Now that a billion-plus transistors can fit onto one processor chip, the race to push CPU clock rates has reached a point of diminishing returns due to increasing power consumption. All is not lost, though.
By turning to parallelism, CPU designers can still leverage all of those transistors to garner more performance. This past year saw the arrival of dual-core x86 CPUs from AMD and Intel, along with the promise of even higher core counts in future processors. Multicore and multithreaded CPU designs are fast becoming the mainstream for next-generation servers, desktops, and even game consoles.
Although the end systems they'll wind up in are just being released, the triple-PowerPC-based CPU developed by IBM and Microsoft for the Xbox 360 and the multicore CELL processor chip for the Sony PlayStation3 were unveiled earlier this year (Fig. 1). The CELL processor, a joint development between IBM, Sony, and Toshiba for the PlayStation3 and other systems, combines a PowerPC CPU and up to eight high-performance programmable coprocessors (see "CELL Processor Gets Ready To Entertain The Masses" online at www.elecdesign.com, ED Online 9748).
In the nonvolatile memory arena, this year saw the introduction of the first 16-Gbit flash-memory chips from Samsung. These chips are the highest-capacity devices yet released and can replace many of the micro hard drives now used in portable music players (Fig. 2). Thanks to 50-nm minimum features and a novel 3D transistor architecture, Samsung's designers shrank the cell size to just 0.00625 m2/bit—the smallest yet achieved for a commercial device.
SEEING IS BELIEVING
A more realistic gaming experience—on a PC or gaming console—is the charge of advanced graphics chips from ATI and nVidia that deliver 3D images that are more lifelike. The ATI Radeon X1800 graphics engine packs 16-pixel shader processors, eight vertex shaderprocessors, a 256-bit eight-channel GDDR3/4 memory interface, and a native PCI Express 16-lane interface. The chip provides full-speed 128-bit floating-point processing for all shader operations. It can execute up to 512 simultaneous pixel threads.
The nVidia Quadro FX 4500 and the GeForce 7800 GTX also offer 128-bit floating-point support through the rendering pipeline. They support the company's scalable link interface (SLI) frame rendering, which allows users to link two of the cards in one system to accelerate graphics rendering. Both chips feature a 256-bit memory interface. The GeForce 7800 delivers a memory bandwidth of 38.4 Gbytes/s, while the Quadro FX 4500 produces a bandwidth of 33.6 Gbytes/s as well as a fill rate of 10.3 billion pixels/s.
MANY OTHER OUTSTANDING PRODUCTS
In addition to CPUs, memory, and graphics, many other noteworthy products help to drive next-generation system designs. Multi-megagate FPGAs with 10-Gbit serializer/deserializer ports, low-power media processors, and high-resolution imaging chips are among the many candidates.