After seeing a 750W PSU coupled with a Core i5 and GTX 960 for the thousandth time, inspiration struck to compile one of our most ambitious benchmarks to-date. This analysis compares watt consumption across various GPUs, CPUs, and complete system configurations, resulting in a loose template answering the question of “how many watts do I need?”
It all feeds into one of the most common PC building mistakes: going overkill on power supplies, often buying larger PSUs for sake of certainty or under the pretense of “room to upgrade.” This is a fine pretense, but is often done to the extreme. The fact of the matter is that most mid-range gaming PC builds can run on 450-600W PSUs, depending on the GPU, with a good deal of them landing ideal wattage around the 500-550W range. Buying a power supply that more closely fits the usage curve of a system will improve power efficiency, reduce build cost, reduce cost-to-run, and allow builders to buy PSUs that put the cost toward more relevant features than just wattage – like efficiency, protections, PFC, and so forth. Think of this as redistributing the cost of purchase; it's not always that simple, but we'd generally rather have increased efficiency ratings and power protections than more watts. It all depends on the build, of course, and that's what we're dissecting here.
Let's first talk power supply basics: How PSUs rails are divided, voltage ripple, how many watts are required, and power efficiency, then we'll dive into individual component power consumption benchmarks. We've tested the majority of the current nVidia and AMD GPU lineups for power consumption, the AMD & Intel CPU watt draw, and templated system power consumption. Our system templates were built in a fashion that should fall within range of reasonable configurations for “real” builders, and will help in determining how many watts you need for common go-to builds.
For this article, we deployed a high-end, 89-94% efficient Enermax Platimax PSU with 80 Plus Platinum certification and 1350W of power. In a bout of irony, the PSU is, in fact, overkill for all of the hardware tested here. “Do as I say, not as I do” comes to mind – we wanted the overhead to ensure none of our tests bottlenecked on power availability, at time of start being unsure what hardware we'd end up testing.
Disclosure: The test PSU was supplied by Enermax, who sponsored our time-intensive research process and content. 100% of research, analysis, and content creation was conducted independently by GamersNexus.
PSU Spec Samples
First, a couple of sample PSUs. This is what we typically see when looking at PSU specifications on a retail website, like Amazon or Newegg.
|Model & Wattage||Modular?||Efficiency & Features||Price|
|Enermax Platimax 1350W||Fully||80 Plus Platinum, 89-94% Efficiency
HeatGuard continues fan operation after shut-down to ensure GPU/CPU are not left hot.
Multiple 12v Rails
|Corsair RMX 850W||Fully||80 Plus Gold, 87-90% Efficiency
Zero RPM Mode
Single 12v Rail
|EVGA SuperNOVA GS 650W||Fully||80 Plus Gold, 87-90% Efficiency
Fans Off Until >25% Load (162.5W).
Single 12v Rail
|XFX TS 550W||No||80 Plus Bronze, 82-85% Efficiency
Single 12v Rail
|Corsair CSM 450W||Semi-Modular||80 Plus Gold, 87-90% Efficiency
Single 12v Rail
There's a lot more to it than that. Hundreds of power supplies exist, most from the same group of OEMs (similar to CLC OEMs), and they fill all manner of mainstream and niche demands. Silence, thermals, modularity, form factor (SFF not featured above, but important), efficiency, wattage, endurance, cap and electrical design quality, and warranty are a few more items to take into consideration. For sake of simplicity, the above list provides a look at a wide range of PSUs we think would be commonly found in gaming desktops. This list isn't definitive and doesn't represent the “best” power supply in each price range, but provides a launching point with a wide selection of quality PSUs.
The goal is really just to throw some specs in front of everyone – priming for the content. Those listed, let's move into other pre-benchmark topics.
Factors in Power Supply Selection
The scope of this article is to outline wattage requirements for various system configurations, concluding on a finalized list of “template” PC builds (read: builds we think would realistically be constructed). The greater topic of PSU selection – that is, regarding topics beyond wattage – is out-of-scope for this content, but it's worth going over a few basics. This should hopefully get the conversation of PSU selection started, hopefully spinning-off a few research ideas for PSU shoppers.
A couple major items to consider, in no particular order:
- Cable management / modularity
- Cable options – do you have all the PCI-e headers you need?
- Power efficiency & efficiency curve
- Cooling & fan noise
- Voltage ripple, PFC, & clean voltage / electrical design
- Form factor
- Voltage (ensure PSU is either switching or uses a native voltage compatible with your country)
- Stability (overclocking, overvolting, and components with exceptional power supply demand will best couple with higher-end PSUs)
- Price, of course
Today, we're focusing on that first one.
