This article will serve as a reference for those who would like to compare results with our reviews, whether you are writing an article for the site or just testing the performance of your new GPU. The plan is to update the article as changes in our testing procedures occur. On to the gritty details…
Our test systems consist of mainstream parts since that is what the majority of people will be using. This allows our results to be relevant to a wider audience, and also allows for that audience to repeat our tests to compare with their own systems. The required parts and clocks are as follows:
|Test System Components|
|Motherboard||ASRock X370 Taichi, ASUS ROG Maximus X Apex|
|CPU||Intel i7 8700K @ 4.7 GHz / 4.3 GHz Cache|
|CPU Cooler||EVGA CLC 240|
|Memory||2×8 GB G.Skill Trident Z 3200 MHz CL15-15-15-35|
|SSD||Toshiba OCZ TR200 480 GB (OS + Applications)|
|Power Supply||EVGA 750W G3|
|Video Card||@ Stock (419.67 driver as of 4/2019)|
A special thanks goes out to EVGA for providing the CLC 240 CPU Cooler and 750W G3 Power Supply to cool and power the system, G.Skill for the Trident Z DRAM, and Toshiba OCZ for the 480GB TR200 SSDs storage running the OS, benchmarks, and games. With our partners helping out, we are able to build matching test systems to mitigate any differences found between using different hardware. This allows for multiple reviewers in different locations to use the same test system and compare results between reviewers minimizing system variance.
Synthetic benchmarks are very consistent and easily repeatable, making them some of the best tests for direct card-to-card comparisons. We have dropped support for Unigine Heaven and Valley and will continue with 3Dmark Fire Strike Extreme and Time Spy.
3DMark – Fire Strike Extreme
- Fire Strike – Extreme setting
- “Include Demo” can be unchecked since it doesn’t affect the score
3DMark – Time Spy
- Default setting (you do not have to run the demo)
We have updated our gaming suite to get rid of some of the older titles. Now all of the benchmarks are “canned” tests, making them as consistent as possible and more synthetic-like than typical gameplay. In general, we do our testing with the commonplace resolution of 1920×1080, all of the graphics settings maxed, and any manufacturer specific features disabled.
World of Tanks EnCore
Set “Ultra” defaults, the in-game graphical options should match the following:
Setup the video configuration as follows using “Very High” defaults with TAA, AF X16, run Australia track and show the FPS counter. The video configuration should look like the screenshots below, then run the built-in benchmark located in the options.
Shadow of the Tomb Raider
Run DX12 with “Highest” preset and RTX when available. The video configuration should look like the screenshots below, then run the built-in benchmark located in the options.
Far Cry 5
Setup the video configuration as follows using the “Ultra” preset. The video configuration should look like the screenshots below, then run the built-in benchmark located in the options.
Tom Clancy’s – The Division 2
Run DX12 with “Ultra” preset, V-Sync changed to off – the rest default. The video configuration should look like the screenshots below, then run the built-in benchmark located in the options for all three tested resolutions.
Ashes of the Singularity: Escalation
Run DX12 with “Crazy” preset and GPU focused. The video configuration should look like the screenshots below, then run the built-in benchmark located in the options. Ashes of the Singularity: Escalation
Run with DX12 enabled, V-sync off, DXR enabled with Ultra presets. The video configuration should look like the screenshots below, then run the Tirailleur episode under War Stories. Play the entire scene until the bunker is taken and you are notified of moving on. Frames are measured using FRAPS which is started after exiting the truck and the battle begins until this battle is complete and the hill is taken.
The following are a few things that will provide additional value and detail to the testing but are not required.
Measuring typical temperatures to expect and/or testing cooler performance.
- Record ambient temperature in degrees Celsius (preferably close to the intake fan of the GPU)
- GPU-Z, PrecisionX, Afterburner, etc. will be used for monitoring temperatures during the test
- Record the GPU’s idle core temperature in degrees Celsius after the card has been on and in the idle state for a few minutes
- Run Shadow of the Tom Raider and F1 2018 (5 laps) at the settings above using 2560×1440 resolution and record the peak temperature in degrees Celsius with the default fan profile (other manual fan speeds can be tested as well).
- When comparing temperature results, they all need to be normalized to the same ambient temperature (Currently 23C).
System Power Consumption
Kill-a-Watt meters (or similar) are used for measuring at-the-wall system power consumption.
- Record the system power consumption at idle
- Record peak power consumption during the following two benchmarks: Shadow of the Tom Raider and F1 2018 (5 laps) at 2560×1440 resolution.
Sound level meters are used to measure dBA as a means to quantify noise and perceived loudness.
- Tested on an open bench, which gives the worst case scenarios.
- Minimize all external sources of noise: Turn off A/C, turn off ceiling fans, turn off TVs, turn CPU fan off or as low as possible, etc.
- Manually set fan speed in varying increments via software such as PrecisionX or Afterburner.
- Record sound level at each increment. Be sure to get readings within your meter’s most accurate dBA range, regardless of distance. The most affordable meters are accurate at high dBA ranges, meaning the meter will have to be placed close to a quiet source for an accurate reading.
- Readings can be estimated at dBA ranges outside of your meter’s capability to get estimated sound level at longer distances.
- Typical linear percentages cannot be used when comparing dBA measurements because dBA uses a logarithmic scale. So, the rule-of-thumb is every 1 dBA difference is roughly equal to a 10% difference in perceived loudness.
The following equation is used for estimation of sound level at different distances.
L2 = L1 – 20 * log10(r2/r1)
- L1 = Sound level at the reference distance
- L2 = Sound level at the desired distance
- r1 = Reference distance
- r2 = Desired distance
Nope, this definitely isn’t the “conclusion” of this article, it will be slowly evolving, just like our GPU testing procedures. We’ll post comments with changes as they take place, so be sure to keep an eye on this, the comments, and the forum thread for updates. We are always open to feedback and suggestions as well. So, if you think there is something worth testing that’s not included, then feel free to speak up in the comments below.
Please leave any comments or suggestions in the comments below.