Ace’s Hardware took an Athlon FX, froze it, and got 2.8GHz out of it. It then ran some benchmarks.
While you probably shouldn’t take all too seriously some of the claims as to PIV equivalence made in the article; it is pretty reasonable to say that to beat a 2.8GHz FX, a PIV will have to get seriously above 4GHz to compete.
It is also pretty reasonable to say that the frozen FX of today will be the more modestly cooled FX of tomorrow. I don’t think 2.8GHz is at all unreasonable to expect from a 90nm Hammer; I suspect 3-3.2GHz will eventually be more like it, and if that’s the case, now we’re looking a neighborhood around 5GHz for the PIV.
I also suspect that the Opteron boards being used for the FX aren’t exactly great gaming boards either. The impression I’ve gotten looking at all the gaming benchmarks is that the difference between the FX and the 64 is rather closer than it ought to be. I suspect a socket 939 motherboard designed for desktop use will give the second generation FX an extra boost over and above what any new technologies might provide, maybe 5-7% more.
All this points to a second-generation FX platform that is substantially better than today’s AthlonXP platforms, and rather formidable by today’s standards.
But it won’t be judged by today’s standards, but tomorrow’s.
What will tomorrow bring?
As we’ve mentioned before, Prescott is going to have a limited lifespan and very limited ramping for a new generation of Intel product. The fastest Prescott planned today is 4GHz.
What we don’t know for sure yet is whether that 4GHz represents a rough guide to how far this chip can go, or just how far Intel will go with it.
So far, we’ve assumed the first, and at this point most of the indicators are pointing that way.
Should that be the case, it is very hard to see how Prescott could beat or even stay even in any overclocking race with a second-generation Hammer.
And that’s before considering the possible impact of x86-64.
However . . . .
If you look at Tejas, which will be made towards the end of 2004 and uses 90nm process technology, too, you see much more of a ramp than with Prescott. The latest numbers being tossed around are 5-7GHz, which ought to make Tejas a more than worthy competitor to any 90nm Hammer.
Indeed, Tejas may be a rather different CPU indeed, as this recent Inquirer article points out.
We’ll talk tomorrow about the concept of a modular CPU, which certainly would be revolutionary from at least a design perspective, but in an age where companies are having tremendous problems getting regular CPUs and GPUs to work right using smaller process technologies, one has to wonder about the growing pains of something like that.
But even the core of such a CPU would be an effective answer to a 90nm Hammer.
What we are likely to see over the next eighteen months (from an overclocking perspective) is Intel and AMD leapfrogging over each other, but not by much.
Let’s give AMD a leapfrog with FX today.
When Prescott comes (and obviously, if it doesn’t, there’s no leapfrogging), that ought to let Prescott at least get over the hump of FX. What will be the first important test of this new race will be by how much Prescott can leap over the current FX. This test will come in two parts: first, in December with the initial Prescott, then in February with the first cheap Prescotts.
If the first and especially second set of CPU have problems getting much above 4GHz, even assuming usual improvements with subsequent stepping, Prescott is going to have a real problem against Hammer the rest of the year.
If the numbers are more like 4.5GHz, then Hammer is the one that’s facing some problems.
I find the first more likely to happen than the second.
The next big potential event is 90nm Hammer (at default speeds, you could have FXs and Prescott swapping places a couple times, but let’s deal with major changes). Should that get launched next June without problems or delays; it probably will be able to leapfrog over Prescott (and by rather more than Prescott leapfrogged over the initial FX).
Once AMD irons out all the details as to what works with what, that’s the platform AMD fans should be looking forward to.
Then, perhaps 4-6 months later than that, we have this maybe modular Tejas, which ought to leapfrog over FX2.
Beginning around this time, multiple core CPUs will begin to be seriously talked about (which might be what this modular Tejas is really about).
The real problem with this timetable is not the what but the when. Will AMD get 90nm processors out in time? I wouldn’t bet on it. Can Intel come up with this revolutionary chip by the end of next year (much less next summer, like the Inquirer article indicates)? I doubt that at least as much (though I can see why Intel would very much want to do that).
If AMD gets delayed with 90nm, and Intel can accelerate Tejas, that would be horrible, maybe fatal news for AMD (especially if AMD has problems moving first-generation Hammers).
If neither get delayed, AMD gets some time to shine and recover at Intel’s expense, which should buy them a few more years of life.
If AMD gets out on time, and Intel doesn’t, that widens the window of opportunity.
At least the first half of 2004 (and perhaps the whole year) will likely be troublesome times for both companies. Both companies will basically be selling transitional products soon to be obsoleted. Intel has to fear AMD being dominant the second half of the year, and giving x86-64 a chance to get entrenched.
AMD has to fear getting 90nm out on time and somehow persuade people to buy a lot of expensive first-generation Hammers while they’re doing that.
There are a lot of unpredictable possibilities throughout all this: motherboard reliability, availability of DDR2, availability of x86-64 software, and many others.
We continue to advise giving your wallet a siesta through all this.