FX versus Phenom: Performance per Clock

After seeing how the new core architecture scales with multiple threads, I wanted to see how much the IPC (instructions per clock) of Bulldozer has changed from the Phenom II chips currently on the market.  The easy definition of processor performance is IPC x Clock so it isn’t entirely fair to judge the performance of the AMD FX processor based on IPC alone, but this is interesting information that will again help us understand results later in the review.

For this test, I took the same POV-Ray benchmark used on the previous page and tested it using just a single core, cores 1 and 2 and then with all cores enabled to get the maximum result.  Both processors, the FX-8150 and Phenom II X6 1100T, were clocked at 3.30 GHz and fixed there.

Clock for clock, single threaded performance is about 19% slower on the FX / Bulldozer architecture than on the Phenom II CPU.  Obviously this is an important issue and indicates that AMD has to have at least a 20% frequency advantage over Phenom II just to stay relevant in the lightly-threaded application workloads.

Note as well that enabling all cores on this test still didn’t allow the Bulldozer CPU to outperform the Phenom II; considering that the FX-8150 is running with 4 modules / 8 cores while the X6 part uses six full size cores, you can see how the module design is very similar in some circumstances to HyperThreading technology used by Intel.

« PreviousNext »