Why you can trust Tom's Hardware Our expert reviewers spend hours testing and comparing products and services so you can choose the best for you. Find out more about how we test.
How We Test
Test System
In order for us to compare results between cards reviewed by any of our reviewers, Tom’s Hardware standardized the test bench that we all use for 2015. By doing this, each of our locations can have different sets of cards and still compare from the pool of results that any of us have obtained. Starting with an MSI X99S XPower motherboard, each test bench has an Intel Core i7-5930K overclocked to 4.2GHz, 16GB of Crucial Balistix Sport DDR4, two Crucial 500GB SSDs and an 850W power supply from be quiet!.
Software and Drivers
Our tests on the PowerColor Devil R9 390X took advantage of AMD's 15.7.1 Catalyst drivers. The comparison charts also include Sapphire's Nitro R9 390, which utilized Catalyst 15.7.
Results from the Sapphire R9 Fury Tri-X were pulled from our review earlier this summer. That card was only sent to us temporarily, prior to the publication of 15.7. So, those numbers were generated using the 15.15 driver package that was available back then.
Our GeForce GTX 970 results were taken from the Nitro R9 390 review, and generated using version 353.30 of the company's software. The GTX 980 and 780 Ti results were pulled from our GeForce GTX 980 Ti review from May. We used version 347.25 in that piece.
Noise
Audio measurements are taken in a completely silent room. Typically, we record our readings two inches from each card's rear I/O bracket. But with PowerColor's closed-loop cooling system, two measurements were needed: one at the rear of the card and a second at the radiator exhaust point.
The meter I have at my disposal is only sensitive to 35 dB, so anything lower than that registers as zero. If the card makes an audible noise that the meter does not detect, I mark it as 34 dB. If no audible noise is observed, then it receives a 0 dB result on the graph.
Power
To test power consumption using our reference platform, a bit of creative math is needed. Since Haswell-E processors don’t have integrated GPU cores, we can’t boot the system without a discrete board installed to get a baseline. We are able to estimate consumption based on the approximate power draw of the test bench, though. In our observations, we’ve found that the approximate power draw from everything other than the GPU is 120W. By deducting that from the recorded wattage reported on our in-line power meter, we can calculate the approximate draw of the GPU.
Comparison Graphics Cards
MORE: Best Graphics CardsMORE: All Graphics Content