There’s a chipmakers’ conference going on, and one speaker told them that clock speeds would only double or triple over the next 10 to 14 years.
Now there’s a Christmas present for a speed demon! And it’s the gift that keeps on giving!
Sarcasm aside, this isn’t a drought, this is a climate change.
Just how much of a slowdown is this? The best way to compare is to see the difference between today’s processors and those made ten and fourteen years ago.
Today, the fastest desktop processor made today is the FX-57. It runs at 2.8GHz (and doesn’t have the elongated, IPC reducing pipelines of the PIV, which makes it a better candidate for comparison).
What was the fastest processor available ten and fourteen years ago?
In November 1995, the fastest processor you could buy was the Pentium 133.
In November 1991, the fastest processor available at the time was the Intel 486DX 50 MHz.
Let’s assume the best, and assume clock speeds will triple in the next ten years. That means it will go up 200%.
What actually happened between 1995 and today? Clock speeds are twenty-one times higher in 2005 than they were in 1995. That means clock speeds increased 2000% between 1995 and 2005.
Let’s be a little less rosy and assume clock speeds will take fourteen years to triple. Again, that’s a 200% increase.
What happened between 1991 and today? Clock speeds are fifty-six times higher today than then. That’s a 5500% increase.
And these are the better scenarios!
Obviously, clock speed is not the only parameter for performance. However, no matter how you slice or dice it, it has been the King Kong of performance up to now, and that role will lessen only slowly and gradually over the next few years, so any slowdown in progress this dramatic will have a huge impact on the computer business.
Yes, in 2020, we might have swarms of CPU mites in our computers, but that is going to a whole new world that will be a lot more different from today than today is different from 1995 or even 1991.
Yes, performance can be increased by means other than clock speed, but anything you can do will be a lot more difficult, painful, and slower in implementation than pumping up the volume.
Increase data width (aka x86-64), and you need new software. You haven’t seen a big rush on that despite both AMD and Intel having shifted, now have you?
Increase the number of processors, and you really need new, usually written-from-scratch software. So far, there’s been a lot of wailing and gnashing of teeth, and little action.
It’s not that these things won’t happen, they will, but they’ll take years and years to become the norm, and you’re just not going to get people excited with Word or AIM performance with eight processors and ultra-wide data bandwidth.
What about overclockers? Well, just look at the name. If there’s no rapidly-increasing clock to over, it’s hard to see people buying a new system every year or so for years on end. It won’t much affect first-time overclockers, but in the long run, most people will do it once or twice, then you find something else to do.
After all, besides increasing the clock (and doing whatever to make that possible), there’s not much else one can DIY. You can’t make your CPUs breed, nor can you can’t fatten up your data paths.
Will this end overclocking, or shrivel it? I don’t think so, but it will end overclocking as we’ve known it. It will dramatically change who does it, how long the average person does it, and what actions they’ll take and money they spend to do it.
This world will still exist; it will just be much different.