How Many Watts Does a Gaming PC Really Need? Exhaustive PSU Usage Benchmarks

By Published November 30, 2015 at 10:13 pm
  •  

 

How Many Watts Do You Need for High-End Graphics Cards? (& OC Power Draw)

The above chart includes most of the cards we've worked with in the past year. Individual GPU reviews we've posted have contained various states of this chart, but the above is the compiled listing. Power draw shown above is total system power when under peak “gaming” load, defined in the methodology, and is not the power draw of only the GPU.

Video card power draw depends first on the GPU provided by AMD and nVidia, then depends on the design of the card. Some boards use non-reference designs that are capable of driving significantly more power than the reference cards, often for overclocking, and will thus front a higher power consumption than what the GPU manufacturer originally stated.

The charts reveal that, on an i7-4790K and with the other specs defined above, all tested single-card configurations (pre-OC) fall below 450W total power draw. MSI's pre-overclocked R9 390X is one of the most power-hungry cards, drawing 416.3W out-of-box and 456.8W after our own overclocks. Assuming a target PSU consumption of 60% load, a comparable load to the 456.8W overclocked test would ideally use something in the range of 750W (450W / 0.60 = 750W, so 60% load of 750W is 450W). Pushing closer to 80% PSU load, which is perfectly reasonable even if technically less efficient, the PSU wattage would ideally be around ~600W (563W, but not much exists between 550W & 600W).

watt-draw-bench-gpus

It isn't until we get to multi-GPU configurations that power consumption nears and exceeds 600W, with 2x R9 Fury X cards approaching 700W. This is where 1000W+ PSUs start to see actual use and saturation. The SLI and CrossFire configurations tested pull approximately 50% more power than their single-card counterparts.

At the low-end, a GTX 960 would get by – assuming a comparable CPU coupling, like a Core i5, and 50% load target – with PSUs as low as ~400-450W. This depends on other components (450W to be safely general, we'll get more granular below) and is assuming a 50% consumption target on the PSU.

How Many Watts Do You Need for CPUs?

watt-draw-bench-cpus

CPU power draw is a direct correlation to vCore and frequency, with advanced power saving settings as a means to mitigate consumption during low-load periods.

At the far-end of the chart is AMD's FX-9590 CPU, a 220W TDP component that requires high-end cooling and more complex power management on the motherboard. Under the load generated from our HandBrake encode pass, the complete FX-9590 system pulls 382.4W from the wall, a disparity of 61.25% over Intel's i7-4790K CPU. Running an i7-4790K draws considerably more power than the same-generation (Devil's Canyon) i5-4690K, a gap of 23.74%. Idle performance of the two is effectively identical.

Running lower-end stuff, like the i3-4160, we're seeing just under 120W total system power draw (100% CPU load), not a bad increase over the fairly hot G3258 (109W load). Skylake's i5-6600K SKU pushes 124W load (70W idle), with the i7-6700 non-K SKU at 132W (71W idle).

We talk a lot about load times on this site – particularly this article – because it's the most exciting to look at for gamers. We want to know what is required of a PSU to drive a gaming machine when it's actually playing games. Truthfully, though, idle power draw is almost more important (depending on the deltas between CPUs under consideration). A system spends most of its time idle. Our media machine and primary production rig both have 24/7 uptime, most of that being idle but awake (S0). The time spent under load is a fraction of this. Just during typing this post, this unspecified production rig is operating at ~90W. Not much. If planning for long uptimes and targeting low power consumption, remember that power management (UEFI utilities, C-States, APM, etc.) and idle draw are both critical to consumption savings.

Templated System Performance

The full system builds are on the methodology page, but we're showing some of the 'core' specs with each system below. These tests seek four metrics: Idle watt consumption (from the desktop, after a period of inactivity), gaming watt consumption (DiRT Rally bench on loop), synthetic watt consumption (3DMark FireStrike), and full load consumption (100% GPU, 100% CPU).

A few things:

These tests, unlike those above, are not meant to be comparative or hierarchical. The objective is not to say that System X is objectively superior to System Y on the basis of power consumption. To this end, despite our provision of one toward the bottom, the configurations are not linearly comparable due to methodological differences. For our “gaming” benchmarks, we selected settings that seemed appropriate to the configuration in effort to achieve the most realistic result. We targeted a 60FPS framerate. This sometimes meant Ultra + 1440p settings, sometimes it meant Ultra + 1080p.

