Everybody’s excited about Intel’s new 45nm chip architecture and its first iteration, the Penryns. They’re going to be in the MacBooks, they’re going to be in your desktops, they’re basically taking the joint over and everyone’s pumped. You’re probably wondering why. Isn’t it just another bump in processor speed? No, it’s a bit better than that – read on and find out why 45nm is a major step for Intel.
First, it’s important to understand what 45nm means. This new process is essentially a further miniaturization of transistor technology, which all computer chips are based on. 65nm has been the standard for a couple years now, and Intel’s chips have been, in their multitudes, based on the same base-level recipe – or rather, on the same cake mold. The various versions – mobile, desktop, server, multi-core, all have the same underlying transistor technology, but customized for different tasks. In any case, 45nm is the new basic building block they’ll be working with. We’ve been shrinking the things for years, and each shrinkage makes for improvement in several areas, the two of which you care about are:
Transistor density- Smaller transistors naturally means more of them in any given area. This means higher clock speeds at the same chip size and on the same interface. It also means they can fit more custom circuits and sub-architectures for special instructions.
Energy efficiency- Transistors leak a bit of current. A leaked electron here and there is a pretty insignificant problem on its own, but when you’ve got 500 million transistors leaking a hundred times a second, it starts getting noticeable. A smaller and more efficient gate means that there is less chance for electron leakage, so less power is needed for each transistor. Translation: better battery life in laptops, lower power consumption for desktops.
There are other improvements as well: Intel is taking this opportunity to improve built-in special instructions for all kinds of basic computations. There are other sets of instructions: you may have seen “MMX” and “SSE” being mentioned before. Well, the new iteration is SSE4, and the upgrades are mainly on the raw computation side. For instance, they’ve doubled the speed of simple division, and added optimizations for vector modifications. Sounds minor, but so many CPU functions rely on derivatives of these kinds of instructions that there can be some major improvements seen. Think about the math involved in stretching a 1080p video to fit your screen, or the amount of calculation involved in ray-tracing shadows or cloth simulation. Any improvement in basic calculation snowballs into a major saving later on as long as programmers are aware of how to take advantage of the special instructions.
Another thing you’ll see is a larger L2 cache. This is used by the CPU as a sort of personal RAM supply for switching things in and out super fast without sending them all the way to your system RAM. Penryn dual core processors will have 6MB of L2 cache, 50% more than current ones, and it should be faster as well. There are even more improvements in there, but honestly some are too technical for me to even try to understand, though I trust others in their judgment that they’ll be helpful.
Basically, 45nm is a good thing, and it’s not a minor change. It’s one of the bigger steps that can be taken with processors, and it has both immediate benefits and the promise of future improvement. Now, if only they’d send me a sample.