Modern GPUs are running hot, and that's inevitable. They are pushing higher frames than ever, and that's not possible without generating a ton of heat. Now, while advancements in process nodes and architecture mean that they aren't running too hot compared to older generations, there's still a problem, especially with CPUs, SSDs, and RAM picking up as well. With all that heat combined, undervolting emerged as a widely used tweak for anyone running mid- to high-end GPUs from the last few generations. But another tweak that should be just as relevant is power limiting.

The role of undervolting in modern GPUs

Optimizing the efficiency of your card

With modern GPUs sustaining 250–350W (and even beyond) under gaming workloads, undervolting serves an important role. Cards employ dynamic voltage-frequency scaling, so they run only as fast as needed. At the same time, chip-makers tune the voltage to accommodate the broadest possible range of PCs, but there are diminishing returns at the top end of V-F frequency curves. And no matter what, cards still have to operate within a defined thermal and power envelope. In most cases, you can reduce your GPU's temperature and power consumption without sacrificing performance. In other cases (for very power-hungry cards), you can improve performance by keeping the card at the same clock speeds without triggering thermal throttling.

While power consumption is definitely reduced, undervolting does not limit the actual amount of power (TGP for Nvidia and TBP for AMD) your GPU can consume. This means that the maximum possible power draw under a given workload might not drop as dramatically as expected if current or clocks remain high. Also, if the card was already hitting its power limit before significant thermal throttling, lowering voltage alone might not reduce total power enough to change boost behavior and improve performance.

The role of power limiting

The other half of the equation

A power limit is a hard cap on the power your GPU is allowed to draw. It directly controls the energy budget available to the card for voltage, frequency, and boost behavior under load. The principle is much the same as undervolting: the upper frequency limit is achieved at a much higher voltage (and power) cost. A power limit forces the GPU to reduce the voltage/frequency when it hits that limit. Thus, a small power limit (e.g., 80–90%) can reduce power consumption more than undervolting alone, with a minor performance loss (e.g., 5–6%).

A GPU with an unchanged power limit can still draw near its maximum allowed power if the workload and frequency demands are high. Thus, reducing the power limit yields greater reductions in temperature and wattage than undervolting alone, because it directly limits the ceiling on energy consumption, whereas undervolting only indirectly influences it via efficiency improvements. Combining undervolting and power limiting together helps find a sweet spot where there's a minor performance drop for much less power, heat, and noise.

Why you might want to power limit your card

It has a place in many builds

In my opinion, undervolting and power limiting the graphics card is great for virtually any build because it can truly maximize your card's efficiency. But if you don't want to lose any performance, you'll probably have to skip power limiting. In many PC builds, though, it becomes more of a necessity. For example, if...

• You're running a small form factor (or otherwise any airflow-limited) build where every watt that becomes heat inside the case matters,
• You live in a hot climate, and your room's ambient temperature is high,
• You're particularly noise-sensitive, or your GPU's cooling simply isn't good enough,
• Electricity savings are a concern,
• Or undervolting just isn't good or efficient enough with your card (for example, that was sometimes the case with the RTX 4090),

…then you should probably power limit instead (or combine it with undervolting, depending on what works best for your card). The RX 9070 XT, for instance, is a card that benefits from combined undervolting and power limiting more than solely the former.

The optimal configuration isn't universal

Even across two of the same GPUs

The results of undervolting and power limiting vary across cards, even within the same model or variant, due to silicon variation. A good starting point is to look for tutorials or other users' experiments, which you'll find plenty of on YouTube and Reddit. The process itself is pretty simple: you apply a certain percentage of power limit and a negative voltage offset (or adjust the V-F curve on Nvidia cards). If you want the absolute best efficiency, you might spend a few hours doing this until you reach the most stable, high-performance-per-watt setting.

Undervolting is awesome, but the importance of power limiting can't be understated

While undervolting is the go-to tweak for reducing your GPU's heat output, power limiting (either standalone or combined with undervolting) is incredibly effective on most modern GPUs, mid-range and above. Undervolting may not be equally effective on all cards since there's a possibility that your card might still hit its power limit at the reduced V-F budget. This is where power limiting steps in to cap the maximum power the card can consume. Though at the end of the day, what works best for you will depend on your card, the type of build you're running, and other conditions.