For 3DMark synthetic testing, we used FireStrike Ultra on loop in all instances, sans the system with the 750 Ti (ran Extreme). This test was the “combined” 3DM Physics + GPU benchmark. The full load benchmark ran Graphics 2 (GPU loaded almost exclusively), with HandBrake live encoding a 4K gameplay video in the background (CPU loaded almost exclusively).

Ultra Budget / Low-End: 760K + GTX 750 Ti

watt-draw-bench-760k-750ti

The 760K, now deprecated and championed by the 860K, was used with an A88X FM2+ motherboard and relatively low-power GPU. The GTX 750 Ti uses no power headers from the PSU, instead sourcing the entirety of its watt requirement through the PCI-e slot (which can afford upwards of 75W).

Idle sits at an impressively low 60.15W. Gaming performance, in our real-world setting, demands 160.12W. 3DMark (synthetic physics + graphics test) eats 165W. At the absolute high-end, in scenarios where the user is actually fully saturating the system – this is uncommon for mainstream users and “gamer only” types – we see a draw of 192.17W. It's good to allow for that ceiling in our PSU calculation.

Depending on the PSU's efficiency curve and other specs, something with a wattage of 400W would be ideal for this build (~50% consumption for peak efficiency, as a general assumption). The wattage could actually go lower and trend us toward a 60-80% PSU utilization, but there's a bizzaro-land of PSUs in the sub-~400/450W range that highly consists of low-quality products (think: OEM supplies for systems without a GPU). We're trying to avoid bad caps and leaking / rapidly aging components, so it is sometimes necessary – at this price-range – to forgo ideal wattage for higher overall build quality. To this extent, if a 400W PSU of reasonable quality is not available in your region of purchase, there's an increasing offering of 450W PSUs within the market. Enermax, EVGA, Corsair, Rosewill, SeaSonic, SilverStone, and plenty of others all offer units we'd recommend in the 400W-450W range. Take these, for instance:

Example Choices
PSU Modularity / Notes Efficiency Price
EVGA 100-W1-0430 430W No 80 Plus, 80% $40
Corsair CX430W No 80 Plus Bronze, 82-85% $43
Rosewill ARC-450W No 80 Plus Bronze, 82-85% $45
SilverStone SFF SFX450W No
Use for SFF Builds!
80 Plus Bronze, 82-85% $75
Enermax Revolution X't 430W Semi-Modular 80 Plus Gold, 87-90% $78

Entry-Level: i3-4160 + GTX 950

watt-draw-bench-i3-950

This system, with a mind-blowing two sticks of RAM (and i3-4160 + GTX 950), is configured as a sort of “entry-level gamer” system. Now, realistically, I'd probably spec this with 8GB of RAM – but we wanted to play around a bit.

The system drinks about 180W when gaming, but about ~206W under our full load test. At this power consumption, the above listing of PSUs is still perfectly viable – something in the 400-450W range is what this deserves for assumed peak efficiency at 50% load. Again, we can push load percentage higher (decrease watt availability), but options below ~400-450W are sparse and generally poorer quality-to-dollar ratios.

Mid-Range: i5-6600K + GTX 970

watt-draw-bench-template-i5-970

The i5-6600K is still new and, thanks to supply & demand, a bit overpriced. Still, it'll eventually work its way into systems and replace the current 4690K CPUs. The 6600K + GTX 970 rig uses four sticks of DDR4-2666 memory and a similar fan/SSD load-out as the other configurations. This build draws about 65W idle – appreciably low, given the relative performance of the build – and ~256W when gaming. 3DM synthetic testing puts us just a bit higher, 266.7W, but is surpassed by our full load test at ~290W.

In its worst case scenario, 290W draw (uncommon in gaming, but possible), the system would load a 500W PSU to about 58%, a 550W PSU to about 53%, and 600W to 48%. 500W is acceptable here and lands us close enough to ideal operating range that it'd be a go-to choice. Were I to personally shop for such a configuration, I'd include 450-550W PSUs in the search, then pick based upon merits (not just wattage) of the options. I would trend higher if leaving room for future GPU upgrades. They'll all fully support the build as is, and with near peak efficiency. Best not to limit options. 450W would work – 68% loaded – but leaves less room for higher TDP upgrades, overclocking, runs marginally less efficiently, and hotter given the increased load on the unit. Regarding upgrades, we've anecdotally experienced that the average user rarely upgrades a system in a way that substantially increases the watt draw bottom line. Overclocking, however, absolutely increases power draw (see GPU charts above).

