How We Test
Our reference system hasn’t changed from a performance standpoint, though we did alter its cooling solution. Instead of the closed-loop liquid cooler, we're now using an open-loop solution by Alphacool with two e-Loop fans from Noiseblocker.
We changed our test system for two reasons. First, the near-silent setup enables us to take more accurate noise measurements. Second, it provides our new and even higher-resolution infrared camera with an unobstructed 360-degree view of the graphics card. There are no more tubes or other obstacles.
We follow the general trend and use Windows 10, which keeps us current and allows us to use DirectX 12.
|Contact-free DC measurement at PCIe slot (using a riser card)|
Contact-free DC measurement at external auxiliary power supply cable
Direct Voltage Measurement at Power Supply
Real-time infrared monitoring and recording
|2 x HAMEG HMO 3054, 500MHz digital multi-channel oscilloscope with storage function|
4 x HAMEG HZO50 current probe (1mA - 30A, 100kHz, DC)
4 x HAMEG HZ355 (10:1 probes, 500MHz)
1 x HAMEG HMC 8012 digital multimeter with storage function
1 x Optris PI450 80Hz infrared camera + PI Connect
|Intel Core i7-5930K @ 4.2GHz|
Alphacool water cooler (NexXxos CPU cooler, VPP655 pump, Phobya balancer, 240mm radiator)
Crucial Ballistix Sport, 4 x 4GB DDR4-2400
MSI X99S XPower AC
1x Crucial MX200, 500GB SSD (system)
1x Corsair Force LS 960GB SSD (applications, data)
beQuiet Dark Power Pro 850W PSU
Windows 10 Pro (all updates)
|Driver||AMD: 15.11.1 Beta (press driver)|
Nvidia: ForceWare 358.91 Game Ready
|The Witcher 3: Wild Hunt |
Grand Theft Auto V (GTA V)
Metro Last Light
Middle Earth: Shadow of Mordor
Ashes of the Singularity
Graphics Card Comparison
We’re using AMD’s Radeon R9 Fury Nano as the top of our range, allowing us to draw direct comparisons between it, the Sapphire R9 380X Nitro, a PowerColor Radeon R9 390, and a “smaller” MSI Radeon R9 380. These graphics cards should cover the entire range of AMD’s offerings in this segment. We were planning to add the Radeon R9 390X, but it’s just too close to the 390. In the end, we wanted to include at least one faster card for the higher-resolution tests.
MSI's GTX 970 4G and Gigabyte's GTX 960 Windforce represent Nvidia's portfolio. Interestingly, there's a large gap in the line-up between those two offerings. Or, seen from AMD’s point of view, there’s an opening to exploit that Nvidia created with its somewhat weaker GeForce GTX 960.
Benchmark Settings and Resolutions
The benchmarks are set to taxing detail presets, since that's what we expect someone who buys a graphics card in this price range to run. In order to demonstrate differences between the cards at progressively higher resolutions, we’re using Full HD (1920x1080) and QHD (2560x1440). AMD's new graphics card is specifically targeted toward the latter, it says.
Frame Rate and Frame Time
We completely updated how we represent frame time variance. In the end, percentages just don’t tell the whole story for longer benchmarks, which can have very different sections when it comes to rendering speed. We’ve settled on two ways of conveying the results. First, we show how long it takes to render each individual frame, telling you a lot more than bar graphs or an FPS graph based on averages. Second, we plot two different evaluations of each frame’s time.
We start by normalizing each frame time by subtracting the average of the overall benchmark’s frame times. This puts the curves for each graphics card at a common average on the x-axis. This allows us to more easily spot outliers. After doing this, we assess the curve’s smoothness, which is to say that we’re looking at the relative differences in render time between the frames. This helps us to find subjectively annoying stuttering or jumps more easily without having the actual frame time influence the curve.
Power Consumption Measurement Methodology
Our power consumption testing methodology is described in The Math Behind GPU Power Consumption And PSUs. It's the only way we can achieve readings that facilitate sound conclusions about efficiency. We need two oscilloscopes in a master-slave setup to be able to record all eight channels at the same time (4 x voltage, 4 x current). Each PCIe power connector is measured separately.
A riser card is used on the PCIe slot (PEG) to measure power consumption directly on the motherboard for the 3.3 and 12V rails. The riser card was built specifically for this purpose.
We are using time intervals of 1 ms for our analyses. The equipment cumulates the natively even more high resolution data for us so that we don't completely drown in the sheer amount of data that this system generates.
Infrared Measurement with the Optris PI640
We’ve identified a method to confirm what our sensors tell us and to spice up our usual temperature graphs a bit in the form of the PI640 by Optris. This piece of equipment is an infrared camera developed specifically for process monitoring. It allows us to shoot both videos and still shots at a good resolution, providing us with not just peak temperatures, but also a good view of any weak points in the graphics card's design.
Optris' PI640 supplies real-time thermal images at a rate of 32Hz. The pictures are sent via USB to a separate system, where they can be recorded as video. The PI640’s thermal sensitivity is 75mK, making it ideal for assessing small gradients.
As always, we use a high-quality microphone placed perpendicular to the center of the graphics card at a distance of 50cm. The results are analyzed with Smaart 7.
The ambient noise when our readings were recorded at night never rose above 26 dB(A). This was noted and accounted for separately during each measurement. The setup was calibrated on a regular basis as well.