- Joined
- Mar 7, 2008
I got a 5070 Ti at launch, and today put it through some testing with both overclocking and power limiting. The model is the MSI Ventus. In MSI afterburner it has 83-100% power limit, mem oc maxes out at +2000, and core has more adjustment range than I need.
For the following I tested at stock, with the lowest 83% power limit which is 250W, stock 300W. And the overclock I used was +2000 mem, +500 core. I did try +525 core but that crashed as soon as a benchmark ran. Driver 572.70. Windows 11. 4k resolution.
3DMark Steel Nomad - DX12: 6169 stock, 6930 OC (+12%), 6068 PL (-2%).
Monster Hunter Wilds Benchmark - Ultra, DLSS Quality, FG off: 70.7 stock, 80.2 OC (+13%). Didn't run PL as it never went over 250W so would be same as stock.
Black Myth Wukong Benchmark - Very High, DLSS 50%, FG and motion blur off: 70 stock, 78 OC (+13%). Didn't run PL as it never went over 250W so would be same as stock.
Final Fantasy XIV Dawntrail v1.1 Benchmark - Max: 89.9 stock, 100.5 OC (+12%), 89.4 PL (-1%).
Shadow of the Tomb Raider - Highest: 123 stock, 140 OC (+14%), 122 PL (-1%).
Watch Dogs Legion - Ultra, DX12: 78 stock, 87 OC (+12%), 77 PL (-1%).
Blender 4.3.0 - overall score: 7121 stock, 7951 OC (+12%). Didn't run PL as it never went over 250W so would be same as stock. Subscores scaled similarly.
If I wasn't so lazy I'd make a nice little chart!
This was quite consistent, everything was 11 to 13% faster with applying the overclock. I did find it interesting that several of these didn't even cross the 250W mark at stock, so they wouldn't be affected at minimum power limit. For those that did cross 250W when running, they were hardly impacted in performance typically 1% slower.
Typical core clocks were just under 2500, which is right were the nominal boost clock of 2482 sits. With the +500 core overclock, it was just under 3000. I wasn't seeing that limiting earlier for power. Going from 2500 to 3000 is +12%, which lines up with the OC benchmark results. VRAM clock went from 1750 to 2000 which is +14%. So that also scaled with the change in core and balance is maintained. TechPowerUp claimed to be able to set +3000 but I wonder how? That could help show if there is any further scaling from VRAM. It seems unlikely, as I did run stock core and +2000 mem on Steel Nomad for about +0.5% difference.
For the following I tested at stock, with the lowest 83% power limit which is 250W, stock 300W. And the overclock I used was +2000 mem, +500 core. I did try +525 core but that crashed as soon as a benchmark ran. Driver 572.70. Windows 11. 4k resolution.
3DMark Steel Nomad - DX12: 6169 stock, 6930 OC (+12%), 6068 PL (-2%).
Monster Hunter Wilds Benchmark - Ultra, DLSS Quality, FG off: 70.7 stock, 80.2 OC (+13%). Didn't run PL as it never went over 250W so would be same as stock.
Black Myth Wukong Benchmark - Very High, DLSS 50%, FG and motion blur off: 70 stock, 78 OC (+13%). Didn't run PL as it never went over 250W so would be same as stock.
Final Fantasy XIV Dawntrail v1.1 Benchmark - Max: 89.9 stock, 100.5 OC (+12%), 89.4 PL (-1%).
Shadow of the Tomb Raider - Highest: 123 stock, 140 OC (+14%), 122 PL (-1%).
Watch Dogs Legion - Ultra, DX12: 78 stock, 87 OC (+12%), 77 PL (-1%).
Blender 4.3.0 - overall score: 7121 stock, 7951 OC (+12%). Didn't run PL as it never went over 250W so would be same as stock. Subscores scaled similarly.
If I wasn't so lazy I'd make a nice little chart!
This was quite consistent, everything was 11 to 13% faster with applying the overclock. I did find it interesting that several of these didn't even cross the 250W mark at stock, so they wouldn't be affected at minimum power limit. For those that did cross 250W when running, they were hardly impacted in performance typically 1% slower.
Typical core clocks were just under 2500, which is right were the nominal boost clock of 2482 sits. With the +500 core overclock, it was just under 3000. I wasn't seeing that limiting earlier for power. Going from 2500 to 3000 is +12%, which lines up with the OC benchmark results. VRAM clock went from 1750 to 2000 which is +14%. So that also scaled with the change in core and balance is maintained. TechPowerUp claimed to be able to set +3000 but I wonder how? That could help show if there is any further scaling from VRAM. It seems unlikely, as I did run stock core and +2000 mem on Steel Nomad for about +0.5% difference.