Don’t Expect A Terrier to Become A Greyhound With A Different Handler
Most reviews nowadays throw a ton of benchmarking numbers at you and think that’s a review. Sometimes that’s a good idea. Sometimes it’s not.
We’ll provide numbers in part two, but enough people have cranked out enough numbers elsewhere to give anybody a good enough idea of how this board compares to others.
What may not be so obvious or addressed is why the numbers are the way they are, and that numbers don’t even matter much sometimes.
As we’ve said in the past, different companies using the same components are going to get pretty much the same results. The only real exception to that is the occasional botched implementation.
The KT266A chipset represents a modest incremental advance over earlier DDR chipsets. Nice to have, but hardly “drop what you’re doing and buy this.”
If you’re a mobo maker working with a particular chipset, there’s just so much you can do to tweak the thing. If you can get your board to do 5% better than the other guy’s KT266A board, that is truly stupendous work (though hardly stupendous results for that work).
So don’t expect the benchmark numbers to improve very much when other companies come out with KT266A boards. Just not going to happen.
When the numbers are close, put the numbers aside. Other factors become more important in a buying decision: features, flexibility, reliability.
The numbers on this board come out to just about the same as the leading KT266A contenders at the moment. When’s all said and done, it’s not going to trail the eventual leader by very much, not enough to mean anything in real life.
So I’m putting the numbers second because they’re a wash, and looking at the other factors first, because they’re going to be the decisive ones.
The New Isn’t Here And The Old Won’t Leave
We are in a transition period. The next jump is going to be .13 micron processors and DDR333. The present
is .18 micron processors and DDR266. The products we’re seeing now try to push the old to its limits without
necessarily being able to dance very well (or even at all) to the new tunes.
If you need or want something new now, or you’re a hobbyist who likes playing Mobo of the Month, you may not care, and that’s fine.
But if these matters concern you, and you have something relatively recent, this is not a must-have.
The Via KT266A boards fall in between the cracks. This is what Via should have released six months ago. If they
had, it would have gotten AMD out of the chipset business last spring, and it would have been “SiS Who?” Maybe nVidia
wouldn’t have even tried to get into this business if Via had done the job right the first time.
Coming when it does, the KT266A is a case of “too little, too late” for many people. It’s certainly not bad. If you need to build a
new computer now, this is likely your best bet.
But if you already have a DDR board, the improvement you’ll get isn’t really worth the money over what you already have, and
while most these boards are at least theoretically capable of DDR333 speed and beyond, memory is a bottleneck and will be for
some time to come. Yes, you can slow the memory settings down and reach all kinds of FSB speeds with current RAM, but what’s
the point if your memory scores are lower at 180Mhz than 150Mhz?
The Palomino in many ways is much like the Via chipset. This also would have been tremendous six months ago, but AMD dropped
the ball, too. This is another effort to push old technology as far as it can go.
One big indicator that limits are being reached are the relative performance of the speed grades. While it’s still too early to draw
definitive conclusions, it’s beginning to look like AMD is testing and effectively binning their CPUs. That’s a fancy way of saying the higher-rated
processors run faster than the lower-grade processors.
I’ve been working with a 1500+, and 1600Mhz with high-end air is all it’s going to do at 1.85V, AGKGA code or not.
The higher grades seem to be doing a bit better, and the 1800+ers look to do around 1.7GHz (again, high-end air).
The Palominos also seem to like voltage. I hate the Frankenstein approach, but 1.85V doesn’t look to be quite enough for serious overclocking.
Up to now, getting more meant getting a soldering iron, but some new approaches apparently remove that requirement. More on that later.
In a sentence, the AK31 version 3.1 appears to be a fundamentally solid no-frills motherboard for a bargain price. However, that doesn’t necessarily make it a good overclocker’s board, and there are enough immmediately obvious flaws to make one cautious about an immediate judgment.
Hardware Features (and lack thereof) That Matter
ICS 94228 Frequency Generator This clock generator is capable of /5 and /6 divisors, and has a maximum capability of 233Mhz.
