Conclusion
While cache size only had a limited impact on the synthetic benchmarks such as PCMark05, the performance difference in most real-life benchmarks was significant. This was surprising at first, because experience tells us that performance differences can typically be found in most synthetic benchmarks, while little of it is eventually reflected in real-life benchmarks.
To answer the question in my title: Yes, cache size has become important, at least for the current Core 2 Duo processor generation. We used a 4 MB Core 2 Extreme X6800, a 2 MB Core 2 Duo E4400 and a Pentium Dual Core E2160, which effectively is a Core 2 Duo with only 1 MB L2 cache. All of the devices were operated on the same test system at a 266 MHz Front Side Bus speed, and with the multiplier set to 9 to reach 2400 MHz. The only difference is the cache size, since all current dual core processors except the old Pentium D are created equal. Yield issues or market demand determines whether Intel decides that a particular die can become a Core 2 Extreme Edition or a Pentium Dual Core.
If you compare the benchmark results of the 3D shooters Prey and Quake 4 with typical gaming benchmark results in our CPU Charts, the performance difference of 1 MB vs. 4 MB L2 cache roughly equals one clock speed increment. The same applies to the video transcoding benchmarks for the DivX 6.6 and XviD 1.1.2 codecs and file compression using WinRAR 3.7. The CPU-intensive benchmarks of 3DStudio Max 8, the Lame MP3 Encoder or the H.264 Encoder V2 by MainConcept don't benefit much from increasing cache sizes, though.
Still, Intel's concept of utilizing the available silicon real estate, which becomes available by shrinking dies to 65 nm and soon to 45 nm, makes a lot of sense for the Core 2 Duo architecture. The L2 cache has been known to be very efficient, especially since it is shared by both processor cores. Hence, it can level the impact of different RAM speeds and prevent Front Side Bus bottlenecks. And it does so very well, as we could see how the test processor's performance with only one megabyte second level cache fell clearly behind.
From this perspective, upgrading the L2 cache from up to 4 MB to a maximum of 6 MB for the upcoming 45-nm dual core Penryn processors (Core 2 Duo E8000 series) makes a lot of sense. Not only does the shrink from 65 to 45 nm give Intel more headroom to increase the cache size, but the company will again offer more performance thanks to the increased cache size. However, the most important benefit is due to how Intel can offer more processor variants with6 MB, 4 MB, 2 MB or even 1 MB L2 cache. In doing so, Intel utilizes an even higher percentage of the dies on a wafer despite some scattered defects that might have forced Intel to throw dies away in the past. Large cache sizes seem to be both important for both performance and Intel's balance sheet.
