Information Age: News, analysis & insight for IT & business leaders

In April 2005, the chip industry celebrated the fortieth anniversary of Moore's Law. Ever since Gordon Moore, co-founder of microprocessor giant Intel, first relayed his theory in the April 1965 edition of Electronics magazine, the industry has striven to fulfil its bold predictions.

Moore said back then that the number of transistors that could be packed into a single integrated circuit would double every 18 months, while the cost remained roughly the same. This was subsequently refined; the more recognised version states that processing power will double every 24 months while prices remain constant.

This has had a profound impact on the chip industry. Being late to market or producing a 'slower' chip can be a fatal episode for a vendor. But, that single-minded focus on speed and performance can leave business customers feeling under-served. Increases in raw power might appeal to gamers, but CIOs have more pressing concerns - not least in dealing with the side effects of chip upgrades.

"There are always segments of the computing end-user community that are looking for increased performance," says IDC analyst Shane Rau. "However, for mainstream users most systems are meeting their immediate needs."

The chipmakers insist they are serving real business needs. Next-generation applications will be far more processor hungry, they say. Take, for example, the move towards service-oriented architectures (SOA): XML, the protocol used by web services, consumes many more cycles than HTML.

Alan Priestley, strategic marketing manager for Intel's enterprise server group, notes that "there is an overhead in performance as you package that depth of functionality. You're not going to evolve to an SOA with existing hardware." He says the increasing depth of business intelligence analysis tools could also demand considerably more processing power. Others suggest that the growth of multi-tasking on the desktop alone - having several applications open in many different windows - is enough to justify higher performing chips.

But there is a snag: the increased heat generation and power consumption of faster and denser processors have posed problems for data centre managers and chip manufacturers alike. Today's highest performing chips require up to 130 watts of power and with the number of transistors built into a processor nearing 2 billion, overheating could cause systems shutdown or errors due to current leakage.

Such physical limitations have caused a radical rethink of chip structures which may be to the advantage of the corporate IT manager. Energy-hungry chips mean higher electricity bills and require more complex and expensive cooling systems. But innovations in 'multi-core' chips may hold the solution.

By putting two smaller and lighter processors onto one chip, the same tasks can be run in parallel, sharing the processing require-ment and consuming less power. Additionally, transistors can be packed less tightly than in a single core, generating less heat. Multi-core processors manage to balance an increase in performance without the heat and power overheads, so fewer servers - and fewer air conditioning units - are needed for the same tasks.

Much of the innovation stems from the development of mobile devices. Dean Freeman, an analyst at IT industry advisor, Gartner, says that developing smaller laptop chips "woke Intel up in a big way" to new ideas about processor design.

"They found they couldn't just take a standard mainframe Pentium processor and make that into a mobile chip," says Freeman. "It needed too much power and the operating speeds were not fast enough. So they completely redesigned the chip. Previously there was a set design, a cookie cutter that they just changed slightly, to make manufacturing high-volume and scalable. Centrino [Intel's laptop chip] opened Intel's eyes to the fact that they were maybe going down the wrong curve with brute force of scaling and made them bring dual core to the forefront of their product roadmaps."

Intel's Priestley also notes that microswitching, a technology developed for laptops, that allows processors to dynamically slow down or speed up according to the level of activity, now saves power in servers too. AMD's 'cool and quiet' technology similarly adjusts processor power within milliseconds of user actions; the company claims 65% power savings at the CPU level, cutting 30% at the overall system level.

Dual-core has only recently made it down to the 'x86' level, the standard CPU architecture used in commodity hardware. The technology will appear in Intel's Itanium and AMD's Opteron chip families by mid-2005. IBM, Sun and HP have already released RISC-based dual-core processors for their own proprietary platforms, which are used only on high-end servers.

Chip slicing

Another chip innovation capturing the attention of the corporate IT buyer is that of virtualisation. This allows single processors to be partitioned into smaller virtual chips, allowing different operating systems to run on the same processor. While virtualisation in the data centre is not new, at the desktop level it has hitherto been rare. Previous software-based approaches have created substantial demands on processor power, impairing performance in both partitions.

Both AMD and Intel are looking at virtualisation at the processor level, dubbed Pacifica and Virtualisation Technology (VT) ,respectively. This approach "reduces the overhead of handling the CPU's resources," says Dave Everitt, product and platforms manager at AMD.

Everitt emphasises the potential for using desktop virtualisation to ease operating system upgrades: "It's a brilliant migration tool. If [Microsoft's next operating system] Longhorn comes out, but a company has some software that only runs on Windows NT, with virtualisation it can maintain that environment in its entirety."

Intel's VT marketing manager, Patrick Bohart, says the technology offers potential improvements in client manageability. "Once you break the rule that there is one set of platform hardware and one OS that runs on that, there is really a whole Wild West of things you can do."

Security management, for instance, could be made easier by providing every PC with a dedicated management partition for running packet inspectors or installing patches. This would enable remote management of terminals that may go offline if they suffer a virus attack or other error - or even if the user simply switches it off when IT needs to do maintenance. Security tools are another potential drain on performance which virtualisation spares the main user.

And just as processor virtualisation improves efficiency by bringing software capabilities into the hardware, so too are security functions being built into chips. AMD and Intel both offer technology which makes it harder for viruses and malware to run on their silicon.

However, virtualisation and the move to multi-core could have unexpected consequences. Some IT directors have voiced concerns about the cost of running software across servers using these technologies: Does a dual-core chip mean users owe software vendors twice the licence fees?

"Virtual environments and multi-threaded processors are going to make it much harder to define what a CPU is and how much a particular piece of software is using it," says Rorie Devine, infrastructure director at online betting exchange BetFair. "The application vendors will have to change their basic business model to adapt to this change, which they know is coming." There are similar issues around the move to 64-bit processors.

Currently the issue has created a schism in the software vendor community. Oracle and IBM have taken an initial position that for licensing purposes, dual-cores are equivalent to two processors - meaning two licences. In contrast, Microsoft says its customers will only need one licence for each dual-core processor. Microsoft's position pressures competitors into following suit, although it is too early to say whether they will.

Moore or less?

Meanwhile, the chipmakers are devising ever-more innovative ways of increasing processing power. At the very cutting edge are chips that end users could reprogram like software, printable semiconductors that could be used on plastics and other nano-technological innovations. Gartner's Freeman is confident that it will be technically possible to shrink circuits from the current 90 nanometre-level down to at least 22 nanometres over the next six to eight years. He is less confident that the cost structure of Moore's Law will be so stable: "We can continue to shrink silicon down but will the economics of that allow us to put it into mass production?"

So despite the advances being made, there are serious doubts over whether Moore's Law can continue to hold true. Even Moore himself believes that the time frame for the doubling of chip power may begin to extend.

Victor Zhirnov, a research scientist at the Semiconductor Research Corporation, which coordinates early-stage, pre-competitive research into chip technologies, says that power and heat concerns will ultimately limit scaling. "We believe that heat is the most fundamental problem," he says. "We are confident that we can make much smaller and much faster transistors but at this point we do not know how to get the heat out. Multi-core processors and design tricks to put processors to sleep will work for a while but ultimately, we still have to face the heat problem."