A lot of PC builders size their power supply like this: add the CPU's TDP to the GPU's TDP, add a bit for "the rest", and buy the nearest wattage. That usually boots. The trouble shows up later, as a random black screen or reboot in the middle of a demanding game, while temperatures look fine and nothing in the logs explains it.
Very often the cause is not average power. It is peak power.
Average draw vs. what the PSU actually sees
A monitoring tool that reports "GPU power: 220 W" is showing you an average over a sampling window, often hundreds of milliseconds. Modern GPUs change their power draw much faster than that. When a heavy frame starts, clocks and voltage jump and the card can pull well above its rated board power for a very short moment, a few milliseconds or less. These are usually called transient spikes or power excursions.
Your software never sees them. Your PSU does. If a spike is large enough to trip the PSU's over-current or over-power protection, the PSU does exactly what it was designed to do: it shuts off. To you, that looks like a crash.
This was widely reported with some high-end cards in the RTX 30 and RX 6000 generations, where systems with PSUs that looked "big enough" on paper shut down under load. It is one of the reasons the newer ATX 3.x PSU standard explicitly requires power supplies to tolerate short excursions above their rated output. An ATX 3.x unit helps, but it does not make a too-small PSU the right size.
The CPU number is also bigger than the box says
On the CPU side, the number on the spec sheet is often not the number that matters:
- Intel lists a Processor Base Power and a higher Maximum Turbo Power. Under a sustained all-core load with default limits, the chip can run near the higher figure.
- AMD desktop chips have a TDP and a higher PPT (Package Power Tracking) limit, which is the actual socket power ceiling.
If you plan with the base or TDP number, you are planning for a CPU that is not running flat out. Games rarely load every core, but shader compilation, loading screens, and background tasks can.
Everything else adds up too
RAM, SSDs, fans, RGB, the motherboard's own VRM losses, and USB devices are small individually. Together they can be 50 to 100 W, more on a full-featured board with lots of fans.
Where the "add ~30%" rule comes from
Once you sum the realistic peak numbers (CPU peak limit + GPU rated board power + the rest), a common rule of thumb is to add roughly 25 to 35% on top and round up to the next standard PSU size. That margin covers:
- Transient GPU spikes that last too briefly to show up in software.
- CPU and GPU peaking at the same time, which the sum assumes but real loads sometimes exceed.
- Ageing and heat, since a PSU's real-world capacity at high ambient temperature is a bit lower than on day one.
- A quieter, cooler PSU: most units are more efficient and quieter in the middle of their load range than near 100%.
It is not about "wasting" watts. A 750 W PSU does not draw 750 W from the wall; it supplies whatever the system asks for. Extra headroom mostly costs a bit more up front.
A worked example
Say you are pairing a Ryzen 7 7700X with an RTX 4070 Super:
| Part | Planning figure |
|---|---|
| Ryzen 7 7700X | 142 W (PPT) |
| RTX 4070 Super | 220 W (board power) |
| RAM, 2 drives, 3 fans, motherboard | ~80 W |
| Total | ~443 W |
Add 30%: about 576 W. Round up: 600 W.
Then do one more check: compare with the GPU maker's own recommended PSU wattage. For this card NVIDIA suggests 650 W, so 650 W is the sensible floor. The vendor figure assumes a fairly power-hungry CPU, which is exactly why it is a good safety net.
If your number lands between two sizes, pick the larger one.
Quick checklist
- Use the CPU's peak limit (Max Turbo Power / PPT), not base power or TDP.
- Use the GPU's rated board power, then add headroom for spikes.
- Add 25 to 35% and round up.
- Never go below the GPU vendor's recommended PSU.
- Prefer a reputable ATX 3.x unit if you are buying new, and use the cables it ships with.
- Remember this is an estimate. Overclocking, silicon quality, room temperature and peripherals all move the real number.
If you do not want to do the math by hand, I put together a free PSU wattage calculator that follows the same steps: it uses peak CPU limits, adds a 30% margin, rounds up to the next 50 W step, and never goes below the GPU maker's guidance. It is a planning estimate, not a lab measurement, so treat the result as the minimum size to shop for.
What PSU are you running, and have you ever chased a shutdown that turned out to be power? I am curious what others have seen.
Top comments (0)