90nm Semprons

90nm Semprons are beginning to show up in the U.S., Newegg, for instance, has them. XBitLabs took a 1.6Ghz 128K model and pushed it to a little over 2.5GHz.

A few observations need to be made at this point:

It’s not how far you go; it’s where you end up A 50%+ overclock may warm the cockles of many souls, but it shouldn’t. After all, it doesn’t matter how much faster you’re going, but how fast “faster” ends up being.

If you had a choice between an 1.6GHz $80 processor which overclocked 60% to 2.6GHz, and a 2.5GHz $100 processor which only overclocked 20% to 3GHz, which would you take? You ought to take the latter, since you get a nice chunk of extra performance for $20.

Of course, you’ll say that the 2.5GHz processor would cost more like $500 than $100, and that 2.6er was a better bang for the buck, and you’d be right, but you wouldn’t be right based on overclocking percentage, you’d be right based on price/performance, which has nothing to do with OC percentage.

2.5+ GHz really isn’t all that fast for a 90nm chip; it’s no better than what XBitLabs was previously able to do with a 130nm Sempron.

There are also different kinds of socket 754 Semprons. One type has 128K cache, one has 256K cache. While it’s not yet clear how much difference the difference in cache makes in real-life apps and games, given that the dollar difference between the two is only $10 at the 1.6GHz level, even a 2-3% difference would easily justify the extra $10 on a bang-for-the-buck level.

That 2.5GHz speed may not be all the CPU’s fault, though, because . . . .

Multipliers and Mobos Matter

The 1.6Ghz Sempron has an 8X multiplier. That means the motherboard has to run at 300Mhz for the CPU to reach 300MHz, 350MHz to reach 2.8Ghz, and 375MHz to reach 3.0GHz.

Yes, nForce4 boards have been reported to get up to 350MHz without too much problem. The problem is, there aren’t any nForce 4 boards for socket 754 yet, only nForce 3 250GB boards, and, as XBitLabs pointed out in its test, getting 300MHz or more out of an nf3 250GB is something of a challenge.

We think any prospective socket 754 buyer ought to wait until stable nForce 4 boards (or Via equivalent) become available. We also think for at least the moment that a 9X multiplier should probably be the minimum multiplier one should get (unless the future mobos get better than 350MHz and/or the E stepping Hammers aren’t quite as good as expected).

The Tail Wagging The Dog

Those looking for a cheap gaming Hammer system are likely to find to their horror that the video card will cost as much or more than the CPU and mobo combined, maybe a whole lot more. If you think you’re going to do this piecemeal, that’s not going to work too well this time around. Upgrading to Hammer while hanging on to anything equal to or less than a Radeon 9800 Pro is going to help you little in Doom 3. On the other hand, if you lay out big for the video card, it will be stunted 20-30% by the older infrastructure.

Those looking at socket 939 may not end up in that position, but that’s just because they’ll pay a good deal more for the CPU and mobo (especially an SLI one), but what was true for a piecemeal upgrade for socket 754 will be even more so with socket 939.

Adding to the problem of a piecemeal upgrade is the shift from AGP to PCI Express. You really don’t want to lay out big money for an AGP card now, that would be extremely short-sighted if you plan to hang on the card for a couple years.

Now And Then

Whenever I write an article like this one, I also get a few emails from people who just bought something and take some offense at anyone suggesting they might not have been geniuses for doing so.

Well, it’s been a long time since we gave a thumbs up, and it’s been tough times. Hammer has taken forever to deploy. It started offering not much more than socket A at an outrageous prices. Then it went to offering not much more than socket A at not so outrageous prices. Right now, the end is nigh for socket A days, but Hammer still doesn’t offer tremendously more bang for the price (and most of the extra bang is really coming from the video card end).

Under those circumstances, and given the higher price tag a thorough Hammer upgrade is going to carry, the most prudent thing to do, at least from this angle, has been to wait to get every bit of improvement in this process generation, and that turns out to be the E stepping. Not that it by itself will be any great improvement, but a couple hundred more MHz will get Hammer around 50% better than overclocked socket A systems.

So what will waiting a few more months get you?

Rather than getting a 2.5-2.6Ghz processor, you’ll get hopefully a 2.8-3Ghz one. You’ll get a motherboard that will probably be capable of 10% more overclocking. For those interested in Sempron, you’ll likely get x86-64 as a bonus.

We are getting close. Obviously, if you’re running a PIII or Athlon Thunderbird and a Ti4200, waiting is probably not worth it, you’re hurting too much now.

But given the price tags, going through multiple generations of Hammers won’t be as fiscally painless as it was for socket A, so making one purchase at the right time makes more sense for most.


Be the first to comment

Leave a Reply