Someone sent me this graph, which apparently comes from this website:
(Normally, I would link directly to the file, but this website is in Chinese, and I haven’t been able to find it on the site itself. I’ll be glad to do so if
someone else can find it. The website also has an advance review with benchmarks for Sempron socket A and 754 processors, so please take a look at that.)
There are three items to note here:
We explained the ramifications of the first a few days ago. It means you’ll be able to get a socket 939 processor at a fairly reasonable price then.
The second item confirms what AMD has indicated for some time now, it’s the end of the road for socket 754 as a cutting-edge platform.
The third item is new, and at first glance rather odd. Why would AMD list within the product line of a single socket the same speed grade twice?
It’s not a typo; the 4000+ that is supposed to come out in Q2 ’05 has a dash in front of the 4000+, as does the 4200+ promised in Q3 ’05.
The only explanation I can come up with is that the 4000+ that will be introduced at the end of the year will be a 130nm processor, while the 4000+ that will be introduced six months after that will be a 90nm processor.
In other words, AMD will take the pick of the litter from 130nm and sell them as 2.6GHz, whether FX or Athlon64. They’ll be priced skyhigh; demand will be slight, so low yields of chips capable of that speed won’t matter.
That of course means that AMD can’t at the moment make a decent enough supply of 90nm Hammers that will run at 2.6GHz, and doesn’t expect to until nine months plus from now.
This is not surprising since AMD’s partner IBM has been having problems getting their 90nm SOI chips past 2.5GHz.
We know that AMD can make at least some 2.2GHz 90nm chips, a few have already seen and tested them. So they can make 90nm, they just can’t make them quite that fast yet.
We’ll no doubt see 90nm chips from AMD in a few months. Most likely, there will be at least Opterons and Mobile Athlon64s. Around the same time (but perhaps more quietly), AMD will start slipping 90nm Athlon64s wherever they can into the desktop range: 2.2GHz would be a safe bet, maybe a little more, maybe a little less).
Those waiting for 90nm processors ought to take note that if AMD can’t make enough 90nm processors that hit 2.6GHz, odds are that overclocking results from the initial chips are likely to be underwhelming.
The proof of the pudding will of course be actual testing, but that’s what the advance indicators point to.