Each component in the system, no matter how small, requires some amount of power. The main power connections in a system operate on rails of three voltages – 3.3v, 5v, and 12v – and are used to drive devices of differing power requirements. Some PSUs run multiple 12v rails, something we talk about here. The motherboard's 24-pin header consists of 3.3v, 5v, and 12v lines, but also uses a few signaling pins to communicate with the board and PSU, namely the PWR_OK and PS_ON pins (gray and green, respectively). EPS-12V 4/8-pin headers (or more, on some overclocking boards) provide 12v power to the CPU, PCIe 6/6+2-pin headers power the GPU (all +12v, GND, and sensors), MOLEX can push 12v or 5v (though fan controllers will use resistors to modulate voltage to the fan), and so forth.
The motherboard takes power, the CPU takes power, the GPU takes power – all of the components require some wattage. Searching for “[product] TDP” will often find the thermal design power of the device, which – depending on the manufacturer and component – will tell us how much power is required to adequately cool a product. This is not necessarily a power supply / wattage requirement indicator, but provides a starting point.
Some components are powered directly by the PSU, some receive power through the board, and some are a split of the two. GPUs, for instance, can draw about 75W of power through the PCI-e slot (see: GTX 750 Ti with 0 power headers on-card), but will also drink from the 6/8-pin cables. M.2 SSDs use PCI-e lanes for data and power, eliminating the requirement of SATA power. The CPU is powered directly through the socket, which sources its power from the 8-pin EPS-12V connector (sometimes 4-pin).
Modern CPUs and GPUs use dynamic voltage to mitigate power draw during idle or low-load times, as seen in Intel's S0iX (“Active Idle”), which throttles clockrate in-step with voltage reduction – sort of a microsleep. NVidia GPUs have received cooling solutions (MSI, ASUS, EVGA in particular) that spin-down fans when under less than ~30W load, a threshold below which passive cooling is adequate for the GPU's thermals. This is another small power reduction, though it's primarily a decision made in favor of noise levels.
Most GPUs and some CPUs will recommend a minimum wattage for operation. Depending on the device, this sometimes makes assumptions of other system components – if it's a higher-end GPU, the minimum PSU spec may be assuming higher-end accompanying components, which will draw more power. It's ultimately the objective of these specs to provide a one-size-fits-all operating range, where almost no one will be under-powering components. This can also mean that the wattage suggestions err a little on the high side for safety.
But that's OK, because modern power supplies will draw only as much as required to drive the system. It's not ideal to buy too high on wattage – we sink on the efficiency curve – but it's not going to actively damage anything by purchasing a 1300W PSU and using it with a 500W draw build. That'd mostly be an awkward allocation of resources.
A note on “Continuous Power” product labeling, from our PSU dictionary (entry authored by Michael Kerns, ed. Patrick Stone):
“Continuous power is generally referred to as the Maximum Power (Watts) a PSU can output safely continuously. Some power supplies may be able to output 800 Watts safely for a moment, but can only provide 600 Watts of continuous power. Most of the time the labeled wattage is continuous power.”
And a note on “Maximum Power” ratings for PSUs, also from our dictionary (entry authored by Patrick Stone, add. ed. Michael Kerns):
“Simply enough, the maximum power listed for a PSU is its maximum combined output wattage that the device is able to spit from all rails in ideal circumstances. Efficiency comes into play here, but we'll talk about that in more in the 80 Plus definition. Many users grossly over-estimate the required size of a gaming power supply, so we always suggest running your components through a power supply calculator first (though they also over-estimate, in general).
The formula for wattage is voltage*amperage=wattage. […] At the low-end [of the efficiency curve], a disproportionate amount of power is consumed in the transformation; at the high-end, heat generation eats up the efficiency.”
Power Efficiency & Efficiency Curve
Before diving into this one, you can read about 80 Plus ratings in our PSU dictionary. Here's a table of 80 Plus Certified ratings:
|Efficiency at Rated Load: 115v (230v)|
|80 Plus Certification||10%||20%||50%||100%|
|80 Plus||--||80% (82%)||80% (85%)||80% (82%)|
|80 Plus Bronze||--||82% (85%)||85% (88%)||82% (85%)|
|80 Plus Silver||--||85% (87%)||88% (90%)||85% (87%)|
|80 Plus Gold||--||87% (90%)||90% (92%)||87% (89%)|
|80 Plus Platinum||--||90% (92%)||92% (94%)||89% (90%)|
|80 Plus Titanium||90% (90%)||92% (94%)||94% (96%)||90% (94%)|
Power efficiency curves illustrate a power supply's performance and scalability against load from the system. Manufacturers who decide to publish some sort of efficiency curve illustration will show a percentage on the X-axis (often 20-100%, stemming from 80 Plus certification requirements) that represents % of available wattage loaded; the Y-axis will show still more percentages (efficiency) in the form of a plotted line. Efficiency is generally lowest toward the far ends of the graphs (again, often 20% and 100%), with peak efficiency resting around 50% load. We're going to break one of our rules and generalize for a second: For most consumers and gamers, there is negligible efficiency loss between, say, 50% and 80% PSU load. With an 80 Plus Gold certified PSU, that means a hit from ~90% efficiency (115v, US outlets) to ~88% efficiency. In some sort of enterprise or serious production environment, a 2% loss of power efficiency means money; we're talking thousands, tens of thousands, or more in additional annual power expense, depending on how large a company it is and how much power they consume. For a consumer, it could be literal pennies and dollars difference. For example:
- 500W demand / 0.9 efficiency = 550.56W draw
- 500W demand / 0.88 efficiency = 568.2W draw
- Difference: 18W
- Assume 24-hour uptime & assume 11.63 cents per kWh (NC, 2015)
- Cost difference: 5c per day, or $18.25 per year (24/7/365).
