Benchmark Results: Synthetics
Synthetic benchmarks are often forward-looking measures of performance that predict the direction of things as the software community gets better at taking advantage of today's hardware features. PCMark Vantage centers entirely on Windows Vista-based apps, though, so its results are actually a bit more meaningful than some of the unrealistically-demanding workloads seen in the past.
Here we see the Phenom II X4 965 trounce Intel’s Core 2 Quad Q9550S and even best the stock-clocked Core i7-920, which gets a nudge up to 2.8 GHz with Turbo Boost. Most interesting, though, is the fact that the overclocked Phenom II gains nothing, while the X4 965 is quite a bit faster than the X4 955 operating 200 MHz slower.
A gauge of gaming performance, 3DMark Vantage also favors the latest Phenom II in its overall suite score. The CPU AI and physics tests overwhelmingly favor the Core i7’s Nehalem micro-architecture.
Despite losing out to Core i7 in Sandra’s arithmetic tests, Phenom II dominates the multi-media metric.
Changing clock speed doesn’t really do anything to the Phenom’s memory bandwidth performance (although taking away one core seems to cost some throughput). Core i7 and its triple-channel memory controller serve up the biggest bandwidth numbers, while Core 2 Quad’s northbridge-based controller hampers performance in comparison. When pressed for an explanation as to why the triple-core Phenom II reflects less throughput, SiSoftware's engineers had this to say:
"Single-core test cases cannot utilize the full bandwidth of the integrated memory controller. It is an issue related to local resources in the internal tables of the core. You need to hold an entry in the table allocated to memory access until the memory returns with the data. Since the table size is limited, you can’t issue a new request to the memory subsystem until one of the previous requests is returned (only then you can allocate the table entry needed to hold the data). This problem would go away if you had an infinite number of entries."
