Benchmarks
The following are a series of benchmarks, each comparing 6 settings:
- The factory settings
- Manual undervoltage by 100mV
- The “saving power” preset
- The 24/7 “Eco” Fashion (Sweetspot)
- Undervolting + simultaneous overclocking
- Maximum overclocking
3D Mark Fire Strike
Clock | Asic | Wall | GFX Score | Watt % | Score % | Score/Watt | |
Floor | 1550 | 264 | 501 | 23812 | 100 | 100 | 47,53 |
-100mV | 1590 | 257 | 494 | 24286 | 99 | 102 | 49,16 |
Eco | 1300 | 199 | 408 | 21354 | 81 | 90 | 54,36 |
UV Eco | 1500 | 178 | 393 | 23683 | 78 | 99 | 60,26 |
UV OC | 1590 | 210 | 440 | 24938 | 88 | 105 | 56,78 |
Max OC | 1710 | 340 | 635 | 26562 | 127 | 112 | 41,83 |
Red Dead Redemption 2 Benchmark 1440p “HUB Optimized” Settings
Clock | Asic | Wall | Fps | Watt % | FPS % | FPS/Watt | |
Floor | 1570 | 264 | 510 | 74,1 | 100 | 100 | 0,145 |
-100mV | 1607 | 220 | 460 | 75,2 | 90 | 102 | 0,163 |
Eco | 1460 | 198 | 420 | 70,9 | 82 | 96 | 0,169 |
UV Eco | 1507 | 164 | 373 | 74,9 | 73 | 101 | 0,201 |
UV OC | 1577 | 186 | 419 | 78,3 | 82 | 106 | 0,187 |
Max OC | 1714 | 303 | 607 | 82,7 | 119 | 116 | 0,136 |
Forza Horizon 4 1440p “ultra” Preset
Clock | Asic | Wall | Fps | Watt % | FPS % | FPS/Watt | |
Floor | 1630 | 240 | 491 | 104 | 100 | 100 | 0,212 |
-100mV | 1617 | 195 | 437 | 104 | 89 | 100 | 0,238 |
Eco | 1503 | 198 | 443 | 99 | 90 | 95 | 0,223 |
UV Eco | 1519 | 136 | 362 | 102 | 74 | 98 | 0,282 |
UV OC | 1602 | 162 | 399 | 106 | 81 | 102 | 0,266 |
Max OC | 1728 | 265 | 536 | 112 | 109 | 108 | 0,209 |
A Plague Tale: Innocence 1440p >high< (Einführung 60 Sek) (einführung=”” 60=””></ (Einführung 60 Sek)>
Clock | Asic | Wall | Fps | Watt % | FPS % | FPS/Watt | |
Floor | 1470 | 264 | 507 | 66,0 | 100 | 100 | 0,130 |
-100mV | 1577 | 264 | 508 | 70,5 | 100 | 107 | 0,139 |
Eco | 1230 | 198 | 406 | 57,0 | 80 | 86 | 0,140 |
UV Eco | 1490 | 193 | 395 | 68,7 | 78 | 104 | 0,174 |
UV OC | 1571 | 219 | 447 | 72,1 | 88 | 109 | 0,161 |
Max OC | 1692 | 349 | 645 | 76,4 | 127 | 116 | 0,118 |
The measurements carried out lead to the following findings
- Undervolting by 100mV brings 2-7% more FPS and 0-10% lower power consumption.
- The Wattman power-saving mode costs 4-15% FPS, but brings 10-20% lower power consumption.
- Undervolting + Overclocking brings 2-9% more FPS and 12-19% lower power consumption.
- Maximum overclocking brings 8-16% more FPS, but leads to 9-27% higher power consumption.
- The “Sweetspot” brings up to 5% more FPS and leads to 22-27% lower power consumption.
Summary and conclusion
Maximum overclocking is also maximally inefficient and for about 25% additional consumption there is hardly over 12% more FPS. In view of the exorbitant consumption of well over 600 W (total system) and the associated gigantic cooling effort, I can only expressly advise against this. The power-saving mode costs a lot of performance, because by reducing the power limit without simultaneously reducing the voltages, the higher P-States are no longer reached under load. As a result, the HBM may not clock up constantly and, in addition to lower FPS, frametime spikes can also occur, which feel very unpleasant.
For tinkerers, it is definitely worth looking for the sweetspot of his card. I suspect it will basically be around 1500 MHz at approx. 1 V voltage, which turned out to be the best compromise between power consumption and performance. Operated in the sweetspot and with optimized memory, the card is at least as fast as with factory settings and needs an average of approx. 25% less energy.
So what’s wrong with the myth of undervolting? Is it really possible to save power and are graphics cards with reduced voltage really faster? I can answer both with “yes”, but I also have to step on the euphoria brake in the same breath. Of course, 25% more performance and 25% lower energy consumption are not possible.
There are two options for undervolting: either you save electricity (same performance as factory settings with lower consumption) or you gain performance (more clock with the same consumption as with the factory settings). If you combine both options and optimize the memory at the same time, you can also increase the performance and at the same time reduce the consumption. However, the increase in performance always to a certain extent with the Increase in power consumption in scale. Undervolting from Vega makes it a much more efficient (or even faster) graphics card – but you can’t make it into the “power-generating 1080 Ti Killer”.
Kommentieren