But this example is extreme and imperfect. 500W demand? That's not going to happen at idle, which is most likely what a 24/7 uptime is primarily comprised of for consumers. Let's look at something more realistic:
- 70W demand / 0.9 efficiency = 77.78W draw
- 70W demand / 0.88 efficiency = 79.55W draw
- Difference: 1.77W
- Assumed 24-hour uptime & assume 11.63 cents per kWh (NC, 2015)
- Cost difference: 0.494c per day ($0.00494), or $1.8 per year (24/7/365).
Again – not perfect. Now we're assuming 24-hour idle, which also is unrealistic. The difference falls somewhere in that range, pursuant to use case and number of hours dedicated to higher load generation.
Depending on use case and user, the difference between a more effectively loaded (and efficiently designed) PSU may make itself up within a few years of purchase. Or it might not, if you're active in using sleep states or shutting down and don't spend considerably uptime under load.
Efficiency loss is often through heat (this is why cooler PSUs, all other specs identical, will also perform better). Each manufacturer and PSU supplier will have its own efficiency curve that fits the specs of the power supply, but a good, general assumption is that 50-60% load (demand for power) on the PSU will warrant its peak performance (efficiency). This doesn't mean system builds all have to run 50% load levels for peak efficiency – sometimes there's better cost/benefit and value in opting toward higher load levels (60, 70, 80 percent) in favor of a less expensive PSU. We'll talk about that more below.
Saturating 95% of a PSU's total watt allowance makes it work harder, run hotter, and ultimately take a dive on the efficiency curve. Power supplies have the job of converting mains electricity into something that the PC can understand and utilize. The power from the wall is AC (alternating current), which feeds into the PSU through its power cable and gets converted to DC (direct current). In effect, everything going into the “back” of the PSU is AC, and everything coming out the front (where the PC's power cables connect) is DC, split into various voltages as defined above. This conversion process generates heat, contributing to the efficiency of the PSU's actual wattage output. Connected to a wall meter and under identical conditions, a PSU with no 80 Plus certification (poor efficiency) will draw more power than a PSU with, for instance, 80 Plus Platinum efficiency, as found in the Enermax Platimax PSU we used to test.
Here's a sample power efficiency curve:
In this curve – which exaggerates the delta by having only an 8-point range – we see 87% efficiency in worst case scenarios (115v) and 90% efficiency in best-case scenarios.
Voltage ripple is a measurement of voltage fluctuation down the PSU's various rails, something we wrote about in-depth previously. The voltage supply down the 12v rail, for instance, won't read-out as exactly 12v when considering the PSU's switching behavior. Depending on the quality of the PSU, voltage can make small ripples – from 12.10v to 12.14v, for instance – that are measurable on an oscilloscope, but don't really impact daily operations of the PC. Lower quality power supplies can exhibit greater voltage ripple and can begin impacting stability of the system, particularly as it relates to overclocks that stress power management and delivery systems.
(Above: O-Scope screenshots sourced from Corsair blog)
Here's an excerpt from our previous voltage ripple article, written by GN's Michael Kerns:
“Taking the example of the voltage going back-and-forth from 12.10 to 12.14, the power supply would be described as having a 40mV ripple. The picture below is a photo of an oscilloscope measuring voltage ripple. A good power supply will keep voltage under control with self-regulation.
While excessive voltage ripple is bad, some amount is perfectly normal and okay. All ATX power supplies have some amount of voltage ripple simply due to their design. For this reason, the ATX specification has limits for voltage ripple. It varies from rail to rail, resultant of voltage differences on each rail. The ATX specification requires that the 12v have a maximum ripple of 120mV, and 50mV for the 3.3v and 5v rails. While this is the maximum voltage ripple allowed, many -- myself included -- have their own standards. I generally like to see around or below 60mV for the 12v rail, and 35mV for the 3.3v and 5v. This drives-up cost, but allows for greater OC stability.
Capacitors and other components are meant to withstand some voltage ripple, as it is a necessary evil for modern day PSUs.”
All right. Let's talk tests. Continue to Page 2 for the tests executed, methodology, and parts lists.