DrVomact's Journal: Intel LGA 1156 socket: disastrous design, or geek paranoia?

I drank the koolaid, and decided to build myself a new box based on the Intel 1156 socket CPU. (I chose the i5 because it was cheap, and I figure 4 physical cores will be enough for now, thank you.) I'll spare you the intricate painful details of my constructus interruptus experience; the point is that as a result of discovering an obvious defect in my "gently pre-used" board, I did some googling on the 1156 socket, and found out things that make me feel very queasy about this new socket design. I just wish I'd done the research before I ordered the replacement board.

First of all, this is NOT about the overclocking burnout issue . Or at least not primarily; I have no intention of overclocking my new machine. It's not even about the claim that Foxconn makes bad sockets (apparently, all the magic smoke at Ananandtech issued from boards that were equipped with a Foxconn socket). No my friends, my deepest fears center on the nightmarish conviction that the 1156 socket is one of those "innovations" that will be go down in the history of technology as a really, really major screw-up—in the same league as Windows ME, Chernobyl, and the Apple Newton. The significance of the overclocking failures with the 1156 is much like that of a canary keeling over in a coal mine—no, the gas hasn't gotten to lethal levels for humans yet, but you don't say to your fellow miners that they shouldn't worry because they're not canaries; you have to ask why the canary died.

When I examined the socket in my 1156 board and the CPU, I realized that Intel had done something that they probably thought was really smart: the 1156 CPU does not have the usual pins sticking out the bottom. Instead, the CPU has contact pads—flat surfaces that are intended to be contacted by little bits of copper wire sticking up out of the socket. This is not merely a gender reversal on the old socket-and-pins idea. The 1156 CPU doesn't have recesses for pins (just flat pads, remember), and those things sticking up from the socket can't be dignified with the word "pins"—they are just very fine pointy pieces of copper wire that are supposed (in theory) to make contact with the pads on the CPU.

Why does this give me the heebie jeebies? Well it seems to me that it's awfully optimistic to expect each and every one those copper cilia to be lined up exactly with a CPU pad, and to meet it with the amount of firmness needed for a proper contact. And indeed, some of the pictures in the Anandtech article seem to indicate that many pads in their sample CPU didn't make proper contact with their appointed cilia. Worse, it seems really dumb to make an electrical connection through a point contact. The old pins made contact along the length of the pin that was inserted into the copper-lined socket. It's possible that the poor contacts contributed to the overclocking failures; regardless of whether they did or didn't, I think poor or no electrical contact in parts of a circuit that are supposed to be connected is bad.

There's also the issue of verifiability. It's poor design to make parts that are intended to fit together in such a way that you can't positively verify that the fitting has been done successfully. With the old-style pins, you could tell if you hadn't placed the CPU properly into its socket—there was a definite tactile sensation when all the pins lined up and the CPU dropped into place. You could then assume with a reasonable degree of confidence that each of those pins were making physical contact with the conductor-lined hole in the proper socket; you could positively verify that this step had been performed correctly. Now you can only tell that the chip fits into the socket; who knows what makes contact? Who knows how good each of those contacts is? Sure, you could take the chip off and look for dimples in the pads that indicate they were touched by a cilium. But that's like checking to see whether your refrigerator light stays off when the door is closed by opening the door. How do you know that when you put the chip back in the socket, the contacts are the same this time around? How many times can you keep doing this before those little bits of wire lose their springiness and give up? Or get tangled with another, as had happened on my pre-used board?

I'm sure this new idea makes chips a bit cheaper to manufacture for Intel. In effect, Intel has gone out of the pin business, and they are probably congratulating themselves about it now. However, I think it's a horribly bad electromechanical design; if I had to do it over again, I'd go AMD.