Test System And Test Conditions
Introduction
We chose the benchmarks carefully, because we wanted to avoid favoring one company's architecture over another. So, the breakdown is evenly split between them. We also wanted to include multiple levels of performance testing across versions of DirectX. Our goal is to provide an assessment that is as balanced and as possible.
Technology
| Test System | |
|---|---|
| CPU | Intel Core i5 2500K @ 4.5 GHz |
| CPU Cooler | Prolimatech SuperMega + Noiseblocker Multiframe PWM |
| Motherboard | Asus P67 Sabertooth Rev. B3 |
RAM8 GB Kingston HyperX 1600 "Genesis"
SSD256 GB Samsung MZ5PA256HMDR-01000Power SupplyCougar GX 1050, 80 PLUS GoldTotal1050 W
Combined Power 3.3V/5V160 WCombined Power 12V1008 WEfficiency93 %Operating SystemWindows 7 x64 UltimateMeasurement DevicesCurrent and Power Consumption Measurement
Energy Logger 4000 (Conrad Electronic)• Long term measurements• Monitoring• Measurement of power consumption up to 1.2 KWVoltcraft
SBC-500 (Conrad Electronic)• Precise measurement of milliwatt• Measurement of power consumption up to 500 WNoise Measurement
Voltcraft SL-400 (Conrad Electronic)• Noise Measurement• Long-term recording• Monitoring
Benchmark selection
Synthetic Tests
We begin our tests with 3DMark 11, which we use to assess processor-independent graphics workloads. The results are interesting and relevant, as the benchmark is relatively new. We deliberately exclude the 3DMark06 and 3DMark Vantage benchmarks because they more out-dated and they include optimizations for things like PhysX, which can really throw off comparisons. Our second synthetic benchmark is Sanctuary v.2.3 from Unigine. It does not support DirectX 11, which we feel is not necessarily an issue today: many games on the market either don't explicitly support DirectX 11 or they rely on DirectX 10 features.
Games
Our primary goal with game selection is to provide a balanced overall picture that is as relevant as possible for real gaming. A lot has been written on tessellation and physics, but at the end of the day, these features only really matter so much as they pertain to real games.
| Synthethic Benchmarks | |
|---|---|
| 3DMark 11 | Graphics Tests 1-4 |
| Unigine Sanctuary v.2.3 | Dynamic lightsHDR renderingParallax occlusion mappingAmbient occlusion mappingTranslucenceVolumetric light and fogParticle systemsPost-processingInteractive experience with fly-through mode |
| Gaming Benchmarks | |
| DirectX 11 | Metro 2033Aliens vs. Predator |
| DirectX 10 | Call of JuarezS.T.A.L.K.E.R.: Call of Pripyat |
| DirectX 9 | Mafia II |
Performance Categories: Gamers and Enthusiasts
After much consideration, we decided to organize our test results into two performance categories, which also reflect pricing. The Gamer category encompasses a screen resolution of 1680x1050 pixels, while the Enthusiast category is based on a 1920x1080 resolution. After much internal discussion and consulting with some of our readers, we decided not to use the oversized 2560x1600 and smaller 1280x1024 resolutions, as they don't seem to be relevant to make readers any more. If anything, moving forward, we're going to try getting more multi-screen configurations included for high-end graphics card reviews.
Performance Index and Calculation Base
Instead of using a cumulative frame rate for overall performance classification, we create two new performance indexes, the Gamer index and the Enthusiast index. Each card can be assessed in terms of its performance in both the gamer and the enthusiast segments. Therefore, a card that performs poorly in the enthusiast segment may be fully acceptable for gamers. Cumulative frame rates say nothing about a card's capability under certain conditions. Hence, we replaced this measurement with our more informative indexes.
So how do we get our results? We take a representative card from each manufacturer's mid-range segment (AMD Radeon HD 6870, Nvidia GeForce GTX 560 Ti) as well as from their top-end (AMD Radeon HD 6970, Nvidia GeForce GTX 570). Performance evaluation happens across a total of 15 games and we group the results accordingly. We measure at the appropriate quality settings and define the highest achievable frame rate as 100%, for both the Gamer and Enthusiast indexes. It might sound complicated, but it's actually quite simple. If you stick to common game settings, our indexes provide good indications of the price/performance ratio.
Power Consumption and Temperature Measurements
After several previous attempts, we decide to take a different approach this time around, hoping to avoid fluctuations. We push our CPU to 100% load using Prime95, but let these threads run in the background with a low priority. Then we measure the power consumption of an entry-level graphics card, the results of which we are well acquainted with. We do this once in idle state (graphics idle) and a second time with FurMark running. By subtracting the values for power consumption in idle and peak load mode from the total power consumption, we get the total power consumption of the system at peak load without a graphics card. Our values are accurate up to 1 W, so we only need to subtract the 131 W system idle power from all subsequent readings to obtain the effective power consumption of each graphics card. Of course we look at the real-time power consumption results as a control and check for possible variations.
We measure temperature both at idle and under full load. A constant room temperature of 72°F (22°C) was maintained in the test environment.
Noise Level Measurements
We use our standard method here, measuring the noise from a defined, unchanged distance to the center of the graphics card. All other noise sources are removed from our test lab during noise testing.
