Your next PC

James Wang, Bennett Ring and Leigh Dyer reveal where your next upgrades will take you.

The power and speed of today’s components are just stepping stones to tomorrow’s tech. What will your next PC look like in the years to come? James Wang, Bennett Ring and Leigh Dyer reveal where your next upgrades will take you.

The central processing unit of tomorrow will be a totally different beast than that of today. It will no longer be a processing ‘unit’, but rather multiple units. Nor will it be ‘central’, but rather spread over multiple cores. It will, however, still do processing, but many truck loads at once.

CPU performance increases in the last few decades have come from two sources: higher frequency and improved architecture. Future CPUs will still use these two techniques but will no longer rely on them solely. Future CPUs will have multiple cores on the same silicon chip; each CPU (as we’ll continue to call them) will in fact be multiple CPUs.

Having multiple CPUs work as a computer system is not new. Symmetrical Multi-Processing has been around for many years and allows two or more CPUs to be installed in the same box. Clusters allow even more CPUs to work together by linking up multiple computers on a network. What is new with multi-core CPUs is that for the first time, mainstream CPUs are integrated on the same silicon chip.

How is this significant and what does it tell us about the future of the CPU? The most significant aspect of having on-die integration is that thanks to Moore’s Law, the number of cores on a CPU will increase dramatically.

 When we had multiple CPUs on the motherboard level, it never grew beyond four CPUs per motherboard. After all, there’s only so much space on the motherboard and only so many hardware slots can be provided. When the CPUs are integrated on the same chip, this problem disappears.

Pat Gelsinger, Intel's senior vice president. He likes CPUs.

Moore’s (revised) Law predicts that the number of transistors on a chip will double every two years. This means that if we see dual core CPUs in 2005, we can expect to see quad-core CPUs of the same type arrive in 2007. Following on this pattern, by 2017, 128 cores of the original chip can be fitted on the same area of silicon. Inevitably there will be the overhead of communication logic, but they won’t be so complex as to halt the growth.

The number of cores on future CPUs depends mainly on the complexity of each core. If they remain identical from one generation to the next (unlikely), the doubling will occur roughly every two years. If each core becomes more complex, progress will be slower. On the other hand if they are simplified, we can more than double the number of cores every two years.

In the coming years, you will observe two kinds of multi-core chips. The first type will feature two or more cores of fairly complex nature. They will most likely be two regular CPUs crammed on the same die. Intel and AMD will be in this category. It makes sense for them to do this as single threaded applications will run just as fast if not faster while allowing for future programs to exploit multi-threading.

The second group will have many cores but each one will be far simpler and a lot more specialised. These CPUs will serve a market where the applications are already heavily threaded (servers) or where applications are yet to be written (consoles). Sun’s Niagara chip and the Cell processor falls into this category.

Will Moore’s Law continue? Without a doubt, for at least another 15 years. At this year’s Intel Developer Forums, Intel reaffirmed that Moore’s Law is not slowing down and expects to move to new process nodes every two years; 45nm in 2007, 32nm is 2009 and 22nm in 2011. For future CPUs, this means slower increases in clock speed but leaps and bounds in integrating multi-cores. Next year, Intel is expecting 85% of their shipping CPUs to be dual core.

What: Quad-core Pentium Processor
When: 2008

With four Pentium processors (either Pentium 4 or Centrino based) on a single chip, there’s little you can’t do. Thanks to Intel’s Vanderpool technology, you can run multiple operating systems at once; Windows and Linux can finally live together without compromise. Assuming Hyper-Threading is enabled on these cores, this mega CPU will be able to run eight threads. Multi-threaded applications will simply fly.

What: Dual-core Athlon 64 processor
When: 2005

The dual core Opteron server chip will come in the first half of the year followed by the desktop Athlon 64. With an on-die memoty controller to feed the CPUs, AMD's dual-core chip is much more mature and well thought out than Intel's.

In true AMD fashion, the chip will be compatible with most currently available motherboards too.

Beyond silicon What: Molecular computing When: 2025 and Beyond In his celebrated book The Age of Spiritual Machines, Ray Kurzweil showed that the exponential growth of computing didn't start with silicon chips but rather much earlier. Computers evolved through five stages: electromagnetic, relay, vacuum tubes, discrete transistor and integrated circuits. When he plotted the computing power available for each era, they roughly joined up to form an exponential curve. From this he argued that computing power in general has been increasing exponentially. He also made the intrigue observation that just as one technology reached its limit, another totally different computing technology would come in and carry on the growth. He therefore postulated that when silicon reaches its limit, a new technology will pick up where it left off. That technology could be biological, nanotech, quantum or something else. Kurzweil argues that DNA and quantum computers, even if possible, would make poor general purpose computers. Instead, he believes three-dimensional molecular computing will be the next big thing.