AMD Bulldozer / FX-8150 Desktop Performance Review

Computer hardware enthusiasts have literally waited for years for AMD's Bulldozer architecture to come to market and we finally see this today in its desktop form, code named Zambezi, brand named AMD FX. In this article we share with you our analysis of Bulldozer's performance in synthetic benchmarks and desktop applications.


Synthetic Benchmarks

Synthetic benchmarks.....are.... well....synthetic and do not get your work done for you. These don't necessarily mean anything when it comes to desktop performance in real life, but certainly there is value to be gleaned from these by enthusiasts. These are also the easiest benchmarks to run when it comes to seeing if your system at home is measuring up.

Stock processor clocks used with 1333MHz memory clocks. Overclocked processors use 1600MHz memory clocks. Our exception is the overclocked i7-920 which uses 20*200 CPU clock and 1680MHz memory clock.

Article Image

All of these are dual channel DDR3 systems with the exception of our i7-920 and i7-990X systems which are of course triple channel DDR3. No surprises are shown here. Surely though, the new Bulldozer memory controller is capable of delivering much more memory bandwidth than the older Thuban core. While the Bulldozer memory controller is not quite up to catching the Sandy Bridge technology, AMD has certainly closed what was once a gaping difference.

Article Image

Sandra Dhrystone which calculates integer math, no floating point math is done, shows us a bit different picture of Bulldozer. Here we see our FX-8120 and FX-8150 act more like you might expect. The 8120 pulls ahead of the 2500K and the 2600K edges out the 8150. The $1000 i7-990X dominates the board though with 6C/12T performance.

Article Image

We are again visiting Hiper Pi / Super Pi which calculates decimal places of Pi out to 1,000,000 places using a single thread. In this set of tests our stock processor have Turbo turned on unlike the previous comparison.

As you can see, even with stock clock Turbo turned on and our CPUs overclocked we see the same scaling.

Article Image

The Prime is a lie! wPrime actually calculates and verifies square roots, so more straight math. No prime numbers involved. We use it a bit differently than Hiper Pi though in that we run wPrime across all cores and threads available.

Our $1000 990X comes out on top here, with the 4-core i7-920, 2500K, and 2600K close behind in a pack. Much like we saw with Hiper Pi, our 1100T still outpaces our new Bulldozer 8120 and 8150.

We show you these synthetic benchmark numbers because we know you expect these. According to these, for the most part, Bulldozer is not competitive with Intel's current Sandy Bridge and Nehalem architectures. But let's move on to some scenarios that will represent a more real world basis for comparison.