Multi-core Processors – Have They Arrived?
The Challenge
The last few years have seen PC processors as well as those in servers and other systems become increasingly power-hungry. The response of designers has been to deploy multiple cores on a single chip to achieve the same or greater levels of performance as previously but at lower power levels. These multi-core processors (MCPs) offer the potential of very large performance increases without the expected increases in power consumption and radiated heat. The drawback is the need for new management and applications development technologies to deal with the issues that they raise.
MCPs herald a sea-change in processor usage. The migration to MCPs by mass-market vendors represents one of the biggest architectural shifts in hardware for a generation. Industry majors such as Intel, AMD, IBM and Sun are already building and/or delivering MCPs and announcing four and eight-core processors and longer term plans for processors with even larger numbers of cores. PC suppliers including Toshiba, Sony and Dell are already shipping MCP-based systems; and existing operating systems can exploit them, even if not to the fullest. The industry is promoting MCP technology, but is it ready, or able, to make use of it yet?
While the front end of the new systems may look conventional, there will be changes in the way processes are managed and executed. Much of the work, undertaken in the last twenty years or so on how to use clusters and networks of processors by among others the scientific community, is relevant to the use of MCPs. One of the big questions surrounding the use of MCPs is whether this work will be ignored or will influence the way in which MCPs are used. Not all the techniques used are directly applicable, but many of the lessons are.
The Market
In October and November of 2006 Concertant LLP undertook a survey of would-be MCP users drawn from across the spectrum of prospective users and developers, from strategists to programmers. The results are summarised in Concertant’s report “Multi-core Processors –Usage and Languages”. The full survey investigated areas including applications, programming languages, operating systems, hardware and market size, giving a unique insight into the technology.
Almost sixty percent of respondents were clear that MCPs were already making their impact felt on the desktop (see Figure 1). Only eight percent felt that MCPs were in the far future for them, while those in the embedded systems market were at some pains to point out to what extent MCPs were already part of their life!
While MCPs may be being delivered to the domestic and commercial markets, it is not at all clear that there is, as yet, a “killer app” or a new class of killer applications. The survey group felt that early applications areas are going to be gaming, some areas of finance, scientific applications, streaming audio and video, as well as many deep-embedded applications. For this reason, there is wide support for the idea that at least in the short term MCPs will benefit the wider market through their ability to offer support for applications seeking “more of the same, only faster”.
Figure 1: Is MCP processing reaching the desktop? (By percentage of respondents.)
At the same time the systems market will benefit because of the throughput improvements that MCPs can bring to servers and communications operations. The database and storage arena are already looking to MCPs to provide increases in cutting-edge performance.
The benefits MCPs offer are remarkable and with the continuing backing of major manufacturers are sure to become widespread. While there remain several issues surrounding certain aspects of integration with more conventional systems, market drivers will lead to their resolution. The market appears to be set for substantial growth over the coming years.
Now we await the new breed of killer app.