Dogpiling DDR. . .

Add Your Comments

Without Vision, the People Perish

We’re going to be getting a 133A and DDR board this weekend. Reviewing the 133A board should be no big deal; we expect to see some minor incremental improvements. Certainly not enough to justify buying a new motherboard right now if you already have a KT133 motherboard, but certainly a reason to buy it over a KT133 if you have to build a new system now.

Reviewing the DDR board is going to be a different story; there’s some strange things going on out there.

20% Is Not Enough?????

I just read that someplace about DDR.

If 20% were the required performance improvement for a product; the average hardware site product review would be an annual, not daily event.

That Was Then, This Is Now, What Is Tomorrow?

It is a fearsome thing to see a shortsighted consensus form that takes no heed of past or future. Right now, the knee-jerk reaction seems to be “DDR Sucks.”

Hmmm, last time I heard that about memory, it was “Rambus Sucks.” Look at some Willamette memory figures lately, now that RDRAM has a suitable home?

Rambus the company may suck, Willy may suck, the cost of dual-channel RDRAM may suck, but the memory performance certainly doesn’t. Blows the doors off DDR numbers, as a matter of fact. But despite that, Willy still can’t keep up with an SDR Athlon on most things. This should tell you something about how much faster memory affects performance

Before we turn our brains off about DDR, these conclusions are based on pre-production first generation boards, with no DDR software optimizations (except maybe in one place, see below). Not exactly a mature product to base a final judgment on, is it?

At the same time, though, we now hear about the wonders of KT133A motherboards, which yields half the improvement as of the moment. Less improvement is better?

Of course, cost comes into play. SDRAM is cheap; DDR is expensive. NOW.

DDR production boards are just beginning to reach reviewers, and first-generation boards are rarely the final word on a product. Oh, and the heavyweight in this category, Via, hasn’t even shown up yet.

As I’ve stated recently, DDR isn’t going to be a miracle worker simply because faster memory just doesn’t yield big results with the cached and prefetched up the kazoo processors we have today.

A year ago, I would have estimated an average 12% improvement from PC2100 DDR over PC133. Right now, we’re getting a bit less than that. Once the kinks get worked out, I bet that’s about what we end up with. Now I would say somewhat more for games and certain applications, but still no miracles.

However, that’s certainly better than what we got going PC66 to PC100, or PC100 to PC133. We had to replace RAM back then, too, lot more expensive RAM at that time. Where was the concern then?

The World Isn’t Ending Next Week

Micron plans on selling PC2100 RAM at a small premium over current PC133 by the middle of February. That’s a month from now. How many of you think you or your money won’t be around then?

That will effectively cut the price of DDR by almost half. If your choice next month is buying a mobo and 128Mb of PC133 for $X, or a DDR mobo and PC2100 for $X+ $30 or $35 ($20 more for the mobo, $10-15 for the RAM), guess which is the better bang for the buck, even with just a 5% improvement?

Oh, you just bought a ton of PC133 at firesale prices, and even with the DDR price cut, you don’t feel like going through the trouble of getting rid of it and buying new stuff for 5% improvement. OK, that’s not crazy.

However, what would be crazy to buy a 133A board to replace a 133 right now. You’re only going to get 3-5% improvement for your $150, and you’ve just dug yourself a deeper hole should you find three or six months from now that you now want DDR.

It would make a lot more sense after Palomino comes out to review your options and then decide whether or not you’ll stick with SDRAM then. We suspect you’ll have to replace the mobo at that point to effectively run a Palomino (due to power requirements and/or FSB), but we don’t know for sure yet. However, at that point, even if you decide to go with a 133A board, you’ll get a better, cheaper board.

If you decide to go DDR, that will be one less motherboard you’ll have to get rid of.

If you have to get a machine now and can’t wait even a month, try to get the first dependable-looking 133A board to preserve your Palomino option for later on.

The Breed Will Improve

The initial DDR RAM we’re going to see is rated at CAS2.5. CAS2 (at least officially) is still a whiles away.

We’ll get to see production boards being reviewed, and we expect to see a little improvement over what we’ve seen so far.

A Difficult Transition

Mobo manufacturers have had problems making DDR boards. The more exact timing of DDR apparently have caused more than a few difficulties, and we believe the mobo manufacturers have reacted in part by going with very conservative settings.

A big part of our DDR reviews is to see how far we can tweak these motherboards.

This may take a while. We got a 128Mb engineering sample stick from Micron. We know we have to buy more to give these a thorough testing, but we don’t want to pay $200 a stick, either.

Our initial DDR board is going to be the much recently-maligned IWill KA-266 class board. It hasn’t done too well in reviews lately.

