In a phone conference this morning, Intel gave a little preview of the papers it plans to unveil at the VLSI Symposia on Technology and Circuits this week in Hawaii. Much of the discussion focused on hairy electrical engineering specifics, but one of the speakers revealed a few interesting details about what Intel has in store for its next-generation Nehalem processor architecture.
Intel Fellow Rajesh Kumar opened his expose by throwing around some bandwidth numbers. Nehalem can hit 25GB/s of socket-to-socket bandwidth and a staggering 32GB/s of main memory bandwidthâ€”figures Kumar said are both "about 3X larger than [Intel's] best competition today." To achieve such prodigal bandwidth, Intel implemented a technique called low-jitter clocking, which can "reduce uncertainty in clocks by, in some cases, an order of magnitude compared to what was achieved before."
Kumar also went into a little bit of detail about how Intel designed Nehalem to scale across mobile, desktop, and server segments. The processor runs all of its internal componentsâ€”the CPU cores, memory controller, and I/Oâ€”in a decoupled fashion, so one can tune their respective frequencies and voltages independently. This isn't a new idea, Kumar stressed, but Intel's implementation is new in that it uses a synchronous interface between those components. Most past implementations have asynchronous interfaces, he claimed, which result in both higher latency and indeterminismâ€”"if you test five different systems, you will get five different results." Because of the synchronous approach, Nehalem's memory-to-cache latency is allegedly "drastically smaller" than that of the competition.
Last, but not least, Kumar briefly described Nehalem's adaptive frequency clocking system. If I understood this part correctly, the processor basically adapts its frequency every cycle based on power draw. As a result, Kumar said Nehalem can hit a higher frequency at a given voltage or hit a lower voltage at a given frequency compared to existing chips. However, the CPU also averages the effects of clock speed variations after every few cycles, so from an outside perspective, "you are getting a fixed frequency all the time."
Nehalem remains on track for a launch in the fourth quarter of this year.