This is the third in a short series of articles that Techworld will be publishing over the next few days about multi-core processing and its ramifications. Read the first article here.
Talk of multi-core chips has been around for some time now, and at last the chips are starting to arrive -- albeit slowly, and for relatively limited applications from an enterprise systems perspective. That's not going to remain the situation for long though.
Systems using Intel's first dual-core chip, the Pentium Processor Extreme Edition 840 running at 3.2GHz, went on sale on 18 April 2005. These initial multi-core systems weren't aimed at the business market but rather at computer hobbyists and entertainment enthusiasts. They offer increased capabilities for audio, high-definition video and 3-D visualisation.
According to an Intel spokesman, the company has more than 15 multi-core projects under development and is set to introduce its mainstream, high-volume dual-core Pentium D processor later this year. Incidentally, these multi-core CPUs don't come cheap: the Intel Pentium Processor Extreme Edition 840 is priced at $999 in 1,000-unit quantities.
For the enterprise market, Intel plans to ship a dual-core 64-bit Itanium 2 processor, code-named Montecito, built using 90nm process technology, in the second half of 2005. The next-generation chip will contain more than 1.7 billion transistors and a 24MB cache.
In the first quarter of 2006, Intel intends to deliver two optimised 64-bit Xeon dual-core processors.
AMD's first multi-core releases, the Opteron family, feature one die with two CPU cores. Each core has separate Level 1 and L2 cache hierarchies. The two CPU cores share the same integrated memory controller and HyperTransport technology resources that AMD has used in its single-core Opteron processors.
Dual-core AMD Opteron processors can be inserted into existing 940-pin sockets, provided that the BIOS is updated. AMD announced its first dual-core server and workstation products on 21 April.
Initial reports indicate that AMD's dual-core processors run faster than Intel's. Dual-core client processors are expected to follow in the second half of this year.
Sun has been talking seriously about multi-core processors since 2003. Later this year, Sun is expected to ship its next-generation UltraSparc IV CPU. Code-named Niagara and designed for network-intensive tasks, the new CPU could put as many as eight cores onto a single chip destined for blade servers.
Further down the UltraSparc road map is the Rock family, which will be tailored to more computational tasks, such as databases, ERP systems and encryption, according to David Yen, executive vice president of Sun's scalable systems group.
UltraSparc is primarily used in servers, and it competes with IBM's Power processors and Intel's Itanium and Xeon chips. Sun also plans to market systems using AMD Opteron dual-core CPUs.
IBM's 64-bit Power5 CPU, introduced last year, features two processor cores with separate L1 (64KB instruction, 32KB data) cache but shared onboard L2 (1.875MB) and off-chip L3 (typically 36MB) caches. These are used primarily in IBM's own server families running the AIX, Unix and Linux operating systems.
IBM is also engaged in a completely separate multi-core processor project, the Cell microprocessor, with Sony and Toshiba. Designed for compute-intensive workloads and broadband rich-media applications, the Cell includes a 64-bit Power processor core connected to eight synergistic digital signal processing cores capable of floating-point processing. Cell's multi-core architecture and high-speed communications capabilities are designed to support multiple operating systems simultaneously.