The Inquirer-Home

Intel's LGA 775 socket: Why no pins master?

Part II
Wed Aug 04 2004, 16:06
[Read Part One, here]

THIS BRINGS us to the million dollar question, why the change from pins to no pins. The answer is both simple and complicated. The simple bit is that the contacts you see in the LGA775 sockets have always been there in the older sockets, just in a slightly different form. The reason you never noticed them before is there was this perforated black plastic cover over them. That cover is now gone, and you can see the little metal bits that were previously hidden.

The complex part is the answer to the question of why you needed to remove the plastic cover. It all comes down to pin density. Chips are becoming smaller and smaller, not only in die area, but also in the overall physical package. Measurements taken with a ruler show that an Athlon is about 49*49mm, a s478 Pentium 4 is 35*35mm, Opterons 39*39mm, and the new LGA775 is 37.5*37.5mm. This puts the area of the chips at 2401, 1225, 1521 and 1406 square mm respectively. All but the Opteron have an area in the middle of the package not covered by pins. If you subtract that area, you end up with about 1725, 901, 1521 and 1134 square mm covered with pins.

The Athlon has 453 pins, the P4/478 has 478, Opteron 940, and P4/LGA775 has, er, 775. If you divide the area covered by pins by the number of pins, you end up with the rough area used per pin, really rough if you consider that this was hand measured. That gives you a per pin area of 3.81 square mm for the Athlon, 1.88 for the P4/478, 1.61 for the Opteron and 1.46 for the P4/LGA755. You can see that the newer chips drop in the area allowed per pin, and the trend is toward ever decreasing allowable real estate.

alt='foreway'

If you look at modern CPUs, you can see a dramatic trend. From the top left, we have an Athlon classic, to its right is an Opteron. The bottom row has a s478 P4 on the left and a LGA775 P4 on the right. Notice how much closer the CPUs on the right have their pins? It is a dramatic and visible difference.

alt='opteronand775pinscloseup' Another trend that is less noticeable is the spacing of the pads that the pins attach to, or in the case of LGA775, the pads themselves. On the Opteron, you can see a bit of the ceramic packaging material between the pins. On an LGA775 chip, the pads almost touch. The Opteron is on top, LGA775 P4 on the bottom in this picture.

You can get around this somewhat by making the pins thinner, and a quick eyeball of the chips shows that pins are indeed getting thinner. While this is a solution, it is a short term one. You can only make the pins so thin before they bend under the slightest touch. This leads to a whole other, and very costly set of problems.

The next problem is with the plastic of the socket, each pin requires a hole in it. The more holes you have, the weaker it gets, and the closer you put them, the weaker it gets. Soon enough, you end up with a perforated bit of plastic that does not protect anything, and is in great danger of breaking with any mishandling. Like thinner pins, this has associated costly problems.

And even more problematic is something closely related to pin spacing. Each socket hole has to have a contact in it, and the contact has to be separated from the pin next to the one it is meant to make contact with. Add the moveable parts of the ZIF socket into the mix and you have very little room to make 775 secure connections. You simply run out of physical room to put things.

With 1.46 square mm or so per pin, you end up with about 1.21mm between the centers of the pins. Subtract the pin width, and you have perilously little space to put the contact, separation material, and the clamping mechanism. If you think eyeglass screws are annoying to put in, imagine the engineering involved here.

All these things, and probably a few more, conspired to make the old concept of sockets untenable. Intel possibly could have done LGA775 the old way, maybe it could not have. Even if it could, each successive generation gets harder until it simply is not do-able. The pins simply had to go.

For the first time in years, the physical layout of the socket has been rethought with an eye on making it better for things other than just electricity. The results so far look pretty good, the old way was clearly not sustainable for much longer. µ

[Part III tomorrow, Ed.]

 

Share this:

blog comments powered by Disqus
Advertisement
Subscribe to INQ newsletters

Sign up for INQbot – a weekly roundup of the best from the INQ

Advertisement
INQ Poll

Heartbleed bug discovered in OpenSSL

Have you reacted to Heartbleed?