong third as a result. You can see the results here

Let me tell you what this meant and what it didn’t.

What it did mean was that we have a lot of people who believe in this site and were willing to take some time and effort to tell the world that this is a pretty good place to visit. I thank all of you for those votes of confidence; it bolsters our spirits.

The website must have liked the hits, since it apparently decided to run the same thing all over again. Enough for us, we made our point.

Anyhow, what the results don’t mean, of course, is that we’re the third most popular computer website, and that’s something important to realize when you look at most Web polls.

It’s pretty easy for a group of determined people to get together and push a poll anyway they like, especially on the Web. The people who did best in that poll were the ones who talked about it on their sites.

I went to one place I had never heard of that finished only a bit behind us, figuring I might find something good there. There was nothing there! A few news posts and a forum, that pretty much was it. But, if you looked at that poll and took it seriously, you’d think the place was a powerhouse compared to little also-rans like Anandtech and Tom’s Hardware.:)

Didn’t we do the same thing? Of course, but we know enough not to take it literally. We just hoped a few people would see that, check us out, and stay awhile. That’s all.

What You Can and Cannot Do With Polling

There’s another poll we’re involved in, though you probably don’t think of it that way. Matter of fact, we run it. It’s called the CPU Database.

More than a few of you over time have commented that the database cannot be statistically accurate because people will be more reluctant to report overclocking failures than successes.

You’re probably right, but it doesn’t matter. Here’s why:

This Isn’t A Presidential Poll

We’re in the closing days of a fairly close Presidential election. It matters a lot in this situation to be as accurate as possible, because if you’re off 5% or 10%, you end up picking the wrong guy.

There’s no need to be that accurate in figuring out how likely it is you’re going to be able to overclock a CPU. You don’t look at our database saying “well, if there’s a 72% chance it will work, I won’t, but if there’s a 75% chance, I will.” A few percentage points makes no difference, unlike a Presidential elections.

Secondly, the nature of how CPUs are made means you don’t get close calls. CPUs are designed to reach up to a certain speed. Unless the manufacturing process is really bad, most of the chips sold will reach that speed, no matter what their official rated speed is. Go above that speed, and the success rate drops dramatically.

You just don’t see situations where a CPU will reach a certain speed half the time. It just doesn’t happen. Generally, the CPU will either hit a certain speed 75% or more of the time, or 25% or less of the time. In other words, it probably will or probably won’t, you don’t get maybes, and that’s what you get from the database, not exact percentages.

So if you see that 75% or respondents don’t reach a certain speed, the absolutely correct percentage is probably a bit higher than that, but whether it’s 75 or 80 or 85% matters little, what matters is that it’s probably not going to work for you.

It’s not a matter of “it has to be absolutely accurate, or it’s worthless.” It’s accurate enough for the purpose of determining whether you probably will or won’t succeed.

Of course, there’s a number of factors that can change the likelihood of success, often quite dramatically. A new stepping of a processor is the main one. Reaching 850Mhz with a cA2 stepping 600E was pretty unlikely, but pretty likely with a cB0 stepping cB0.

Cooling can help, too. If your goal is 50Mhz more than the apparent limit of an air-cooled CPU, adding a Peltier or two is going to improve your odds quite a bit. If you stick with your fan, somebody else hitting a higher speed with a Peltier or two isn’t going to improve your odds one little bit.

Refining the Raw Material

It’s harder to interpret the database now than it was in the past, simply because people have more choices. People use water- and Peltier-cooling a lot more than in the past. People just don’t use BX boards anymore; the type of motherboard you use can certainly affect your overclocking results.

A new factor that emerged with Coppermines was that memory often became the bottleneck. If you stuck with your old PC100; that usually didn’t work too well if you were trying to run at 150Mhz. Video cards presented a problem on BX boards too, some of them didn’t take too kindly to 133Mhz+ speed, either.

So what we’re going to do within the near few weeks is to give you more interpretation of what the figures in the database mean. We’re going to analyze the numbers and tell you what you are likely to reach given various setups, so you don’t have to juggle all these things around yourself. In English.

Consider it our early Christmas present. 🙂

Email Ed


Be the first to comment

Leave a Reply