Skip to main content

Part 2: How Many CPU Cores Do You Need?

Synthetic Benchmarks: SiSoftware Sandra

The processor benchmarks demonstrate a linear decrease in performance that is very similar to the Core 2 Quad results we saw previously. Synthetics like this one highlight the theoretical difference in multithreaded optimizations, while many of the applications we test aren't quite as optimized for threaded execution.

The multi-core efficiency benchmark shows that the Phenom II communicates between CPU cores well, no matter how many are utilized. This is a contrast to the results we saw with the Core 2 Quad Q6600, when there was a dramatic drop in bandwidth from four to three CPU cores.

When we previously tested the Core 2 Quad, we saw absolutely no difference in memory bandwidth as CPU cores were disabled—with the Phenom II, we're seeing a definite drop-off that becomes noticeable between three to two CPU cores, with a surprisingly large drop in bandwidth when a single CPU core is used.

We certainly had a second look at these results. After further scrutiny and testing, they seem to hold up—previous Phenom II reviews have shown X3s and X2s have lower memory bandwidth than the X4s, and when we benchmarked the Phenom II X2 550 on our test platform to compare, the bandwidth was right on target.

We showed our results to AMD and asked if it had an explanation for this phenomenon. According to AMD, this is a bug with the SiSoftware app, and it happens from time to time when new silicon is released. AMD says it has communicated this issue to the makers of Sandra.

It's good to know that a software bug caused these results, but even if they were accurate, no single-core Phenom II X1 CPU exists that would be crippled by this theoretical result, anyway. While the X2s also appear to suffer a small penalty, it shouldn't be enough to cause a notable performance hit.

Now that we have lined up some expectations based on the synthetic results, let's move on to the applications.