Some new SPEC CPU2006 results were published last month that might cause some of you to take another look at SPARC. Whether we call it Fujitsu SPARC Enterprise M8000 or Sun SPARC Enterprise M8000, its SPECfp_base2006 result of 11.1 puts SPARC back in the vicinity of PowerPC (e.g. POWER5+). The SPARC Enterprise M8000 server showcases Fujitsu's SPARC64 VI microprocessor, an implementation of the SPARC-V9 architecture. This result shows SPARC64 VI as 2.75 times faster than UltraSPARC IIIi.

Take a look at the throughput performance vs. single thread performance graph here. I really like that kind of graph and I want to see more progress to the right (e.g. Rock). Computer manufacturers tend to talk more about throughput when their single thread performance lags the competition. Single thread performance is important too. Not all applications are, or can be, massively multithreaded. And sometimes, what we care most about is how long we're going to have to wait for the computer to complete a single task.

It's good to see this kind of success with SPARC. Here's a measure of architectural efficiency that I devised: 1/(R&D dollars * execution time). I have a hunch that the SPARC-V9 architecture, because of the relative ease of implementing microprocessor designs based on it and because of performance-enhancing architectural features (e.g. register windows), outperforms the legacy IA-32 architecture (with its expansion to support 64 bits) on that efficiency metric. SPARC R&D is money well spent.