All tests were performed on the following testbed:
- Processor: AMD A8-3850 (Llano, 4 cores, 2.9 GHz, 4 MB L2, Radeon HD 6550D).
- Processor cooler: Thermalright Ultra-120 eXtreme with Enermax Everest fan.
- Mainboard: Gigabyte GA-A75-UD4H (Socket FM1, AMD A75, BIOS F4).
- Memory: 2 x 2 GB DDR3-2133 SDRAM 9-9-9-27 (GeIL GE34GB2133C9DC).
- Hard drive: Kingston SNVP325-S2/128GB.
- Power supply unit: Tagan TG880-U33II (880 W).
- Operating system: Microsoft Windows 7 SP1 Ultimate x64.
- AMD Catalyst 11.7 Display Driver;
- AMD Chipset Driver 8.871.
Practice: A8-3850 Overclocking on Gigabyte GA-A75-UD4H
Let’s try and apply the above described overclocking strategy to a real system. We will also try to estimate how far we can go in our overclocking attempts using Gigabyte GA-A75-UD4H mainboard. In our opinion, maximum overclocking is a mode when the processor and all other system components work at the maximum possible frequencies, but the system remains fully stable. We test system stability using or traditional tools. Processor stability is verified with LinX 0.6.4, and graphics core stability is confirmed in Furmark 1.9.1 and Futuremark 3DMark 11.
I have to stress that 3DMark 11 is an irreplaceable tool during Llano overclocking. It not only allows estimating the practical effect from increasing the processor and graphics core frequencies, but may also be considered an excellent system reliability test. The thing is that even if you confirm stability of the graphics and computational cores separately, it doesn’t necessarily mean that the complete Llano system is going to be stable, too. The best way to perform this final stability check is to run Combined Test from 3DMark11 suite. It simultaneously loads the graphics core as well as the computational cores and the memory controller, and in most cases it crashes the system that has just successfully passed both: LinX and Furmark tests.
Among other utilities that may be helpful during Llano overclocking I should also mention some diagnostic tools. Unfortunately, A-series processors aren’t supported in AMD’s proprietary Overdrive utility for some marketing reasons. So, we will have to resort to third-party tools. For example, there is absolutely no way we could do without CPU-Z, which versions 1.58 and higher report the current frequency of the Socket FM1 processors absolutely correctly.
Things are not rosy with processor temperature monitoring. The existing utilities monitoring CPU temperatures do not work well with Llano. They do have access to the thermal diodes inside the CPU cores, but cannot interpret the readings correctly. Therefore, such programs as Aida64, HWMonitor, CoreTemp, SpeedFan and others do show some processor temperatures, but the readings sometimes look pretty strange and hardly have anything to do with the real processor temps.
Therefore, the best option in this case would be the monitoring utilities offered by the mainboard makers. Gigabyte’s utility is called EasyTune6. Even though programs like this take their temperature readings off the thermal diodes beneath the processor socket, their readings are much closer to reality than what the utilities get from the thermal diodes inside the CPU cores. Another advantage of EasyTune6 is the option for adjusting the BCLK frequency right from the operating system environment and without any system reboot to follow.
So, let’s check out the obtained results now. Our Gigabyte GA-A75-UD4H with AMD A8-3850 processor remained fully stable at the maximum BCLK frequency of 141 MHz.
Frankly speaking, we could increase this frequency even more, as the system was still able to pass most of the tests, but it was at this particular frequency that its stopped passing the Combined Test from 3DMark11 suite. Therefore, we decided to play it safe and to stop at this setting. The graphics core in this case worked at 846 MHz.
When we increased the processor core voltage by 0.125 V and the graphics core voltage by 0.1 V, the system remained stable at the maximum CPU clock multiplier of 26x. In other words, the computational cores proved operational at 3.66 GHz frequency.
At this point maximum processor temperature reached 88°C according to under-the-socket diode, but we didn’t have any overheating issues during the stability tests.
As for the memory frequency, at 141 MHz BCLK we could only use x13.33 multiplier, so our modules functioned as DDR3-1879.
However, GeIL EVO PC3-1700 modules used in this test session can work at a much higher speed of 2133 MHz. That is why we also tried a different overclocking approach: by lowering the BCLK frequency to 133 MHz and switching the memory to DDR3-2133 mode with x16.0 multiplier. In this case the graphics core frequency is 800 MHz, and we can use 27x processor clock multiplier setting the processor core frequency to 3.59 GHz.
However, despite all this, the results in 3DMark vantage are just as good: higher memory frequency makes up for lower GPU frequency and lower CPU frequency, too. So, this alternative approach to A8-3850 overclocking also proves absolutely justified.