However, we also know that all the reviews we’ve seen so far have been on a preproduction board that’s had at least one subsequent hardware revision to it. We know that the IWill engineers were focused on getting the board stable at that point, performance would come later.

If you look at recent IWill products closely, you’ll notice that they tend to be a bit different than the others as the BIOS matures. Most companies’ production boards emphasize performance first, then work on stability. IWill boards tend to be stable first, then improve in performance as the BIOS matures.

For instance, the KV-200-R (a board we didn’t review) performed poorly when first reviewed, but as time went on, that same board closed the gap as the BIOS matured.

Maybe not the best strategy to make reviewers happy, but if we find a relatively slow but stable board, we’re not going to rush to final judgment. We’re going to give it a chance to mature, especially if the competition requires televised appeals for prayer to work reliably. That’s one thing we have to say about the IWill boards we have, they work without constant care and feeding.

We’ll look and treat other DDR boards the same way, too. We think they’re going to need it. When we find a good one, we’ll tell you. Until then, we won’t. Our purpose is not to write up quick reviews of umpteen motherboards. Our goal is to find at least one good one, and we think we’ll have to let a few grow up a bit first. We are going to be in no rush, and neither should you.

Will Palomino Like DDR Better?

Maybe, maybe not. Combine that possibility with the greater likelihood of software optimizations (at least in new games), and it’s probably worth waiting to find out, isn’t it?

On the Other Extreme, the Curious New ZDNet Tests

Up to now, most benchmarks testing (outside of those using apps that are bandwidth-starved) haven’t yielded any dramatic improvement. The “old” ZdNet tests (Content Creation 2000, Winstone show relatively little improvement.

However, there are now new ZdNet tests, Content Creation 2001 and Winstone 2001. These show practically miraculous improvements using DDR, along the lines of 30% or better. We are extremely skeptical about this.

We’ll get those tests in a bit. In the meantime, we’re going to come up with some alternative tests to see just how much improvement you can really expect out of DDR using some of the apps these two programs use.

It’s Not Either/Or; It’s Neither/Or

I know a lot of you hate hearing me say “Wait” and “Not Now.” Be a lot easier and popular to jump on the bandwagon and say “Buy, buy, buy” all the time, but then I wouldn’t be honest to myself or to you.

I do say “Yes.” Not often, but that’s because there are relatively few optimal times to buy equipment.

I personally believe the best time to buy equipment is at the point in the generational cycle when the most of the improvements from a generation of products (including the bugs worked out of the first generation of products) can be had at a fairly reasonable price (i.e, settled down to $250 or below, and subsequent price cuts won’t be that great), and when there still remains a reasonably considerable amount of time before the next generation of products that significantly exceeds it comes along.

Last June, I said the cB0 Coppermines were worth buying, precisely because that point had been reached. At that time, we pointed out that the AMD products were worth a look.

Of course what is “a fairly reasonable price” to you might be different than mine, but the general principle still applies.

This only happens once or twice a year. In the last year or so, it’s been complicated by product delays. We had that with the cB0 Coppermines; they were delayed about two months. We’re having that with Palomino now.

We know we tick off people who want to buy now, but that’s precisely the time when you need to hear a voice telling you what you need to hear, not what you may want to hear at the moment. We lay out the facts and the future the best we can, and we give you our best call based on that.

You may or may not agree with our conclusions, or have a different set of circumstances or needs that result in a different conclusion, and that’s fine. We feel we’ve done our job when you know what your options are, and can make an informed decision.

A Murky Time

I look towards the next six months, and I see an unclear, troublesome time. Right now, we’re gathering an emsemble together, but we don’t know when the leading actor is going to show up.

We’ve seen delays already; we expect to see more. We’re decisively moving away from the BX and Descendants generation of PCs and its accompanying equipment. The PC of two years from now will have little in common with the one we see today, and there will be major birthing pains.

Adding to the technical troubles will be the likelihood of at least an American recessionary period, which will likely chop the resources necessary to aid the midwifing.

Sometime before summer, the horse (Palomino) or the Oregon river (Tualatin) will show up, but even as they show up, the star performers for the second act, representing an even more radical technological change (the Hammer or an improved Willamette river or maybe even an ex-President McKinley), will start grumbling on the sidelines.

However, the really ominous grumbling will come from the audience saying, “We’re short of money. Our stuff runs fine with what we got. Why should we buy you?” and for 80-90% of that audience, they have no answer.

Why do you need a 3Ghz processor to run Word? Think voice recognition is going to cut it in a noisy office?

If only 10-15% of users need and demand the extra power, and 85% don’t and won’t be conned into paying more for it, don’t be surprised to see cheap stuff for the 85%, and skyrocketing prices for the 15%.

Email Ed

Leave a Reply

Your email address will not be published. Required fields are marked *