We’ve said several times in the past that things are slow, and that hasn’t changed. We still think sitting on your wallet this summer is a good idea, especially for those who have to make their purchases count.
This article is for the more impatient.
Not Much Thrill Here
On the AMD side, it’s hard to muster much enthusiasm for Thoroughbred when not even AMD can. While the early Thoroughbreds may be good for folks who want to replace their fire-breathing Thunderbirds, the smart money will wait a couple months or even until we see Barton in what looks to be the fall. At the least, the increase in cache to 512K should provide roughly a 5% boost in performance, maybe a little more.
There are also mumblings, some of them even coming from AMD, that Thoroughbreds will be improved down the road. The AMD statements may mean no more than the additional cache (and some tweaking).
If you try to read deeply into the AMD statements, they look like changes that are going to happen, not have already happened.
There have also been rumors of more radical changes, but this looks to be a combination of wishful thinking and confusion with Hammer features.
It doesn’t seem likely AMD wants to spiff up Thoroughbred too much, otherwise, why buy Hammer?
In any case, we’ll probably not see whatever improvements there are right away. My best guess is early fall.
Per mobos, we won’t see KT400 based mobos until the summer. If you’re looking for a mobo that has to last you longer than toilet paper, you might as well get AGP 8X out of it.
On the other hand, we have Intel.
The Northwood processor itself is still at the B0 stepping, and Intel has not to date notified developers of a new stepping for it. Unless Intel does so within the next week or so, it’s unlikely we’ll see one until late summer/early fall.
However, early stepping or not, people have been doing some remarkable things with the little Northwood runt of the pack: the 1.6A.
It’s pretty hard to ignore 800Mhz, 50% overclocks with at most minor voltage increases. For the bold and the brave, 1GHz+ overclocks look quite possible without getting too insane about it.
Not bad for a $130 processor.
Is 1.6A Northwood overclocking for you?
Well, there’s two ways to approach it. If all you’re willing to do is increase the voltage a little, 2.4GHz is the speed you can realistically expect to get with some reasonably decent DDR2100. That’s not bad, but it’s not something to abandon a socket A system over.
No, to make it worth somewhat worth your while, you probably need to get a little extreme about it.
If 2.7GHz sounds more appealing to you, the following are the sort of things you’ll need to do and consider.
Be ready to go to 1.85V You might be lucky and be able to get away with less, but don’t count on it. If you can’t afford to lose the processor, don’t do it. If you’ll lose sleep over this, don’t do it. On the other hand, if you’ve been acting like the default voltage on the Athlons is 2V, and you go up from there, overvolting the PIV by 20%+ should be old hat.
Most PIV boards do not regularly provide 1.85V. In most cases, you’ll have to make a small modification to the PIV to change its default voltage. Here’s how to do this. You can do it with conductive ink as shown, or you could do the same thing with thin wire.
Watercooling Would Be A Really Good Idea While a few have managed the feat with air, most who have already done this have been using water.
You Need Serious RAM Remember that the PIV is bandwidth-hungry. The initial Willies were meant to absorb 3.2Gb a second from RDRAM, and even the fastest DDR is trying to feed a faster processor with somewhat less data.
However, MHz don’t matter as much as CAS when it comes to feeding this beast. You not only need memory that runs at 166Mhz or more, you need memory that will do so at the quickest possible settings. This page shows rather graphically (if a bit confusingly) that you get little to no boost if you run at 166MHz, but have to scale back from CL2 to even 2.5.
Memory is only just now becoming available that is supposed to be able to do so right out of the box. While pioneers have coaxed current modules into the feat, they’ve been generally doing so with more than a bit of extra voltage. For instance, most of the testing done here was done by overvolting the memory by about 20%.
This is one good reason for waiting a few more weeks before trying this. To hell with the raw MHz numbers. You want something that can run at least 167MHz out of the box at CL2 with all the other numbers down as low as they go. If you’d like to see some examples (and codes) of these soon-to-come modules, look here.
The other, even better reason for waiting a bit is . . . .
Better Mobos Are Coming
Again, while this chart is incomplete and confusing, it does show that future chipsets will give a substantial real boost to performance over current ones, somewhere in the neighborhood of 15% in the case of Quake. That’s an excellent reason to wait a few weeks.
(Incidentally, the RDRAM PC1066 and 1200 scores on that chart also provides the first confirmation of what I said the other day about what might be expected from dual-channel DDR. RDRAM PC1066 would provide the same bandwidth, and it improved Quake scores by 15% over the (improved) single-channel DDR score.)
Which is this “future chipset?” It could be the Intel 845E or the Via P4X600. It might be the SiS 648. Whichever one it is, it looks like something to wait for.
Update: It looks like the article was originally more forthcoming on these mystery chipsets. Take a look at this. No matter how poor your Finnish is, you should be able to read the part about the 845G in the middle of the page.
Hmmmm. Sort of like nForce all over again. And yes, there will be 845G boards out there with an AGP slot.
In a month or so, the summer overclocker competition will be set. It will be a system like this vs. the initial AMD Thoroughbred systems. While performance will probably be close (though Intel is likely to extend its lead in bandwidth-sensitive programs), I suspect that overclockers who take the Northwood approach will get more of a thrill out of a 1GHz overclock than Thoroughbreeders will from 400MHz or so.
A cheap thrill? Yes, but what else is overclocking? 🙂