     |
 |
 |
 | | | | |
 |
|||
 |
||||||
          |
 |
 |
||||
|
A Glimpse Inside the Cell Processor
Remember your first time? Programming a processor, that is. It must have seemed both exciting and challenging. You ain't seen nothing yet. Even garden-variety microprocessors present plenty of challenges to an experienced programmer or development team. Now imagine programming nine different processors all at once, from a single source-code stream, and making them all cooperate. When it works, it works amazingly well. But making it work is the trick. That's the task that IBM, Toshiba, and Sony took on several years ago in creating Cell, an amazing new multiprocessor microprocessor chip that will debut later this year in Sony's PlayStation 3 video game console, Toshiba's high-end televisions, and IBM's blade servers. Cell is the beginning of a new family tree for all three companies, and it promises to branch into consumer, computer, and embedded systems for many years to come. Let's take a look at what's inside the first Cell processor; for an insider's look at Cell's programming model, see Alex Chow's article [to appear on Gamasutra next week]. Insanely Complex, Insanely Cool There aren't many businesses where manufacturing technology exceeds design technology. Throughout human history we've been able to dream up things we can't yet build, like spaceships, skyscrapers, jet packs, underwater breathing apparatus, or portable computers. But in the semiconductor business the situation is reversed: chip makers can build bigger and more complicated chips than they can design. Manufacturing prowess exceeds design capability. We can fabricate more transistors than we know what to do with. Cell goes a long way toward ending that pattern. It's an insanely complex chip with about 250 million transistors (compared with 125 million for Pentium 4) that runs at more than 4GHz. With just the right conditions Cell can crank through 256 billion floating-point operations every second, according to IBM. What's remarkable is that Cell wasn't developed for scientific applications, military computers, or code breaking. Instead, Cell is primarily intended for entertainment. It says a lot about embedded systems when three of the world's top multinational corporations devote millions of dollars in R&D and thousands of personnel into developing one of the world's most complex processors--for toys. That's not to say that Cell can't also be used for more "serious" applications. As we'll see, Cell lends itself to parallel processing, which can be useful for scientific simulations and medical imaging. The fact that it also plays Gran Turismo is just a bonus. Figure 1 shows a top-level block diagram of the first processor in the Cell family, more formally known as the Cell Broadband Engine, or CBE. Cell is based around a single 64-bit PowerPC processor surrounded by eight identical coprocessors. The central PowerPC processor can execute two instructions at a time. That's nice, but not unusual. Intel, AMD, and others also produce dual-issue processors, even 64-bit dual-issue processors. So the heart of Cell is actually--dare we say it--fairly pedestrian.
Note that although the central processor is based on IBM's PowerPC architecture, it's a new design, not an existing PowerPC core. Cell's central processor is similar to the current PowerPC 970 chip, although it's not an exact match. The central processor includes the VMX (visual media extensions, similar to AltiVec) instruction-set extensions to the base PowerPC instruction set. The central processor has a pair of 32K first-level (L1) caches and a unified 512K second-level (L2) cache. This should keep the processor humming along at 4GHz.
|
|||||||||||||||||||||||||||||||||||||||||||
Features
Introduction|
|
|