There’s been two projects I’ve promised that I haven’t really mentioned for a while, and this is to let you know what’s happening with them.
Comparison of KT266A Boards
I’ve run some tests and comparisons of the boards, but plenty have done that and more.
The point of this website is not to do what everyone else does, but rather to do what they don’t do.
In this area, I don’t think the important issue is whether or not one board does 2% better than another running at current speeds, but rather how well they handle future demands.
The obvious one is FSB and higher RAM speeds. The next jump is to DDR333, and while hobbyists have cranked up their FSB to that and beyond, they’ve generally had to make the self-defeating move of cranking down their memory settings to do so. This really doesn’t test the FSB all too well.
There’s a few vendors out there offering what they’re calling DDR333, but up until very recently, it usually has just been high-end DDR266.
Even manufacturers that say they now have DDR333 are engaging in some questionable practices. For instance, Kingmax says they have DDR333. However, if you look at this forum thread, you’ll see that
Kingmax DDR333 modules use two different types of memory chips.
One type is marked -7A, the other -5, and which one you get is a roll of the dice (and whether those -5 are really -5 is questionable, too).
It also doesn’t help that most of the places offering it have less-than-admirable ResellerRatings.com numbers, and the one that does says 5ns in the Pricewatch ad, but doesn’t say that in their actual description of the product on their website.
(Helpful Hint: When this sort of situation comes up, it is important to see and keep a copy of the exact description of the item being offered for sale. In this particular case, since they didn’t say they were going to give you a module with -5 chips, if they ship you one with -7A chips, they’ll likely stick you with a restocking fee if you return it because “you thought” you’d get the other. If you order from a place that has -5 as part of the description, you have proof that they actually did send something other than what they offered the wrong part and thus can’t stick you with a restocking fee).
Fortunately for my piece of mind if not my wallet, Corsair announced a whole two days ago that they now have DDR333 modules. Even better, the CM64SD256-2700CX2H module is rated at CAS2 for that speed. Guess what I ordered today.
Not saying you should; I’m sure we’ll see cheaper, and see it sooner rather than later, but at least I’ll be able to seriously push the boards.
The less obvious one is the increased bandwidth available for hard drive/PCI devices. The best way to see what’s going on there is by using RAID to push the PCI and ATA specifications a bit. Indeed, there’s been whispers that the Via chipsets don’t handle these circumstances too well at the moment. To do that, you need hard drives that can do such a thing, and until now, there really hasn’t been.
Fortunately, again, the IBM 120GXPs have just become available, so I did a Noah’s Ark of 40GBers.
I should be getting these about the middle of the week, then I should be able to tell you something you don’t already know about these boards.
I asked about this a while back, and got about eight trillion parameters to try out. Not going to happen, folks. I’d like to get this done before our Sun becomes a red dwarf.
The idea is to come up with something simple that does a reasonably good job measuring quality in a reasonably objective way. So while I’ll probably use some of the suggestions I got as a sort of cross-check (and we may well find out there is no simple test that’s any good), we’re not going to crank out reams of data nobody will read.
Another reason for delay has been our inability to get the GF3 we wanted. To make a long story short, the card we wanted got discontinued, and we were shipped the wrong one. Given that the GF4s will be announced in a month, I decided to wait for those since nVidia is changing the reference specifications to require high-quality RF filters for them.
Now before the Radeon fans start screaming, if ATI comes up with a new Radeon by then, we’ll certainly look at that, too.
(I have been using a Radeon 8500 in dual-monitor mode the last couple weeks, and while my old Matrox G450 does a better job handling two monitors (the Radeon does a little tearing of webpages and the second monitor flickers when I scroll webpages on the second monitor), the difference in image quality is at most slight, and the Radeon runs 3DMark2001 over six times faster.)
In the meantime, what I’ll be doing over the course of time is exploring what does matter or doesn’t in visual quality.
My suspicion right now is that most of any real, general problem is due to RF filters. Most complaints on that score have been with nVidia, so I suspect that at least for 2D, that question will become moot with the GF4s all using good filters.
This will still leave how well the various vendors handle 3D modes, and that could prove very interesting indeed.
However, the impression I’ve gotten from comments I’ve gotten and read in various forums is that a huge chunk of “visual quality” is in the eye of the beholder. For instance, a video card may actually be less accurate in rendering pixels, but provide more intense coloring and thus look “better” to many people.
Some Good Reasons For The More Timid To Delay
The last .18 micron Palomino will be the 2000+. From preliminary testing so far (I’m sure we’ll see much more Monday/Tuesday), there’s not a lot of overclocking room on this CPU. If you’re interesting in this primarily because you don’t want to futz around with unlocking, you would be much better off waiting for the .13 micron Thoroughbred that should be coming out at the same speed in three months.
A few Northwood reports are appearing here and there. There was a report of 3GHz being reached, but what wasn’t as noticed was that it took 1.9V to do it. Since the PIV DDR boards usually have a maximum range of 1.7-1.75V, unless voltage mods don’t faze you (nor throwing 25% more voltage than recommended into a $600+ CPU), I wouldn’t count on 3GHz from the first bunch of these at all.
It’s not even clear yet whether 2.67GHz (which is what a 2.0A Northwood would do at 133Mhz FSB) is doable, though that seems to be a lot more likely.
Unfortunately, the Northwood reviews we’re likely to see in the next few days will probably have precious few overclocking results, and probably will use the 2.2GHz CPU instead.
Unless these things are really awful, the 2.0A will easily be the best overclocking bet. It’s more likely to hit 2.67GHz with conventional voltages than the 2.2GHz will hit 2.93GHz, and if these prove better than expected, moving the FSB towards 150Mhz shouldn’t be too big a deal for the rest of the system.
Unless Intel has a last-second surprise, the only two Northwoods to be introduced Monday are the 2.2GHz and the 2.0A GHz. There seems to be a lot of vendor confusion about this. At least one place is calling its 1.8 and 1.9GHz chips Northwoods, but if you go to the site and the description of the chips, they’re not.
If it’s a 2.2GHz CPU, it has to be a Northwood.
If it’s a 2GHz CPU; it may be a Northwood, it may not. The Northwood version is supposed to be called a 2.0A. In any case, all Northwoods have 512Kb L2 cache, the early CPUs have just 256Kb.
Anything under 2GHz is not supposed to be a Northwood (if that changes, we’ll change this).
In all likelihood, though, while these chips are supposed to be released at 3GHz by the end of this year, it’s likely that overclockers will have to wait for at least one new stepping to hit that speed without extreme measures.
It looks likely (especially for the Northwoods) that prices a week from now will be lower than they are now. For instance, the low Pricewatch price of the Athlon XP1900+ has dropped $13 in the last couple days.
As noted above, the GF4s are supposed to be announced in a month. The high-end models will be a good deal better, and a lot more expensive. The ones to look for will be in the middle of the pack, since they look better on paper than anything available right now. Of course, the introduction of the GF4s will put downward pressure on GF3 prices.
In the next month, it’s likely that true DDR2700 will start becoming commonplace. They’ll probably go for a premium until Crucial shows up, but even with escalating DDR prices, you’ll probably still save a bit. I don’t think we’ll see $35 for 256Mb sticks anytime soon, if ever, but I think we’ll see prices jump up some more for a little while, then head back down.