The Budget Overclocking Dance
Right out of the box, our A10-5800K spent most of the time at 4.0 GHz with Turbo Core enabled. And really, this is more than enough processor to serve our basic needs. Therefore, you aren’t going to see the aggressive CPU overclocking I'd normally pursue. It simply wasn't worth the risk of damaging our inexpensive motherboard to chase performance we don’t need. Rather, I wanted to maximize gaming potential by boosting the on-die graphics and memory subsystems without taxing AMD’s boxed cooler or overheating the motherboard's power circuitry. We didn’t suffer any losses or major setbacks, but to put it mildly, tweaking this rig was a time-consuming challenge.
MSI’s FM2-A75MA-E35 shipped with an updated v2.0 UEFI, which supports the newer Richland-based APUs. This would have been good news, that is, if the firmware wasn’t belligerent in accommodating my tweaking intentions. And unfortunately, once 2.x is running, there is no way to flash back to an earlier version. Hopes of 1833 MT/s memory at 1.55-1.6 V were dashed when DRAM voltage settings all read N/A. In fact, the UEFI didn’t even list the current memory voltage. Subsequently, utilities like HW Monitor and AMD OverDrive were unable to pick up my data rates either. Worse still, the northbridge voltage needed for stable graphics core overclocking was also unavailable. We only had control over the CPU and CPU-NB voltages, and even those were limited.
It's possible that MSI knows the limits of its own board and purpose disabled some of the tweaking features. But I wasn’t happy with the Trinity architecture's graphics prowess, especially hampered by 1600 MT/s RAM and subjected to SBM game settings specifically tuned for discrete graphics solutions.
So, I decided to continue my exploration. First, I took a rare look at MSI’s OC Genie II auto-overclocking feature. How much performance could we gain from a simple click-and-save approach within the UEFI? Basically, all this does is boost the reference clock from 100 to 105 MHz, affecting numerous frequencies. The CPU clock rate jumps to 3990 MHz (from 3800 MHz), and potentially reaches 4410 MHz under light loads with Turbo Core. The CPU-NB frequency is raised to 1890 MHz, main memory to 1680 MT/s at default CAS 9 timings, and the graphics engine winds up at 840 MHz. After confirming that load voltages and temperatures remained within reason and that the settings were stable, I decided to use this as the main overclock throughout the test suite. For gaming, however, I wasn't ready to wave a white flag just yet.
AMD’s OverDrive overclocking utility isn’t a favorite with enthusiasts, but it can be useful for saving time. And on some budget hardware, it may reap benefits beyond what you can do in the BIOS. Most options in OverDrive were unavailable, and all but a few voltages read zero. However, I did end up with a bit of control I didn't have in MSI's firmware, namely, the ability to bump northbridge voltage. The stock NB VID was 1.175 V, but read 1.2 V in OverDrive. If I raised it manually to 1.2 V and tampered with GPU frequencies, the voltage jumped to 1.275 V when new settings were applied. While I'd prefer to fine-tune this, I was able to squeeze a more respectable 950 MHz from the GPU, rather than topping out at 894 MHz.
After running all of our normal tests, we're going to take a specific look at the performance gains from overclocking the on-die GPU compared to simply running higher-speed memory. I either lucked out with cooperative RAM, or the motherboard was over-volting the heck out of it; I have no way of knowing. But I did find stable timings overclocked to both DDR3-1866 and -2133. I felt 1866 MT/s was safer, and focused on testing that setting.
Tampering with AMD OverDrive did have some detrimental consequences. It broke the ability to specify the CPU voltage or multiplier within MSI's firmware. Whenever I tampered with the GPU, OverDrive kept enabling Turbo Core. Inevitably, some cores would also throttle back down to below base frequencies, on occasion. Nervous about unknown voltages, and the potential for self-destructing VRMs, I didn’t want to overclock the CPU at all. In the end, I didn’t pursue fine-tuning frequencies by adjusting the base clock.