Example Choices
PSU Modularity / Notes Efficiency Price
EVGA 100-B1 500W No 80 Plus Bronze, 82-85% $46
Rosewill Photon 550W Fully Modular
Extra long cables
80 Plus Gold, 87-90% $80
Corsair RM 450W Fully Modular

Low-noise caps & transformers to reduce coil whine.

Additional design focus on fan speed regulation.
80 Plus Gold, 87-90% $90
Enermax EDF550W Fully Modular

Fanless / Passive cooling (0 noise) with continuous power at 40C.

Copper-bridge array transmission ensures clean voltage delivery.
80 Plus Platinum, 92-94% $250

Again, the purpose of the examples isn't necessarily to outline an end-all list – we've left some price gaps in there – but to provide a guideline. Because so many system configurations fall within the ~500W range, more granularity is required to adapt to builds with specific needs. Anyone hoping to perform more extreme overclocks should trend toward higher-end options (read: that 500W EVGA PSU, although fine for its price, isn't the best for high sustained overclocks; the RM450 will more quickly run out of 'room' for OCing & upgrades). Overclocking draws more power and generates more heat, so a reinforced focus on cooling efficiency and power delivery (clean, stable voltage with low ripple) is encouraged.

At the low-end, EVGA's 500W PSU would work, but it's not a luxurious experience. Cable management is more difficult without modularity, overall build quality is lower, noise is higher than on the two high-end choices, but it works. The Photon provides superior efficiency and some luxuries (like 600mm long cables, for use in full towers). The Photon PSUs are fully modular, an added bonus for management, but still lack some of the finer design points of the Corsair RM450 (focus on fan speed regulation) and Enermax EDF550 (which is passively cooled). This isn't to say the EVGA or Rosewill units are bad, they're just targeted toward budget and entry-level markets.

High-End: i7-4790K + 2x R9 380X CrossFire

watt-draw-bench-i7-380x2

Any multi-GPU configuration is going to draw a lot more power. For the 980 Ti SLI and Fury X Crossifre watt draw plotting, we generally see about 50% more power consumption (+/- 4%, depending on configuration) than a single-card equivalent.

Our i7-4790K + R9 380X CrossFire build pulls 88W at idle, 447.38W when gaming, 484.7W when running synthetic tests, and a staggering 565.28W when pushing full load on the CPU and both GPUs. Despite the 23% lower gaming power draw, we should still plan to include the consumption of full load within our purchasing. Even if a game doesn't load this system to 100% CPU, 100% GPU load, spikes can still cause sudden increases in power draw (and games become more intensive, and use scenarios / interests change). It's good to consider the full draw in our plans.

This is a good example of where it is worthwhile to consider a single, high-end GPU as an alternative to dual GPUs. Even with boosted performance or potentially superior perf-per-dollar on the GPU purchases, the perf-per-watt can be driven down comparatively.

If we're buying within spec of full load consumption, we're looking at PSUs in the ~1000~1100W range for 50% load; this is getting a little bit extreme as an attempt at “saving” (so to speak) a few percentage points of power, considering the general price hike of PSUs in this range. Speaking through the visage of practicality, trending more toward 700-800W PSUs is the most cost effective. In general, we'll lose a couple percent off the average efficiency curve (2-3%). A 700W PSU will be 80.7% loaded by the full 565W draw of the system. An 800W PSU will be loaded 70.6%, providing more “comfort room” for overclocking, potential add-ons, LED kits, capacitor aging, and other incidentals.

All PSUs below are fairly high-end. There is less quality disparity than in other price points. We've broken this one into two tables.

Assuming the user is comfortable in the 70-80% PSU load range:

Example Choices
PSU Modularity / Notes Efficiency Price
Rosewill Capstone G 750W Semi-Modular

5-Year Warranty (parts); 1-Year warranty (labor)
80 Plus Gold, 87-90% $80
Enermax Revolution X't 730W Semi-Modular

5-Year Warranty
80 Plus Gold, 87-90% $110
EVGA SuperNOVA G2 750W Fully Modular

10-Year Warranty
80 Plus Gold, 87-90% $125
Corsair HXi 750W Fully Modular

7-Year Warranty
80 Plus Platinum, 90-92% $130

Assuming a user really wants to hit that ~50% efficiency target, here's what some sample options look like:

Example Choices
PSU Modularity / Notes Efficiency Price
Corsair RMx 1000W Fully Modular

