A few months ago, AMD said it was going to standardize on one socket, called M2, for the desktop. Opterons were going to get a different socket, called F. It had a lot more pins, and nobody knew exactly why.
Now we know why. AMD plans to integrate PCIe into Opterons.
There’s a number of technical questions raised by this, but we don’t know enough to try to speculate about them intelligently.
However, there’s one item that sticks out like a sore thumb: Why is this just for Opterons?
This is the socket 754/939 blunder all over again.
If integrating PCIe functions is so good for servers, it ought to be even better for desktops. After all, video is no big concern in the server world, but it is for the desktop, so you know this is going to migrate to the desktop sooner rather than later.
Right now, though, if this technology proves to be great shakes for video, what are gamers going to have to do? Buy Opterons? Are server mobo companies ready to make hot gamer boards?
Just watch and wait. We’re going to have these two sockets: M2 and F. Sometime next year, it’s going to dawn on AMD that they’ve screwed up again, so we’ll get another Opteron that isn’t an Opteron, and we’ll end up with a socket F minus 1 socket nine months later, with socket M2 being shoved into the bargain category.
So AMD will squander most of yet another opportunity. The article indicates that Intel has equivalent technology further down the pike, rest assured they won’t make the same mistake.
What AMD should have done with 754/939, and what they should do now with future sockets, is standardize on one socket. Don’t differentiate the sockets, differentiate the motherboard circuitry. Intel is perfectly capable of differentiating PIVs and Celerons without using different sockets. If cost is really a factor, you don’t have to enable the extra pins for bargain CPUs. Note that this new approach reduces the need for motherboard circuitry; it doesn’t increase it, so it should be relatively simple on higher-end boards to include any HT circuitry, and let enabled new chips either bypass in favor of the direct approach, or use the old HT as a supplement. If extra cost doesn’t stop people from building SLI or Crossfire mobos, it certainly shouldn’t here.
Snatching Defeat From The Jaws of Victory
Events like this are the main reason why I get so ticked off at this company. AMD comes out with great ideas, then they blow the implementation, either by not being able to deliver timely, or even worse, by making decisions oblivious to what will inevitably happen down the road that will force them to reverse course. It happened with 754/939; it happened with x86-64 and Semprons, it’s happening with dual-cores; it happens again and again and again.
That’s bad enough, but when they start repeating the same errors, well, that just raises frustration to a new level.
This is why I get a lot madder at AMD than Intel. Intel ran a CPU design into the ground, but what else can you say after saying, “Don’t buy them until they get out of the ditch?”
With AMD, it’s more like watching somebody dig themselves a (smaller) ditch, jump into it, get themselves out, and as soon as they get out, start digging again. It’s maddening, especially when it’s a race, and you know the competitor is really stuck.