OK, so I have to confess that I wasn't at all convinced by this but decided to put it to the test using the Gears of War 5 benchmark and I have to say that the results surprised me. Tests were carried out with Ultra video quality, 2560x1440 resolution, vsync disabled and no frame rate limits. GPU settings were made in Wattman and power, temp & fan readings taken using HWINFO64. I started the test with 1602 MHz target GPU clock and 900 mV P7 voltage, and increased by 10 MHz and 25 mV increments up to 1702 MHz and 1150 mV. Power limit in all custom tests was 50% but the fan PWM setting for 72C was set at 60% up to 1652 MHz and increased to 80% for higher clocks to avoid excessive temperatures. HBM2 frequency was 980 MHz for all custom tests. I also tested the 3 built-in Wattman profiles for comparison.
I've embedded the graphs and table of results below which can be summarised as follows:
- All 3 built-in profiles produce woeful performance. In all my earlier testing I found that high core voltage with low power limit combinations produced poor results compared to low core voltage and high PL - hence why I am only using 50% PL here.
- Increases in FPS gradually decrease with increasing MHz/mV up to a peak at 1692 MHz / 1100 mV, after which the FPS reduces again, presumably due to throttling. From 1602 to 1692 MHz, the FPS range is a mere 6 FPS or 9.3% with this title & settings.
- The power consumption delta from lowest to highest is 110 Watts or 68%! Interestingly power does not increase linearly with the mV which I'm guessing will be due to internal power limit and temperature management algorithms.
- Watts per frame neatly shows a much more favourable efficiency at lower clock speeds and significantly diminishing returns at higher clocks.
- Fan speed increases in 4 distinct ranges: at 1602-1622 MHz, fan speed peaks around a comfortable 1670 RPM; at 1632-1652 MHz, this increases to a bearable ~1920 RPM; at 1662-1682 MHz, rotation notches up another 200-300 RPM and fan noise becomes intrusive; and finally at the 1692-1702 MHz range, the screaming fan is just offensive.
- GPU temperature rose from 58C to 68C between 1602 and 1662 MHz and plateaued at 70/71C from 1672 MHz. At this max temperature, VRM and hotspot temps were in the 85-92C range (too hot).
So the upshot of all this is that, for me, there is no clear winner setting as all options from 1602 to 1662 MHz are acceptable - it would depend on the priority between FPS vs Power/Temp/Noise which could vary according to the titles being played. If I had to pick a sweet spot setting it would be 1622 MHz / 950 mV since this achieves lowest fan noise, excellent power efficiency, and acceptable performance with 6 FPS above stock and only 3 FPS behind the 1662 MHz / 1050 mV profile. In reality I will probably keep 2 custom profiles: 1662 MHz / 1050 mV for the games that need the extra horsepower, and 1622 MHz / 950 mV for the games that don't. I haven't fully game tested these overclocks yet so further tweaking may still be required
Thanks for your help Dan
. Care to run a shadow of the tomb raider benchmark at 1920 x 1080 high settings with your best overclock?
