Skip to main content

How to Stress-Test Graphics Cards (Like We Do)

Witcher 3, Doom (2016) & Metro: Last Light


The first picture shows the graphics card after idling for ~20 minutes. Some of the temperatures are on the high side, which we'd expect from a card using semi-passive cooling.

The Witcher 3

The Witcher 3 is our go-to title for power consumption measurements because it almost always ends up generating results in line with manufacturer-defined power limits. The Skellige map produces the highest GPU and voltage converter loads that this game offers, so that's what we use. Memory module temperatures are generally average, but The Witcher 3 still serves us well as a stability test for GPU and memory overclocking.

Depending on the performance of whatever card we're testing, we tend to use a resolution of 1920x1080 or 3840x2160. It’s important to make sure that Nvidia’s HairWorks effects are disabled, and the quality settings are set as high as possible while still achieving >30 FPS. Higher settings are better for maximizing loads and temperatures.

Measurement64 °C86.2 °C80.1 °C63.4 °C98.5W
Compared to Maximum98.5%95.7%94.8%87.2%95.0%
Assessment- Very high realistic gaming power consumption- High GPU temperature for cooling tests- High package temperature for stability tests- Medium memory temperature
Use for- Power consumption measurements- Great stability test for overclocking

Doom 2016 (Vulkan/OpenGL)

Doom doesn’t challenge hardware quite as much as The Witcher 3, but frame rates of 40 to 50 at high enough resolutions still heat up graphics cards. Due to lower power consumption figures, clock rates tend to be higher under Doom, exceeding what you'll see under The Witcher 3. Testing the highest possible GPU Boost frequency steps (Nvidia) or P-states (AMD) is easy with Doom.

Doom is also an interesting alternative for testing the stability of overclocked memory, since its load is actually higher than the one generated by The Witcher 3.

Measurement62 °C81.7 °C77.4 °C63.6 °C95.4W
Compared to Maximum95.4%90.7%91.6%87.5%92.0%
Assessment- Medium realistic gaming power consumption- Not suitable for cooling tests of any kind- Medium memory temperature- Allows for very high boost frequencies
Use for- Serviceable stability test for overclocking- Gets better with increasing resolution

Metro: Last Light (Redux)

Metro: Last Light is a mixed bag. On one hand, testing at 1920x1080 is enough to produce some of the highest memory loads and temperatures. On the other, WQHD or UHD are necessary to generate high power consumption numbers. When we use FHD, we end up with lower power consumption measurements than those achieved under The Witcher 3. Again, quality settings should be set as high as possible while still achieving >30 FPS.

Owners of Nvidia graphics cards have the option to further increase power consumption by activating GPU-accelerated PhysX. AMD users can try this out at least in part by using CPU PhysX to increase one CPU core’s load. Using this method, the average host processing load of 19% goes up significantly. With a stronger graphics card, CPU loads exceeding 30% become a distinct possibility.

Measurement63 °C81.5 °C75.4 °C65.2 °C95.9W
Compared to Maximum96.9%95.7%94.8%89.7%92.5%
Assessment- Medium realistic gaming power consumption- High GPU temperature for cooling tests- Medium package temperature- Very high memory temperature for a game
Use for- Good stability test for overclocking of high-performance graphics cards from WQHD upwards- Memory tests

MORE: Best Graphics Cards

MORE: Desktop GPU Performance Hierarchy Table

MORE: All Graphics Content

  • phobicsq
    Doesn't hwmonitor cost money?
  • FormatC
    HWiNFO is free :)
  • Th_Redman
    Great article of information Igor(and Tom's, of course). I use a number of these stress tests and your article listed some I've never heard of or read about, so thank you.
  • Jay E
    But did you eat the egg?
  • Unolocogringo
    Very nice article to point newbie overclockers towards. You have to have some basic understanding if you want to overclock successfully.
    I have overclocked everything possible since my first overclock.A pentium 75mhz I overclocked to 90mhz. This was mid to late 1996. I learned how to do it from This site. Toms Hardware ( back then).
    Since I overclock every thing to stable 100% load 24/7/365 for Folding@Home and occasional gaming, It must be 100% stable for correct folding results. And of course gaming with my son and grandson.
    I use most of the tests and tools you do, except for the fancy thermal images, to achieve this. Nice to know my testing methods are the Same as yours, but mine last 36 to 48 hours on final overclock settings before being put into service.

    Overclocking is a serious affliction , even my non overclockable SuperMicro 2p server board is overclocked from 2.5 to 3.0 on all 8 cores and folding away for years. :)
    Enjoyed your article and testing methodology explained. Thanks
  • ddferrari
    For my uses (gaming, surfing) I see no reason to push a component to its power or thermal limit via synthetic tests. All that does is shorten its life span. I don't care if my OC fails during a multiple hour, unrealistic load. If it runs fine during real-world usage then I'm satisfied.

    All my components are overclocked, and I test them for stability the old fashioned way: I USE them. They key is to overclock only one component at a time and see if problems arise while gaming. I keep bumping up the OC until an issue pops up- then I know where the maximum lies.

    There seems to be a lot of monkey-see-monkey-do going on around the internet these days.
  • stonedwookie
    We dont care about the stats what we want to know is did the eggs taste good?
    what would you rate the eggs ?
  • FormatC
    The egg got only three-stars rating (3/5).

    The reason why:
    It was simply too long for my taste and it is a real pain to look over such a long time at this egg if you are hungry :P
  • Co BIY
    What is the best thermal paste to use for a mining rig omelet pan?
  • FormatC
    Olive Oil. The best taste :)