Power Computing

Superfast energy conserving chip in powerful new computer

IBM has simultaneously launched what it claims as, the fastest microprocessor ever built and an ultra-powerful new computer server that leverages the chip’s many breakthroughs in energy conservation and virtualisation technology. The new server is said to be the first ever to hold all four major benchmark speed records for business and technical performance.

At 4.7 GHz, the dual-core POWER6 processor doubles the speed of the previous generation POWER5 while using nearly the same amount of electricity to run and cool it. This means customers can use the new processor to either increase their performance by 100 percent or cut their power consumption virtually in half.

POWER6 is nearly three times faster than the latest HP Itanium processor that runs HP’s server line.

The POWER6 chip is the mighty heart of IBM’s new 2- to 16-core server, which offers three times the performance per core of the HP Superdome machine based on the key TPC-C benchmark.

But the new server offers more than just raw performance – it is a powerful midrange consolidation machine, containing special hardware and software that allows it to create many “virtual” servers on a single box. For example, IBM calculates that 30 SunFire v890s can be consolidated into a single rack of the new IBM machine, saving more than $100,000 per year on energy costs. According to IDC, IBM has gained 10.4 points of UNIX revenue share in the past five years — versus HP’s loss of 5.3 points and Sun’s loss of 1.4 points (4). IBM will use the new machine to target customers with less-efficient HP, Sun and Dell servers.

Benchmark Grand Slam
Demonstrating its versatility, the new IBM System p 570, running the POWER6 processor, claims the No.1 spots in the four most widely used performance benchmarks for Unix servers – SPECint2006 (measuring integer-calculating throughput common in business applications), SPECfp2006 (measuring floating point-calculating throughput required for scientific applications), SPECjbb2005 (measuring Java performance in business operations per second) and TPC-C (measuring transaction processing capability). This is a first for a single system to have all four categories. The new System p 570 now holds 25 benchmark records across a broad portfolio of business and technical applications.

Design Performance
The performance is largely attributed to the system’s balanced design. IBM succeeded in scaling the new server’s processor performance and system design (cache sizes and bandwidth) in a balanced way. For example, the system bandwidth – 300 gigabytes per second — could download the entire iTunes catalogue in about 60 seconds – 30 times faster than HP’s Itanium. POWER6 has a total cache size of 8MB per chip – four times POWER5 – to keep pace with the bandwidth.

By contrast, many other servers concentrate mainly on processor performance, at the expense of the server’s ability to feed data to the chip at a rate that takes advantage of the processor’s speed.

“Like the success of IBM’s Deep Blue chessplaying supercomputer 10 years ago, the debut of POWER6 proves that relentless innovation brings ‘impossible’ goals within reach,” said Bill Zeitler, senior vice president, IBM Systems and Technology Group. “The POWER6 processor forges performance and energy conservation technologies into a single piece of silicon, driving value for our customers.”

The POWER6 processor is built using IBM’s 65 nanometre process technology. Coming at a time when some experts have predicted an end to Moore’s Law, which holds that processor speed doubles every 18 months, the IBM breakthrough is driven by a host of technical achievements scored during the five-year research and development effort to develop the POWER6 chip. These include:

A dramatic improvement in the way instructions are executed inside the chip. IBM scientists increased chip performance by keeping static the number of pipeline stages – the chunks of operations that must be completed in a single cycle of clock time — but making each stage faster, removing unnecessary work and doing more in parallel. As a result, execution time is cut in half or energy consumption is reduced.

Separating circuits that can’t support low voltage operation onto their own power supply “rails,” allowing IBM to reduce power for the rest of the chip. Voltage/frequency “slewing,” enables the chip to lower electricity consumption by up to 50 percent, with minimal performance impact.

A new method of chip programming that enables POWER6 to operate at low voltages, allowing the same chip to be used in low power blade environments as well as large, highperformance symmetric multiprocessing machines. The chip has configurable bandwidth, enabling customers to choose maximum performance or minimal cost.

Power Productivity
The POWER6 chip includes additional techniques to conserve power and reduce heat generated by POWER6 servers. Processor clocks can be turned off when there is no useful work to be done and turned back on when there are instructions to be executed.

Power saving is also realised when the memory is not fully utilised, as power to parts of the memory not being utilised is turned off and then turned back on when needed. In cases where an over-temperature condition is detected, the POWER6 chip can reduce the rate of instruction execution to remain within an acceptable, userdefined temperature envelope.

IBM plans to introduce the POWER6 chip throughout the System p and System i server lines.

IBM has also unveiled another industry-first with a new feature that provides customers with the ability to move live virtual machines from one physical UNIX server to another while maintaining continuous availability. Coined the POWER6 Live Partition Mobility function, this technology — currently in beta, with general availability planned for later this year — enables customers to move active virtualised partitions without temporarily suspending them. While competing offerings require a disruptive reboot of the UNIX system and software stack, IBM is helping clients to optimise resource utilisation on a broader scale by allowing administrators to think of large groups of servers as a fluid resource rather than focusing on each server as a single entity with a dedicated purpose.