As I predicted last year, Intel is recovering from its mistakes by introducing new processors that are much more competitive with AMD’s chips. Meanwhile, AMD is suffering money woes again, but the scrappy company is developing some very interesting multicore processors with integrated ATI graphics. The coming year will be a great one for PC enthusiasts.
Intel is on track with Penryn, which improves the already-strong Core microarchitecture and shifts to the next-generation 45nm fabrication process. In addition to having smaller transistors, the new process will debut high-k metal-gate transistors, which I described in April. Along with other improvements, the new technology should make Penryn a winner.
You’ve probably heard about Penryn’s SSE4. With 47 new instructions, it’s the largest expansion of the x86 instruction set in seven years—even larger than the recent 64-bit extension. Digital video, 3D graphics, and scientific applications are driving the need for many of these new instructions. In some cases, programmers can replace a whole screen of code with a single SSE4 instruction.
The drawback? Programmers must use low-level assembly language to reap most of SSE4’s benefits. Vectorizing compilers that do the work automatically are still immature. Some SSE4 instruction mnemonics (such as MPSADBW and PHMINPOSUW) are so confusing that I need mnemonics to remember the mnemonics.
Penryn-based mobile processors will have a new Deep Power Down mode that drastically reduces the voltage to minimize power leakage while a core is idle. The OS issues a command to force the core into this mode; another command restores the original state. I’ve been worried that buggy or malicious software could repeatedly trigger this mode, creating a power-thrashing race condition that effectively disables the processor. However, Intel assures me that a special timer prevents a core from entering this mode too often. (We’ll see.)
Another Penryn feature for mobile processors is Enhanced Dynamic Acceleration Technology. Dumb name, clever idea. It temporarily boosts the clock speed of one core to handle extra work while another core is idle. This job-sharing prevents the idle core from having to power up to handle a small additional workload.
Overall, Penryn is a smart design with a mix of power-saving and performance-enhancing features. Sometimes little things add up to a lot.