The AMD FX is the mid-range and high-end CPU family from AMD, using either the “Bulldozer” or the “Piledriver” microarchitectures. Let’s take a look at the main specifications of all models released so far.
For a better understanding of what is new in the “Bulldozer” microarchitecture, please read our “Inside the AMD Bulldozer Architecture” tutorial. The “Piledriver” microarchitecture is a “fine-tuned” “Bulldozer” microarchitecture.
On these microarchitectures, the CPU is comprised of “modules,” each module containing the equivalent of two CPU “cores.” However, these cores only process integer instructions, and there is only one “core” for processing floating-point instructions in each module. Furthermore, these “cores” share the same front-end engine (L1 instruction cache, fetch unit, and instruction decoder). The operating system recognizes each module as two processors, even though these processors are not complete processors as they would be on a true dual-core CPU, where each core has its own front-end engine and floating-point execution engines. So, on a “quad-core” FX processor, the CPU uses two modules.
All AMD FX processors have an automatic overclocking feature called “Turbo CORE.” This feature was first introduced with the Phenom II X6 processors; however, the AMD FX processors have two “Turbo” clocks. The first one, called “Turbo Core,” is used when the CPU “feels” the application needs more processing power, and there is still enough headroom in the CPU’s maximum dissipation power for it to increase its clock rate. For instance, if the processor TDP is 125 W and the CPU is currently dissipating 95 W, it “knows” that it still has 30 W “unused” and, thus, can increase its clock rate. In this mode, all CPU cores may be active.
The second “Turbo” clock is called “Max Turbo,” and is used when there are unused processing cores available. In this case, the CPU puts these unused cores to “sleep,” allowing it to increase its clock rate and thermal dissipation even further.
The main specifications for the AMD FX processors include:
- Socket AM3+
- Bulldozer (models where the second digit is “1”) or Piledriver (models where the second digit is “3”) microarchitecture
- Four (models starting with “4”), six (models starting with “6”), or eight (models starting with “8”) processing “cores”
- 64 kiB instruction L1 memory cache per module (i.e., shared by each pair of “cores”)
- 16 kiB data L1 memory cache per core
- L2 memory cache of 2 MiB per module (e.g., an eight-core CPU has four 2 MB L2 memory caches)
- L3 memory cache of 4 MiB or 8 MiB shared by all cores
- Memory controller running at 2 GHz or 2.2 GHz, depending on the model
- Support for DDR3 memories up to 1,866 MHz supporting dual-channel architecture
- HyperTransport 3.0 bus supporting the higher speeds of 2,400 MHz (9.6 GB/s) and 2,600 MHz (10.4 GB/s), not available on socket AM3 processors
- Two “Turbo” clocks on most models
- Unlocked clock multiplier, allowing overclocking by increasing the CPU clock multiplier
- Support for the SSE4 and AVX instruction sets, feature not available on previous processors from AMD
- 32 nm manufacturing process
Let’s now see all of the models of the AMD FX processor released so far.