The New Moore’s Law
Times They are a-Changing
A new report from IDC lays out what should be obvious to anyone who has followed two of the biggest recent trends in the server market—multicore and virtualization—by suggesting that when you have two popular new technologies that are designed to let customers consolidate multiple servers into a single box, then server vendors can expect to sell fewer such boxes. In other words, when customers begin rapidly replacing multiple servers with single servers, server sales start to dip (http://arstechnica.com/hardware/news/2007/03/multicore-virtualization-and-a-shrinking-server-market-maybe-or-maybe-not.ars)
As some of you may know software has been riding what we refer to as Moore’s Law for a couple of decades. What you may not know is that this “observation” dates back to 1965.
Here’s what I found at Wikipedia:“Moore's law describes a long-term trend in the history of computing hardware. Since the invention of the integrated circuit in 1958, the number of transistors that can be placed inexpensively on an integrated circuit has increased exponentially, doubling approximately every two years. The trend was first observed by Intel co-founder Gordon E. Moore in a 1965 paper. It has continued for almost half a century and in 2005 was not expected to stop for another decade at least.” (http://en.wikipedia.org/wiki/Moore%27s_Law)
It seems that we should all fear progress doesn’t it? Just when Intel/AMD has started to take advantage of silicon in new and imaginary ways, somebody needs to paint a very bleak picture of our future. As if we need more bad news, right?
While I don’t disagree with the fact as IDC sees them I’m not sure about their conclusion and I believe they report is over-simplify the issue. In fact over a century ago, Mark Twain put it very succinctly “The reports of my death are greatly exaggerated”
No one is suggesting the way Moore’s will continue to manifest itself is by continuing to crank up the clock speed, we may actually see the clock speed go down a bit. In fact, both Intel and AMD are going to increase the power of the microchip by adding more cores. How many? How fast? That depends on how much you trust the two chip manufacturers ability to continue shrinking the size of the transistors on their CPU’s.
The public roadmaps from both Intel and AMD clearly shows that they are going to continue to add more cores, in fact everything points to that at least Intel will continue to double the numbers of transistors on their CPU as well as other innovative steps (I’ll get to that later). This would mean that we are going to see the number of cores on Intel CPU’s double every two years (as per Moore’s Law). If we start applying that to PC’s coming in the first part of the next decade, that means we’ll see cell phone with dualcore CPU designs, Laptops with 4-8 cores, desktops (or Workstations) with 12-16 cores and servers with over 512 cores.
In the next couple of years we will likely see Intel and AMD continue to take advantage of their ability to continue to shrink the transistors, but rather that use this to add more CPU like cores to the design, you will see them add more specialized cores to their CPU designs: Dedicated cores for IO, networking, and graphics (check out Intel’s Larrabee).
This will likely provide computer manufacturers with some very scalable designs. Think LEGO (TM) brinks.
We are also going to see Intel and AMD pull a fast one and reverse it. Huh? Well, think about the Atom CPU presently powering the ultra-compact laptops commonly knows as Netbooks. What Intel did what to take a single-core CPU and rather than adding more CPU’s to the die, they shrunk it. Everything we’ve seen so far indicates that Intel will continue that trend. This obviously means that we’ll get more powerful Atom (dualcore and beyond with more powerful graphics capabilities), but also smaller CPU’s as well as PC-on-an-chip designs designed for smaller devices such as smartphones and devices between the Netbooks and Smartpones, commonly known as MID’s (Moblie Internet Devices).
But all of this great stuff is all for naught, at least according to Gartner and IDC, right? No so fast!
Windows Server 2008 R2 and Windows 7 will both dramatically increase their multi-tasking (actually their multi-threading) capabilities when they become available. The redesigned kernel architectures that went into Windows (with 7 and Server 2008 R2) is believe to scale linearly by introducing a new scheduling design as well as a better way to control processor affinity (for more about this, check out this video interview with Mark Russinovich on http://channel9.msdn.com/shows/Going+Deep/Mark-Russinovich-Inside-Windows-7/) . To date the Server team has publicly demonstrated Windows Server 2008 R2 up to 256 cores (or logical processors) in a single machine the only thing holding the test team back is the relative scarcity of these monster servers.
And as most of us tend to run multiple applications, browsers, email clients and background processing such as anti-virus, you will continue to take advantage of the multicore CPU’s by way of Windows multi-threading capabilities. In fact, the real challenge moving forward may not be Windows (or any other OS), but the tools available to write applications. For more information about this check out what our Developer Tools Division is doing with the parallel extension and their recently announced next generation tools and compilers.
All of this is to simply suggest that we will continue to tweak our products capabilities, including Windows, to ensure our customers can take advantage of the power of Moore’s Law. Now, the Virtualization and Virtual Machine (VM) density on servers? That’s a different story!