Four DIMM Slots: Good if you need four DIMM slots. Unlike most other four-slot systems, the AK31 doesn’t require registered DDR if you fill them up. We’ll be testing that claim shortly.
Six PCI Slots: Good.
Fan On The Northbridge: Good, but you might want to add some more thermal grease to the heatsink if you want to get wild.
Lots of Room Around The Heatsink Using a Glaciator at the moment, and still have at least a third of a inch clearance on all sides.
CMOS Jumper In The Middle of Nowhere I rather like this. It’s located just below the Southbridge, to the right of the PCI slots, and left of the drive connectors. Much easier to get to than any other board I’ve seen.
No RAID, No Serious Onboard Audio Then again, what do you want for $85?
Drive Connectors All In A Row Picky, but having them all in one location increases cable clutter and makes it harder to disconnect.
Capacitor Right Next To The Power Connection The power connector is right by the ports and keyboard/mouse connector. Unfortunately, so is a lone capacitor. This is a very tight fit, and you’re going to have to take care when disconnecting it.
The manual is very good for a first-time builder. Not so good for technical specifications. The package is barebones: VIA drivers and that’s that. Not even a system monitoring utility (though MBM 5.1 works fine).
Jumper-Free Settings You do everything from the Award BIOS.
Voltage Settings for DIMMs You can set voltage for DIMMs up to 2.7V. Not as much range as the Epox 8KHA+, but better than nothing.
Automatic Heat Shutdown In BIOS You set the temperature, it’s supposed to shutdown the system when it reaches that point. Will it react quickly enough? I wouldn’t bet on it, but it’s better than not having it.
A Voltage Feature That Doesn’t Work This board promises an adjustable voltage up to +0.275V. As I’ve noted, this would be a very, very feature to have for XPs.
Unfortunately, while there’s a BIOS setting for all these voltage settings, nothing above +0.100 works. If you set +0.125 or better, it just gives you default voltage.
Socket A VIA motherboards have been prone to provide a little more voltage than the setting, usually giving you 0.05-0.10V more than the setting. Not this one. This board gives you .01-.03V less than you’ve asked for.
I suppose from a design standpoint, this is preferable, but it sure doesn’t help when you need more power for a voltage-hungry CPU.
More importantly, sending a motherboard to market with a bogus overclocking feature gets major demerits on the Stroligo Honesty Scale. Better to claim not enough than too much.
I’ve seen some comments in a few forums that have said, “This board is getting a bum rap! What’s a little soldering?”
First, when you sell a product that features increased voltage, soldering not required, this tends to appeal to people who don’t or can’t solder. That probably includes most overclockers. People ought to know what they are or are not getting for their money.
Now if you decorate your own Christmas cards with original solder designs, nothing I say is going to stop you from buying this board, and this is not meant for you. It’s meant for those for whom this is a big deal.
Second, folks have recently come up with solderless means of voltage modification. Clips are used instead. You can read about that here.
But I’m sure that even clips are going to be too much for some, and no matter what you might think of their intestinal fortitude, they still have the right to know that they’re not getting what they paid for.
Very, Very Sloppy BIOS “Sloppy” is the best word to describe it. With one exception, the stuff pretty much works, but it often doesn’t tell you that it is.
Using SPD for Memory Timing Destroys Your Memory Scores Set your memory timings manually, and the AK31 does pretty well. Use SPD with the exact same settings, and your memory scores drop 30%.
There’s something seriously going wrong in a BIOS routine someplace. The fix is obviously easy, but why is it even necessary?
Variety Is The Spice of Life? You may like or dislike AMD wanting the initial boot screen to show just PR. Shuttle has addressed this creatively: they give you variety. Sometimes it gives you PR, sometimes it just says XP, sometimes it gives you a different PR, all without you changing the speed of the chip one Mhz. I’ve even gotten an “Athlon XP 800” message.
Shuttle says the BIOS will show actual MHz once you get passed 1600Mhz. Ha! Get it past 1800+ range, but below 1600MHz, and sometimes you get the Mhz, sometimes you get 1900+, most often, it gives up and just calls it an XP.
