Well, I have a new GPU and as usual i like to undervolt my things. So today marks the first daay of my tests and I would like to share them with you.
Regarding my test/verification setup I ran Port Royale with custom settings and Cyberpunk 2077 three times each and then compared the results:
Port Royale | Cyberpunk 2077 | ||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|
Score | Temp | Power | FPS | Temp | Power | ||||||
Avg | Max | Min | Max | Avg | Avg | Max | |||||
Default | Run 1 | 12140 | 67,2 | 452 | 472 | 49,48 | 93 | 74,24 | 63,2 | 396,2 | 410,711 |
Run 2 | 12164 | 67,4 | 452 | 472 | 52,79 | 107,92 | 74,02 | 63,2 | 398,5 | 410,711 | |
Run 3 | 12183 | 67 | 451,55 | 475,8 | 35,48 | 107,81 | 73,94 | 65,2 | 399,8 | 416,237 | |
Avg | 12162,33333 | 67,2 | 451,85 | 473,2666667 | 45,91666667 | 102,91 | 74,06666667 | 63,86666667 | 398,1666667 | 412,553 | |
vs. 985mV | 1,0075 | 1,0184 | 1,1215 | 1,1062 | 0,8949 | 1,0355 | 1,0093 | 1,0127 | 1,1115 | 1,1127 | |
vs. PT 70 | - | - | - | - | 1,1387 | 1,0562 | 1,0339 | 1,0385 | 1,1963 | 1,2344 | |
vs. PT 65 | - | - | - | - | 0,886 | 1,0608 | 1,0583 | 1,001 | 1,2837 | 1,3266 | |
vs. 60 FPS | - | - | - | - | 0,7887 | - | - | 1,0219 | 1,2619 | 1.1676 | |
Score | Temp | Power | FPS | Power | |||||||
Avg | Max | Min | Max | Avg | Max | Avg | Max | ||||
2800Mhz @975mV | Run 1 | 12048 | 65,4 | - | 416,117 | - | - | - | - | - | - |
Score | Temp | Power | FPS | Temp | Power | ||||||
Avg | Max | Min | Max | Avg | Avg | Max | |||||
2800Mhz @985mV | Run 1 | 12092 | 66,2 | 406,15 | 435,056 | 48,87 | 111,12 | 73,51 | 63,2 | 360 | 371,927 |
Run 2 | ?? | 66 | 403 | 429,71 | 58,1 | 97,81 | 73,19 | 63,3 | 358,4 | 372,785 | |
Run 3 | 12085 | 66,1 | 400,9 | 426,024 | 49,87 | 96,25 | 73,14 | 60,2 | 352,3 | 366,875 | |
Run 4 | 12045 | 65,5 | 400,75 | 423,865 | 55,53 | 99,63 | 73,52 | 63 | 359,8 | 370 | |
Run 5 | 12074 | 66 | 403,15 | 421,142 | 59,42 | 99,6 | 73,93 | 64,5 | 359,5 | 371,445 | |
Run 6 | 12065 | 66,1 | 403,45 | 431,142 | 36,07 | 91,91 | 73,02 | 64,2 | 359,3 | 371,84 | |
Avg | 12072,2 | 65,98333333 | 402,9 | 427,8231667 | 51,31 | 99,38666667 | 73,385 | 63,06666667 | 358,2166667 | 370,812 | |
vs. Default | 0,99% | 0,9818 | 0,9817 | 0,904 | 1,1175 | 0,9658 | 0,9908 | 0,9875 | 0,8997 | 0,8988 | |
vs. PT 70 | - | - | - | - | 1,2725 | 1,02 | 1,0244 | 1,0255 | 1,0763 | 1,1095 | |
vs. PT 65 | - | - | - | - | 0,993 | 1,0245 | 1,0486 | 0,9885 | 1,1549 | 1,1924 | |
vs. 60 FPS | - | - | - | - | 0,8813 | - | - | 1,009 | 1,1353 | 1,0495 | |
Score | Temp | Power | FPS | Temp | Power | ||||||
Avg | Max | Min | Max | Avg | Avg | Max | |||||
2800Mhz @985mV PT 70 | Run 1 | - | - | - | - | 34,57 | 105,94 | 71,84 | 332,5 | 334 | |
Run 2 | - | - | - | - | 40,21 | 95,56 | 71,31 | 64 | 332,8 | 334,432 | |
Run 3 | - | - | - | - | 46,19 | 90,79 | 71,76 | 59 | 333,2 | 334,195 | |
Avg | #DIV/0! | #DIV/0! | #DIV/0! | #DIV/0! | 40,32333333 | 97,43 | 71,63666667 | 61,5 | 332,8333333 | 334,209 | |
vs. Default | - | - | - | - | 0,8782 | 0,9467 | 0,9672 | 0,963 | 0,8359 | 0,81 | |
vs. 985mV | - | - | - | - | 0,7859 | 0,9803 | 0,9762 | 0,9752 | 0,9291 | 0,9013 | |
vs. PT 65 | - | - | - | - | 0,7804 | 1,0043 | 1,0236 | 0,9639 | 1,073 | 1,075 | |
vs. 60 FPS | - | - | - | - | 0,6926 | - | - | 0,984 | 1,0548 | 0,9459 | |
Score | Temp | Power | FPS | Temp | Power | ||||||
Avg | Max | Min | Max | Avg | Avg | Max | |||||
2800Mhz @985mV PT 65 | Run 1 | - | - | - | - | 48,23 | 99,05 | 70,04 | 63,8 | 309,2 | 310,97 |
Run 2 | - | - | - | - | 52,88 | 96,41 | 70,23 | 63,8 | 311,1 | 310,993 | |
Run 3 | - | - | - | - | 53,91 | 95,57 | 69,69 | 63,8 | 310,2 | 310,98 | |
Avg | #DIV/0! | #DIV/0! | #DIV/0! | #DIV/0! | 51,67333333 | 97,01 | 69,98666667 | 63,8 | 310,1666667 | 310,981 | |
vs. Default | - | - | - | - | 1,1254 | 0,9423 | 0,9449 | 0,999 | 0,779 | 0,7538 | |
vs. 985mV | - | - | - | - | 1,007 | 0,976 | 0,9537 | 1,012 | 0,8659 | 0,8386 | |
