Did a little perusing of our CPU database this morning.
First, a few tips on how to look at it:
1) Only pay attention to those entries that have your class of cooling. If you’re using the cooler that came with the chip, don’t look at the Peltier and water-cooling scores. You’re not going to get those kinds of results.
2) Look at what the typical score is (i.e., where a lot of entries get clustered around). A lot of people look at the highest score, and think they’re entitled to at least that. Wrong. Usually, the highest scores among a specific class of cooling come from “lucky chips,” which are chips that just do somewhat better than the norm, just like some people are more talented at certain things than others.
Expecting to get a “lucky chip” is unrealistic. You should base your expectations on what the average person doing pretty much what you’re doing is getting. If that doesn’t satisfy you, you’re probably better off waiting until the chips improve.
For instance, expecting to get 3GHz right now from a PIV chip using high-end air is very unrealistic for 1.6/1.8A, somewhat unrealistic for 2.26s. A few are doing it, but the average person isn’t, and those who are doing it are throwing more voltage at these chips than may be wise.
3) Pay attention to voltage. There’s no point in looking at results at 1.85V if you’re never going to use 1.85V.
If you see a lot of people reaching a certain speed at default voltage, that’s really an excellent indicator of what you can count on, and what the chip is really capable of.
If you plan to throw more voltage at it, fine, but once you get past, say, around a 10% overvolt, you are really forcing the chip into speeds it really wasn’t meant for, which may have some long-term reprecussions.
When a CPU is reaching its limits, you’ll often find that you have to increase voltage a lot to get little gain. It’s not worth it. If you have a PIV that will do 2600 at 1.55V, but needs 1.85V to do 2700; it’s trying to tell you something. If you’re out to show an inanimate object who’s boss, you can make it eat more voltage, but it can stop you dead in its tracks by dying.
4) Look at dates/stepping/codes. If you have a new processor, looking at results from older chips is often useless. For instance, 1.6A PIVs got better starting around week 7 this year. Looking at scores from before then would be misleading and lead you to believe 1.6As were less capable than they now were.
It’s important to look at dates because, if you don’t, good CPUs can be overlooked and/or bypassed. Generally, when a processor comes out (or reaches a somewhat affordable level), early adopters will try it out. If it doesn’t do too well (and especially if it’s more expensive than other alternatives), it gets bypassed for the cheaper alternative.
Should the CPU manufacturer improve the process a bit later on, if you don’t look at dates, the database often looks like the cheaper processor actually outperforms the bypassed processor, which is almost never the case.
This happened with the PIV 1.8A. When it and the 1.6A came out, they both did about the same, so people went to the 1.6A. When the 1.6A got better, there were plenty of new 1.6A entries, but very few new 1.8A entries.
So a casual look at the databases would give you the impression that 1.6A were great, and 1.8A should be avoided, when if fact, they probably were about the same.
Now that the 1.6As have been discontinued (and those remaining carry a bigger pricetag), people are starting to buy the 1.8As, and they’ve “suddenly” gotten “better.”
5) Look at types of mobos, not specific models. Motherboards tend to come in flavors of the month; the prevalence of a particular mobo is often due more to marketing than superiority.
If you see the top fifteen results coming from one particular mobo, and everybody else has much lower scores, that likely means something. However, if the top scores are 80% Brand X and 20% Brand Y, and you don’t find the bulk of Brand Ys way down on the list, odds are the boards are equivalent.
It’s better to look at classes of mobos, say 845E mobos as opposed to say, SiS mobos. This will give you a clearer idea on what really is better.
Now that I’ve taught you how to fish; here’s what I caught this morning.
The most recent entries seem to indicate the 1.8s are doing about as well as the 1.6A, but things start getting dicey above 2700MHz, with most scores above that requiring over 1.7V.
What’s interesting about the 1.8A is not what it is now, but what it will end up being. Unlike the discontinued 1.6A, it will be in production through the C1 stepping change. A C1 stepping 1.8A will probably be the budget CPU of choice for the Christmas season, and should be able to get at least close to 3GHz.
While there’s still not a lot of entries, those that are there clearly indicate that the 2.26 is a cut above the 1.6A and 1.8A.
2800-2900 MHz looks pretty achievable at reasonable voltage levels using high-end air, which is a couple hundred MHz more than its lowlier relatives are getting. That’s enough of a difference to indicate to me that Intel is either binning a bit, or doing something a bit different with these CPUs.
3GHz looks a bit dicey, though; a lot more doable than with the others, but no certainty, either.
The Near Future
Intel will have its big price cut Labor Day weekend. This will reduce the price of the 2.4B to around $200 (the 2.26 is supposed to be the same price). The 2.4B will also be one of the first C1 stepping chips, so this will be the guinea pig to see if the C1 stepping will make 3GHz easy.
What we don’t know (and won’t know for quite some time) is whether the higher rated C1 stepping CPUs will continue to do better than the lower ones, much as the 2.26 does better than the 1.6/1.8. We won’t see C1 stepping 1.8As until October at earliest, quite likely later than that.
If the higher-rated chip do better, then the Christmas overclocker will be the 2.4B (or maybe the 2.26).
If it turns out all C1 stepping CPUs do about the same, then it will be the 1.8A.