GIGABYTE RADEON RX 6950 XT GAMING OC Review
Brandon Dodman / 3 years ago
How We Test
We continue to update our testing methods around once per year. As such, we re-test older hardware to reflect changes over time. These can be driver updates, Windows updates, game patches, and more; all of which have an impact on performance figures. Furthermore, we update our test bench to newer and more relevant hardware. This means that our new reviews aren’t always comparable to those of older reviews, so please compare the testing methodology on older reviews should you be trying to compare them with newer ones.
Replicate Our Results
When it comes to our benchmarks in our reviews, the benchmarks are pretty self-explanatory and kept as simple as possible, although there are a few exceptions. Remember that your choice of a graphics card, CPU, the silicon lottery, and other factors can yield different numbers, and there’s always a margin for error when using any software.
Links are provided below, as well as the settings we use. We encourage you to not just look at how one product compares to any other, but how it compares to your own. If you’re looking to build a new system, you should benchmark your current PC using the benchmarks available to you. You should then look at the percentage improvement from your current hardware to the hardware tested here to give you a ballpark figure of how much of an upgrade this will provide you with.
Graphics Card Test System
Motherboard | ASUS Crosshair VIII Dark Hero (Wi-Fi) BIOS 3904 |
Processor | AMD Ryzen 9 5900X |
RAM | 2x16GB Corsair Dominator Platinum RGB 3600MHz |
CPU Cooler | Noctua NH-D15S |
Power Supply | Corsair RM1000x |
Storage | Mushkin Vortex 2TB |
Chassis | Open Benchtable |
OS | Windows 10 Pro Build 19041 |
Additional Hardware
- Killawatt style electricity usage meter wall plug
- Precision Gold N05CC Decibel meter
Gaming
To gauge performance levels which are easily reproduced time after time, by both ourselves and our readers, we now stick strictly to games default profiles. Given how powerful most modern cards are getting, we’ve began using more demanding profiles for most games, but will otherwise state on a per game basis if we use other profiles beyond High, Ultra, Maximum or their respective equivalent at all resolutions. V-Sync is always disabled in our testing, as are FreeSync and G-Sync technologies or similar. We also exclude brand exclusive texts in tests that compare AMD/NVIDIA, such as DLSS only testing as that’s a one sided battle. Graphics card power usage is set to optimised or balanced, and the system power mode is set to High Performance within Windows and the available GPU driver software. Identical settings are used for all resolutions unless otherwise stated.
Games
- Assassin’s Creed Valhalla
- Borderlands 3
- Dirt 5
- Godfall
- Horizon Zero Dawn
- Metro Exodus Enhanced Edition
- Shadow of the Tomb Raider
- Watch Dogs Legion
Resolutions
- 1080p (Full HD)
- 1440p (2K)
- 2160p (4K)
- 4320p (8K)
Some tests will also include features such as NVIDIA DLSS and Ray Tracing. However, these will be clearly labelled and may only appear in reviews for graphics cards that support those features. 8K testing will only appear in reviews of flagship cards that we consider capable.
Software Used
- 3DMark Fire Strike (download)
- FireStrike
- FireStrike Ultra
- Port Royal DXR
- TimeSpy
- TimeSpy Extreme
- HWMonitor (download)
- GPU-Z (download)
- MSI Afterburner (download)
Noise
Everyone has a reasonable noise level preference when it comes to components on a computer. Some can handle all fans at 100% load to keep temperatures down; some want an entirely silent computer. To accurately gauge the noise output of a graphics card, we position a Precision Gold N05CC one meter above our open-air test bench and take an average reading at idle and load.
Power Consumption
We take power readings during the idle state with no background applications running. Then again at 50% completion of the Unigine Superposition benchmark, using the average as the final published result.
Temperatures
We take temperature readings after 10 minutes of desktop idle with no background programs running, then take the recorded maximum delivered from Unigine Superposition after a 4K optimised run, confirming the numbers are accurate with HWMonitor. Furthermore, the ambient temperature is always kept +/- 1c from 21c.