Multi-core serves up performance

By Ben Williams, Network World
June 13, 2005 12:11 AM ET

This vendor-written tech primer has been edited by Network World to eliminate product promotion, but readers should note it will likely favor the submitter's approach.

As application performance demands grow, data centers require more computing power in a smaller box so that more servers can fit in a rack. Multi-core processing helps address these needs.

Multi-core refers to a CPU that includes two or more complete execution cores per physical processor. The processors and their caches and cache controllers are combined onto a single integrated circuit. This technology evolution allows for increased performance and higher productivity in computers that can simultaneously run multiple complex applications and complete more tasks in a shorter amount of time. These performance gains are accomplished without increasing power or heat.

Processing power

In 1965, Gordon Moore predicted that the number of transistors on a chip would double every 18 months. The next 40 years witnessed this assertion become law (Moore's Law), as transistor size decreased and transistor density and computing power increased.

However, the laws of physics restrict Moore's vision. First, transistors slimmer than 16 nanometers cause unpredictable electron transmission, creating a ceiling for transistor size and density on a single core. Second, performance improvements related to innovations in cache, clock speed, memory access and I/O cause undesirable power consumption and heat generation.

Related Content

To achieve constantly increasing microprocessor operating frequencies, individual transistors need to switch faster and faster. Continuous enhancements in transistor technology enable this, but the trade-off is the amount of power consumed by each transistor.

To operate at very high speeds, transistors draw power constantly. This is referred to as "leakage" or "static" power. With the highest-frequency microprocessors, static power can become a large percentage of the total power consumption.

Creating further challenge, the increase in static power for every extra megahertz of operation is a non-linear function. Where a microprocessor may require 100 watts to run at a certain frequency, to run 10% faster might require 30% or 40% higher power. This creates a problem at the system level where the heat generated by these microprocessors needs to be removed from the box.

Click to see: How it works: Multicore processing

There are limitations on how much heat can be removed using traditional fan and heat-sink air-cooled methods. Moving to more exotic system cooling solutions is a significantly more expensive proposition.

Multi-core microprocessors provide an excellent way to address these issues. The concept is to replicate the microprocessor core and run it at a slightly lower frequency. The net result is almost double the compute power in a given chip at the same power consumption.

The marriage of multi-core processing and x86 64-bit computing represents the next wave in enterprise computing. Performance is maximized by operating systems that can more effectively prioritize and manage multiple application tasks simultaneously - either from many applications or a single application.

Architectural advantages and increased physical resources allow multi-core processors to assign discrete threads (specific applications or operating systems) to distinct cores within the processor, thereby offering greater multi-tasking capabilities to multi-threaded environments. Until recently, higher-end systems and mainframe servers exclusively handled the most demanding applications and workloads, but now the combination of multi-core processing and x86 64-bit computing makes mainframe capabilities available to the masses by drastically lowering the price of performance.