AMDZone had the following to say about this socket 939/940 question:
The 940-pin Athlon 64 will be the variety at launch. The extra one pin is required to work on the 6-layer Opteron motherboards. The 939-pin Athlon 64’s aren’t ready for shipping yet so they will be out a little later on. As for the Opteron situation, that little extra one pin is required for SMP support due to the necessary HyperTransport links.
Let’s look at this:
As for the Opteron situation, that little extra one pin is required for SMP support due to the necessary HyperTransport links.
This seems to say that turning SMP (and the required HyperTransport links) on and off depends on that one pin. That implies AMD is going to this 939-pin layout just to prevent people from taking Athlon64s and treating them like 2XX series Opterons. It also implies that doing just that is doable with the initial Athlon64.
The 940-pin Athlon 64 will be the variety at launch. The extra one pin is required to work on the 6-layer Opteron motherboards.
This would seem to mean that 939-pin Athlon64s will not work on current Opteron motherboards. That would be very bad news indeed for any early adapter who bought an Opteron board and Opteron/initial Athlon64 processor and figured to upgrade it with a cheaper socket 939 Athlon64 processor later.
This would also seem a very odd requirement for any single-processor Opteron board (and God only know what the number of board layers would have to do with it, but see below).
AMD obviously doesn’t want to see any cheap SMP systems around. They want you to buy two-way Opterons instead, which at current pricing at the high end means about $500 extra dollars to enable two pins.
It seems like they’re so hellbent on doing this that they’ll disable socket 939s from working at all. This doesn’t make any sense. Disable SMP if you don’t get a signal from the CPU, but not the whole thing.
Perhaps the “six-layer” comment was some AMD rep’s garbled version of what he had heard, which actually was about dual-processor capable boards needing that pin. But if that were the case, what need would there be for socket 939 when single-processor socket 940 boards would work as well?
The 939-pin Athlon 64’s aren’t ready for shipping yet so they will be out a little later on.
This doesn’t make any sense, either. If they were adding some capability to the CPU, yes, a delay would make sense.
But how much work does it take to omit one pin which disables a feature?
And if they’re so concerned about unofficial SMPing, why are they releasing Athlon64s WITH that naughty pin to begin with?
A better explanation is that socket 939 motherboards won’t be ready in time.
But I can think of a better one still.
A Stall By Another Name
First, this 939-pin stuff came out of the blue recently, just a little more than two weeks ago. That’s pretty fishy in-and-of-itself.
This Inquirer article reveals some rather odd projected shipments for Athlon64s.
In AMD’s heyday, they shipped a little over 4,000,000 Athlons and a little under 4,000,000 Durons a quarter. Now, it’s probably like three million apiece.
According to the article, total shipments of socket 754 chips are supposed to reach 1.25 million in the first quarter of next year. It’s understandable that this figure wouldn’t come close to three or four million, after all the AthlonXPs will still be around, and this is supposed to debut after its big brother (perhaps in November).
So that number looks reasonable.
But look at the projected 940-pin (32,000) and 939-pin (250,000) shipments. Together, that’s only 282,000 CPUs. For the second quarter of a rampup, you’d expect a figure at least several times higher than that, a figure that would at least approach the 1.25 million for socket 754 processors.
The article then goes on to project sales of “big” Athlon64s jumping by a factor of ten the next quarter, to around 2.7 million. The second quarter is also the time when 90nm Athlon64s are supposed to show up.
AMD has projected huge rampups before, just to have them vanish. Even if you’re not skeptical about that, the production numbers for the 1Q seem to indicate continued big problems in making the things. Given the not-too-bad shipment numbers for the simpler socket 754 chip, this would seem to indicate some sort of yield problem with the more complex chip. They can make some, but not a lot.
It seems to me that somebody’s getting excuses ready for a little-more-than-paper launch and early trajectory of the more complex Athlon64.