vs. PT 70 | - | - | - | - | 1,2815 | 0,9957 | 0,977 | 1,037 | 0,9319 | 0,9305 | |
vs. 60 FPS | - | - | - | - | 0,8876 | - | - | 1,0208 | 0,983 | 0,8801 | |
Score | Temp | Power | FPS | Temp | Power | ||||||
Avg | Max | Min | Max | Avg | Avg | Max | |||||
2800Mhz @985mV 60FPS | Run 1 | - | - | - | - | 58,16 | 61,98 | 60 | 62,5 | 319,3 | 353,339 |
Run 2 | - | - | - | - | 58,55 | 61,48 | 60 | 62,5 | 312,7 | 353,339 | |
Run 3 | - | - | - | - | 57,95 | 62,22 | 60 | 62,5 | 314,6 | 353,339 | |
Avg | #DIV/0! | #DIV/0! | #DIV/0! | #DIV/0! | 58,22 | 61,89333333 | 60 | 62,5 | 315,5333333 | 353,339 | |
vs. Default | - | - | - | - | 1,2679 | - | - | 0,9786 | 0,7925 | 0,8567 | |
vs. 985mV | - | - | - | - | 1,1347 | - | - | 0,991 | 0,8808 | 0,9529 | |
vs. PT 70 | - | - | - | - | 1,4438 | - | - | 1,0162 | 0,948 | 1,0572 | |
vs. PT 65 | - | - | - | - | 1,1267 | - | - | 0,9796 | 1,0173 | 1,1362 |
The odd thing at 985mV is that the second test did not have any score so it might have failed which is an indication of an unstable undervolt.
I’m also wondering if I’m setting my curve the wrong way, because I’m trying a new approach. Instead of the classic one where one raises the clock at a given voltage I move the whole curve and lower everything after a specific voltage. The difference is that the former undervolts every clock setting, while the latter ramps up more towards the end. Maybe, the latter is preferable since power consumption and heat generation at lower clocks is negligible. Do you have any ideas/input?
The new approach undervolts every clock setting by the same amount. E.g. given an arbitrary but fixed voltage the clockspeed is bumped by n
MHz.
GPU: MSI RTX 4090 Suprim X
CPU: 5900X
Mainboard: GIgabyte X570 Aorus Master
Edit (27.10.2022): I have updated the table and 985mV seem fine. However, Run 2
still bugs me and I also experienced a system freeze yesterday. This could have been caused by something else, but I’m still wary. I’m thinking of bumping the voltage by 5mV and test everything again. I assume the power consumption should roughly be the same but I should gain a bit stability just in case. Also, please note that somewhere between Run 4
- Run 6
Windows decided to install updates. I assume it was during the last Cyberpunk 2077 run (hence the low min fps) but I’m not sure.
Edit (30.10.2022): I have updated the table with additional information and came to the following conclusion: Initially, I wanted to test 2800MHz @990mV just to see the difference, but I was unable to set the clockspeed to 2800MHz. After testing with 985mV again I noticed that my GPU was also not able to reach 2800MHz, it was rather stuck between 2790Mhz or 2820MHz. While this difference would in practice be negligible it naturally has an effect on performance and power consumption.
Given that 985mV does seem to be stable is settled for this value for now and proceeded with testing Power Target limiting as suggested by der8auer and Optimum Tech. While it does further increase the power consumption of the GPU at a PT of 60% I noticed some peculiar behaviour which is why I aborted my tests and tested 70% and 65% instead. All values can be seen in the chart above.
However, I did come to a different conlusion as both YouTubers and I would not recommend reducing the PT of a 4090. Much rather, I would recommend limiting the FPS to something appropriate. For example, when capped at 60fps a 4090 uses 79% of the average power compared to default settings and is almost equal to PT 65% setting. Contrary to PT 65%, the frametime graph is almost a perfect line and much more consistent with lows being 13% higher. (It should be noted though, that peak power consumption is 14% higher than PT 65%, but still 86% of stock settings.)
Thus, I would recommend to undervolt a 4090 as much as possible and then proceed to limit the framerate to something appropriate.