So long as you’ve made manual settings, it will do what you’ve said, it just won’t always show it.
Bizarre Default Settings Shuttle presumes you won’t use a floppy drive. Now maybe I’m old-fashioned, but I like to use the old setup disk routine, and it’s disconcerting to expect a floppy boot and not get it.
It also defaults hard drives to PIO settings. You have to turn UDMA detection on in the BIOS, if you don’t, it won’t.
The organization of settings is a little odd. If you’re trying to save every IRQ you can and want to disable your serial ports, you won’t find those settings upfront in Integrated Peripherals, but rather within something called “Super I/O.”
Quibbles, for sure, and easily fixed, but if you don’t know to look for these things, you might be frustrated for a pretty long time.
No indication of PCI divisors Admittedly, this problem is becoming commonplace. Epox 8KHA+ owners discuss the mysteries of just when the /5 divisor kicks in much like medieval theologicians discuss the mysteries of the Holy Trinity.
Granted, just typing in a number is convenient, but surely a few sentences in the manual would remove the mystery. From the datasheet on the clock generator, it sure looks the /5 divisor kicks in at least by 166Mhz, which brings us to the next problem.
Maximum FSB Speed 166Mhz The clock generator can do up to 233Mhz. The Epox 8KHA+ allows up to 199Mhz. There’s no good hardware reason why the Shuttle is limited to 166Mhz.
This is shown by a little shareware program called CPUFSB. Use that, and you can set the FSB to way beyond 166Mhz.
The problem with that approach is that the gentleman who wrote CPUFSB would like to get paid for his work, and much of the cost appeal of the Shuttle vis-a-vis the Epox goes away when you have to pay $13 for something the Epox gives to you.
Not that running current memory that high is such a bonus anyway, at least at the moment. I was able to run Prime95 at 175Mhz with Crucial PC2100 with slow settings (but keeping 4-way interleave), but the memory scores were lower than they were at 150Mhz with the quickest settings.
I’m sure I could have gone even higher up the FSB scale if I had turned off 4-way interleave, too, but why?
FSB “Holes” There seems to be at least one instance where you don’t get the FSB you typed in. For instance, when I type in 155, I get 156 instead.
So far, the board seems able to handle anything reasonable I’ve thrown at it. I’ve had no problems setting up OSs or running benchmarking programs so far. Performance seems to be roughly equivalent to other KT266A boards. I’ve had crashes, but they’ve only occurred when I’ve pushed at or beyond the limits I’ve found in other tests.
Delay Before Initial Boot You turn this machine on, and you wait. The boot beep can take in the vicinity of ten seconds. It’s so much longer than the usual that you think something’s wrong until you get used to it. So don’t panic, give it time.
If I had to build a system for somebody who wasn’t going to be overclocking (or at least not much) tomorrow, and if you put a gun to my head; I would be inclined to buy them this board.
Then again, I’ve only worked with it for a week.
What makes me very hesitant to endorse this board quite yet is the sloppiness of the implementation. Yes, outside of the voltage issues, these are all minor matters that could easily be all fixed with a single BIOS change. Maybe Shuttle can even take care of the voltage issue at that point, too, in which case it becomes the leading current contender for overclockers who don’t want to solder or clip.
But these are not esoteric little quirks. Most of these should have been caught before the product went out the door. The fact that such obvious, apparent items weren’t caught is a cause for concern.
Makes you wonder when and if they’ll fix it. Makes you wonder even more
what truly esoteric little faults might be lying underneath to be discovered in the course of time, and whether they’ll get fixed.
This is one of a handful of first-generation KT266A boards. The heavy hitters haven’t spoken yet, and you can’t help but think that we’re going to see better from some of the others, especially from the overclocker’s perspective.
The Epox 8KHA+ seems to be the hot board as of the moment, but it’s a little VCORE-challenged, too, and might well just be the best of the first.
I think a month from now, we’ll have a fuller list of contenders and a better idea of which will end up being best.