I've tried and failed on numerous occasions to overclock my Fury X. Had the core clock at around 1100 and it would crash playing ROTR and show some pretty hardcore blue artifacts running Firestrike. However I never considered increasing the voltage and after reading AMD Matt's post I checked and my card was running at a default of 1.05v at full load. I've just increased the voltage to +6 Mv using MSI Afterburner and set the core clock to 1110. I seem to have achieved a stable overclock as have just played ROTR without issue and ran Firestrike with no artifacts. The voltage is sitting around 1.15v full load now so likely a bit more room to go further based on Matt's advice regarding 1.25 max but I'll leave that for a rainy day. Also gained 2.7 average FPS on the Deus Ex - MD bench mark and increased the minimum FPS by 1.5. Happy with that!
Download AIDA64 and install it. Open it, go to View, enable the status bar and in the bottom left corner where it says the Aida version, right click and select video debug, ATI resistors. This will give you a long report and in that report will be your default voltage and clock states. The lower the DPM7 state the better. Of course it still comes down to the lottery, but it should give you more voltage headroom before you hit diminishing returns, or negative scaling.
Here are both GPUs from each of my Pro Duo's.
Pro Duo 1 Core 1:
------[ GPU PStates List ]------
DPM0: GPUClock = 300 MHz, VID = 0.90000 V
DPM1: GPUClock = 508 MHz, VID = 0.95000 V
DPM2: GPUClock = 717 MHz, VID = 0.95600 V
DPM3: GPUClock = 874 MHz, VID = 1.03100 V
DPM4: GPUClock = 911 MHz, VID = 1.06800 V
DPM5: GPUClock = 944 MHz, VID = 1.10600 V
DPM6: GPUClock = 974 MHz, VID = 1.14300 V
DPM7: GPUClock = 1000 MHz, VID = 1.17500 V
Pro Duo 1 Core 2:
------[ GPU PStates List ]------
DPM0: GPUClock = 300 MHz, VID = 0.90000 V
DPM1: GPUClock = 508 MHz, VID = 0.94300 V
DPM2: GPUClock = 717 MHz, VID = 0.95600 V
DPM3: GPUClock = 874 MHz, VID = 1.06800 V
DPM4: GPUClock = 911 MHz, VID = 1.10600 V
DPM5: GPUClock = 944 MHz, VID = 1.14300 V
DPM6: GPUClock = 974 MHz, VID = 1.18700 V
DPM7: GPUClock = 1000 MHz, VID = 1.21800 V
Pro Duo 2 Core 1:
------[ GPU PStates List ]------
DPM0: GPUClock = 300 MHz, VID = 0.90000 V
DPM1: GPUClock = 508 MHz, VID = 0.95000 V
DPM2: GPUClock = 717 MHz, VID = 0.95600 V
DPM3: GPUClock = 874 MHz, VID = 1.03100 V
DPM4: GPUClock = 911 MHz, VID = 1.06800 V
DPM5: GPUClock = 944 MHz, VID = 1.10600 V
DPM6: GPUClock = 974 MHz, VID = 1.14300 V
DPM7: GPUClock = 1000 MHz, VID = 1.17500 V
Pro Duo 2 Core 2:
------[ GPU PStates List ]------
DPM0: GPUClock = 300 MHz, VID = 0.90000 V
DPM1: GPUClock = 508 MHz, VID = 0.94300 V
DPM2: GPUClock = 717 MHz, VID = 0.95600 V
DPM3: GPUClock = 874 MHz, VID = 1.06800 V
DPM4: GPUClock = 911 MHz, VID = 1.10600 V
DPM5: GPUClock = 944 MHz, VID = 1.15000 V
DPM6: GPUClock = 974 MHz, VID = 1.19000 V
DPM7: GPUClock = 1000 MHz, VID = 1.22500 V
I would recommend overclocking in 20Mhz steps and increasing voltage two notches (+6v = 1 notch) at a time. Using a simple repeatable benchmark (the original Tomb Raider is perfect) and keep going until you start to lose performance. Find the sweetspot and bobs your uncle.
If you have a good sample with a low DPM7 state voltage then you'll be able to hit 1.275v+ before things go south.
