Professional Help: Getting The Best Overclock From AMD's A8-3870K

Test System And Benchmarks

In an effort to best demonstrate multiple configurations, we first tested our system at stock frequencies using 800 MHz memory at 8-8-8-24-2T timings. Those are the same settings used in February's Llano-based overclocking story. Then, we overclocked Corsair's Vengeance kit to 933 MHz (the architecture's highest official data rate) using 9-9-9-24-29-1T timings, but kept stock CPU and GPU speeds.

We're also including a standard overclock using the same 933 MHz memory setting, changing only the CPU and GPU multipliers, but not the reference clock. This gives us 3.6 GHz processing cores and a 960 MHz GPU.

Finally, we include the highest memory overclock, a 117 MHz reference clock, 3627 MHz CPU cores, a 936 MHz GPU, and 1092 MHz (2184 MT/s DDR3) memory at 10-10-10-27-35-2T.

We're ultimately interested in how our all-out overclock compares to a discrete Radeon HD 6670 DDR3. So, we installed it in the overclocked system and disengaged its on-die GPU.

Socket FM1
AMD A8-3870K (Llano), Stock: 3.0 GHz,
Overclocked To 3.627 GHz @ 1.5 V
Asus F1A75-V Pro, Socket FM1, Chipset: AMD A75
Integrated GraphicsIntegrated Radeon HD 6550D, Stock: 600 MHz,
Overclocked To 960 MHz @ 1.3 V
Discrete GraphicsRadeon HD 6670 DDR3
800 MHz GPU, 900 MHz (1800 MT/s DDR3) memory
On-board Gigabit LAN controller
Corsair Vengeance CMZ4GX3M2A2000C10
2 x 2 GB, Max rating: 2000 MT/s, CL 10-10-10-27-2T,
Max overclock: 2184 MT/s @ CL 10-10-10-27-2T
Hard Drive
Western Digital Caviar Black 750 GB, 7200 RPM, 32 MB Cache, SATA 3Gb/s
ePower EP-1200E10-T2 1200 W, ATX12V, EPS12V
Software and Drivers
Operating System
Microsoft Windows 7 x64
DirectX 11
Graphics Drivers
Catalyst 12.7 Beta
Benchmark Configuration
3D Games
Battlefield 3
Version, Operation Swordbreaker, Fraps Runs
Elder Scrolls V: Skyrim
Version, Fraps Runs
DiRT 3
Version, Built-In Benchmark
StarCraft II
Version:, Tom's Hardware Guide Benchmark
Synthetic Benchmarks
3DMark 11Version:, Benchmark Only
SiSoftware Sandra 2011Version 2011.1.17.15, CPU Test = CPU Arithmetic / MultiMedia, Memory Test = Bandwidth Benchmark
Create a new thread in the UK Article comments forum about this subject
This thread is closed for comments
Comment from the forums
    Your comment
  • HEXiT
    cool... nice to get the input from some 1 who actually knows his apples.... rather than putting up a white board and saying this is what will happen in theory...
  • demtibor

    The article is good..but...

    If i can recommend something : Throw out the Vengeance modules, use more potent ones which can do around 2400 or above. As i noticed, Llano's memory controller has around 10% handicap compared to discrete solutions which means it's DDR3-2200 is roughly equal with discrete DDR3-2000. Of course this highly depends on the used timings.

    Here are my 24/7 Llano settings :

    ASUS F1A75-V Evo, 3870K, 2x2GB Geil EVO 2200 CL9, Freezer Pro 13.

    -125 BCLK
    -28x CPU multiplier (3500 Mhz ) with 1.475V
    -9,33x memory ( resulting 2333 Mhz memory clock ) with 9-11-9-28-2T-33-4-4-4 Timings. ( NB is around 1125 mhz as it has 9x ).
    -800 Mhz GPU clock ( resulting 1Ghz GPU in real because of the higher BCLK ).

    At this point i feel that memory performance is still a bottleneck. To be honest the board is capable of operating at 150 BLCK ( especially with the newest 1803 bios which solved the long-boot-time problem ( so far above 110 it took several seconds to initialize and post ) ) but there is no option to lower the north-bridge multiplier, and Llano does not tolerate well NB clock above 1125, even tough with voltage raising. It can operate in windows with 1200 NB, but will fail in 3D immediately. No matter how high voltage you apply.

    I have asked Asus to include that damn divider in their BIOS, unfortunately they rejected my request.