7-Year Warranty

Variable fan RPM
80 Plus Gold, 87-90% $135
EVGA SuperNOVA G2 1000W Fully Modular

10-Year Warranty

Additional design focus on fan speed regulation
80 Plus Gold, 87-90% $150
SeaSonic X-1050 Fully Modular

5-Year Warranty

Variable fan RPM
80 Plus Gold, 87-90% $170
Enermax ERV1000EWT-G 1000W Semi-Modular

5-Year Warranty

Dynamic response & cross regulation reduce efficiency loss during conversion
80 Plus Gold, 87-90% $185

Production: i7-5930K + GTX 980 Ti

watt-draw-bench-i7-980ti

Looking more to the production side of things, the i7-5930K + (reference) GTX 980 Ti consumes higher wattage at idles, resting around 106.2W, with gaming workloads at ~305.6W. Synthetic consumes 361.6W. Full load hits 425.7W.

As above, sticking to a 50% PSU load for peak efficiency would demand an 850W PSU – a bit extreme – but the system is still perfectly operable at 550W (77.5% load) or 600W (71% load) power supplies.

Example Choices
PSU Modularity / Notes Efficiency Price
EVGA SuperNOVA G2 550W Fully Modular

7-Year Warranty
80 Plus Gold, 87-90% $80
SeaSonic SSR-RM 650W Semi-Modular

5-Year Warranty
80 Plus Gold, 87-90% $80
Corsair RM 550W Fully Modular

5-Year Warranty
80 Plus Gold, 87-90% $100
Enermax Platimax 600W Fully Modular

3-Year Warranty
80 Plus Platinum, 90-92% $132

All Templated Systems Compared

watt-draw-bench-all-templates

Conclusion: How Many Watts Do You Really Need?

platimax-psu-2

We're almost at 7000 words. Recapping the entirety of this article is truly terrifying. If you're skipping here to read a brief, all-answering conclusion, we'd strongly recommend revisiting some of the test sections. That is where the true analysis happens.

Buying a quality PSU is one of the best things builders can do for their systems. PSUs govern, to some extent, service lifespan of other components in the system. Power is responsible for sustaining and killing components – best to do it right. A big point we've been trying to make is that it isn't all about wattage – there's so much more to a PSU than picking the cheapest unit with the highest wattage, and we've only scratched the surface. One unique consideration we didn't even touch yet, for instance, is that of cannibalizing old systems for parts: A good PSU might survive a few system builds, migrating from one to the next and saving money each time. There is some level of capacitor aging and decay as PSUs get abused, but a quality product will endure years of service life.

Other considerations are noise compared against utilization: As PSU utilization ramps (% used), heat rises, and the active cooling system (if there is one) will compensate to maintain a target temperature. This often means fan RPM increases or a rise/fall pattern, both of which generate more noise. Depending on how concerned you are with noise levels, this may be a worthwhile consideration in system builds. SFF HTPC builds are particularly critical to keep quiet, given their presence in (normally) living room environments.

Overclocking and overvolting also drive power draw and heat upwards, and should be factored into PSU selection if in the system's life plan.

But let's stick to the scope of the article and close it out: Wattage. It's amazing what can be done with a 550W PSU or 450W PSU. Gaming builds equipped with i5 CPUs and GTX 970s can be powered at around 450-550W with relatively high efficiency. It's not until getting into CrossFire or SLI that the real power draw shows itself, pushing ever toward or beyond the kilowatt range. For single-card systems, especially those with no OC plans, it's worth checking our templated build charts to see how truly low power consumption is with modern components.

Our thanks to Enermax for sponsoring the exhaustive testing conducted for this article. For anyone who requires assistance in PSU selection, please seek one-on-one help on our forums.

If you like this kind of content, please consider supporting our efforts on Patreon.

- Steve “Lelldorianx” Burke.


« Prev Next

Last modified on December 02, 2015 at 10:13 pm
Steve Burke

Steve started GamersNexus back when it was just a cool name, and now it's grown into an expansive website with an overwhelming amount of features. He recalls his first difficult decision with GN's direction: "I didn't know whether or not I wanted 'Gamers' to have a possessive apostrophe -- I mean, grammatically it should, but I didn't like it in the name. It was ugly. I also had people who were typing apostrophes into the address bar - sigh. It made sense to just leave it as 'Gamers.'"

First world problems, Steve. First world problems.

We moderate comments on a ~24~48 hour cycle. There will be some delay after submitting a comment.

Advertisement:

  VigLink badge