PSU Armageddon thanks to RTX 4090? Practical tests with current power supplies from 650 to 1,650 watts
Not only the RTX 3090 causes system crashes due to overload in some scenarios, the comparatively tamer RTX 3080 also has a negative impact with high peak loads. Depending on the overall system and power supply, safety shutdowns occur more frequently with some Ampere cards. Frame limits, undervolting or a modern power supply can help. But has Nvidia learned from it and made things better with the expensive RTX 4090?
Will the RTX 4090 cause more problems with protection circuits?
To anticipate it: Our first impression of the power supply is excellent, Nvidia did a good job after our practical tests. The RTX 4090 runs compared to the RTX 3090 much more stable and merciful in terms of peak loads. And that despite the fact that some manufacturers’ minimum requirements of up to 1,200 watts, which are sometimes far too high, suggest the opposite. Even older power supplies around 700 watts shouldn’t have any problems with the RTX 4090, as long as you keep your system and the total load within reasonable limits – keywords undervolting and efficiency tuning. The feared need to upgrade the power supply is by no means given. Future ATX 3.0 power supplies are far from being necessary with current GPUs. In our random test with different watt classes and manufacturers, we let the RTX 3090 compete against the RTX 4090 in different scenarios.
Power supplies: RTX 3090 vs. RTX 4090 in practice
Our first practical test with the RTX 4090 and various power supplies takes on the watt classes from 650 to 1,000 watts, each with different manufacturers. The Multi-Rail Straight Power 11 Platinum from Be Quiet competes in the 650 watt class, NZXT with a Seasonic foundation is a candidate for 750 watts, and Asus with the older Thor 850-P1 for 850 watts. In the kilowatt range, we used the Corsair RM1000x as an example. Other PSUs tested included the Thermaltake GF3 1650 Watt and Seasonic’s Prime TX-1000, both of which we ran with the native 12VHPWR cable – the results are virtually identical to the Corsair RM1000x. Each scenario listed separately in the table was carried out with a running time of approx. 15 to 30 minutes.
Please note that the selected power supplies were only randomly sent to the power course in the same open test system with the graphics cards. Depending on which system you are working with, the results may differ. The trend with the RTX 4090 and our experience with the now older ampere cards such as the RTX 3080/3090 match the results. In particular, the first ampere cards bring power supply units up to the 850 watt range, depending on the suboptimal load, for safety shutdown. We cannot reproduce the same behavior with the first Ada Lovelace GPU. The peak loads are successfully captured.
Unlimited i7-12700K scenario | Be Quiet Straight Power 650W Platinum | NZXT C750 Gold | ROG Thor 850W Platinum | Corsair RM1000x |
---|---|---|---|---|
RTX 3090 – Stock | ||||
Furmark + Cinebench R23 | Stable (total approx. 690 watts) | Stable (total approx. 700 watts) | Stable (total approx. 690 watts) | Stable (total approx. 690 watts) |
Cyberpunk 2077 with sudden changes | Stable | Stable | Stable | Stable |
PUBG: Main Menu Tab Switch | Unstable | Stable | Stable | Stable |
RTX 3090 – OC with +60 core MHz, +23% power limit | ||||
Furmark + Cinebench R23 | Stable (total approx. 780 watts) | Stable (total approx. 810 watts) | Stable (total approx. 790 watts) | Stable (total approx. 790 watts) |
Cyberpunk 2077 with sudden changes | Unstable | Stable | Stable | Stable |
PUBG: Main Menu Tab Switch | Unstable | Unstable | Stable | Stable |
RTX 4090 – Stock | ||||
Furmark + Cinebench R23 | Stable (total approx. 770 watts) | Stable (total approx. 800 watts) | Stable (total approx. 780 watts) | Stable (total approx. 770 watts) |
Cyberpunk 2077 with sudden changes | Stable | Stable | Stable (total approx. 520 watts) | Stable |
PUBG: Main Menu Tab Switch | Stable | Stable | Stable | Stable |
RTX 4090 – OC with +125 core MHz, +33% power limit | ||||
Furmark + Cinebench R23 | Unstable | Stable (total approx. 910 watts) | Stable (total approx. 910 watts) | Stable (total approx. 890 watts) |
Cyberpunk 2077 with sudden changes | Unstable | Stable | Stable | Stable |
PUBG: Main Menu Tab Switch | Unstable | Stable | Stable | Stable |
RTX 4090 – Undervolted at 2,600MHz at 875mV | ||||
Furmark + Cinebench R23 | Stable (total approx. 680 watts) | Stable (total approx. 680 watts) | Stable (total approx. 690 watts) | Stable (total approx. 680 watts) |
Cyberpunk 2077 with sudden changes | Stable | Stable | Stable (total approx. 400 watts) | Stable |
PUBG: Main Menu Tab Switch | Stable | Stable | Stable | Stable |
System: Intel Core i7-12700K (no power limit, approx. 200 watts under Cinebench), Asus ROG Strix B660-i Gaming WiFi, 2x 16 GiB DDR5-5600 CL36 Corsair Vengeance, Alphacool Eisbaer LT 240. Geforce RTX 3090: Asus ROG Strix gaming. Geforce RTX 4090: Asus TUF Gaming (see table below) and Palit Game Rock.
Remark: Unstable here means the safety shutdown of the power supply unit, the computer suddenly switches off and then starts up again. Stable means that there was no crash (due to load) within the time frame of 15 to 30 minutes. In games, active work was explicitly done with different resolutions and sudden interactions. With undervolting of the RTX 4090, our overall system (consumption at the socket) remained at a total of 400 watts even under Cyberpunk 2077 and Metro Exodus Enhanced Edition (GPU power draw: approx. 230 watts at 4K with DLSS).
[PLUS] PSU comparison test: 12 models with 1,000 watts
Even if even the Be Quiet Straight Power 11 Platinum with 650 watts ran without any problems in the standard setting of the RTX 4090 TUF Gaming and the Palit Game Rock, we explicitly recommend a higher watt class or at least undervolting of the all-round impressive card. With manual overclocking and further raising the power limit, the consumption increases up to 550 (in practice the absolute maximum) or theoretically approved 600 watts, but here too the spikes are less aggressive compared to the RTX 3090 and therefore less aggressive in practice especially merciful. Depending on the housing, permanent overloading of your power supply unit can not only lead to overheating (and thus renewed protective shutdown), but also to higher residual ripple, lower efficiency and other negative effects. The applied practical load should therefore not exceed the rated power of the power supply unit, even if the computer is running without problems. Wear and tear effects on your hardware are certain with a load of around 800 watts from a 650 watt power supply unit, even if this consumption should only occur in exceptional cases (such as an unrestricted thread ripper with a heavy GPU load).
Nvidia’s target of 850 watts for the RTX 4090 is a valid and reasonable suggestion, which you can now easily follow as well. Buying a 1,000 watt power supply is by no means pointless, but offers other advantages, such as buffers for further hardware upgrades, higher efficiency in certain load ranges and much more. General information about this and which power supply would be right for you – including tests with comparisons and tips – can be found in our new Power supply purchase advice. In the context of increasing requirements for CPUs and GPUs, we tested in ours PCGH issue 11/2022 twelve 1,000 watt power supplies put through their paces from a technical point of view, including questions from manufacturers about the 12VHPWR cable. On the second page we go through a few more points, show our stability tests and draw a first conclusion.
Reference-www.pcgameshardware.de