Sign in with
Sign up | Sign in

Test System And Benchmark Selection

Micro-Stuttering And GPU Scaling In CrossFire And SLI
By , Greg Ryder


Test System
Intel Core i5-2500K @ 5 GHz to verify GPU scaling
Prolimatech SuperMega + Noiseblocker Multiframe PWM
Gigabyte Z68X-UD7

8 GB Kingston HyperX 1600 "Genesis"

System SSD
256 GB Samsung 470 Series
Power Supply
Cougar GX 105080 PLUS Gold
Total1050 W
Combined Power 3.3V/5V160 W
Combined Power 12V1008 W
Efficiency93 %

Windows 7 x64 Ultimate
Test Equipment
Current and Power Meter
Energy Logger 4000 (Conrad Electronic)

• long-term measurements
• monitoring
• power measurements up to 1.2 KW
Voltcraft SBC-500 (Conrad Electronic)

• precise measurements down to the milliwatt range
• power measurements up to 500 watts
Sound Level Meter
Voltcraft SL-400 (Conrad Electronic)

• sound level measurements
• long-term recording
• monitoring
Syntethic Benchmarks
3DMark 11
  • Grahpics Tests 1-4
Unigine Sanctuary v.2.4
  • Dynamic lights
  • HDR rendering
  • Parallax occlusion mapping
  • Ambient occlusion mapping
  • Translucence
  • Volumetric light and fog
  • Particle systems
  • Postprocessing
  • Interactive experience with fly-through mode
Gaming Benchmarks
DirectX 11Metro 2033
Aliens Vs. Predator
DirectX 10Call of Juarez
S.T.A.L.K.E.R.: Call of Pripyat
DirectX 9Mafia II

Performance Categories: Gamer and Enthusiast

Although we cover a wide range of settings and resolutions in our launch coverage of various products, and will continue to do so, this piece includes two more generally-applicable performance categories that also take the acquisition cost into consideration. Our Gamer category is tested at 1680x1050, and our Enthusiast category is benchmarked at 1920x1080. We also base the image quality settings on both classifications.

Performance Index and Calculation Method

Instead of using cumulative frame rates to rank a graphics card, this review ranks cards in those same Gamer and Enthusiast classes. A graphics card that may be ranked rather low in the Enthusiast column might be completely usable in the Gamer category. Cumulative frame rates do not shed light on performance under specific conditions. Thus, we're replacing that old system with more transparent rankings.

So, how do we obtain these values? We take one representative from each GPU manufacturer, both for mid-range cards (AMD Radeon HD 6870 and Nvidia GeForce GTX 560 Ti) and high-end cards (AMD Radeon HD 6970, Nvidia GeForce GTX 570), and assess their behavior in 15 games. Then, we compute an average for each of the two card groups, creating two fictional 'Raforce Gedeon' cards. We use these two models as a reference for quality settings in each benchmark test and dub these synthetic frame rates the 100% level for the Gamer and Enthusiast categories, respectively, for each benchmark. The sum of all performance percentages, divided by the number of tests, yields two indexes. As the 'Raforce Gedeon' card is manufacturer-agnostic, advantages held by specific cards are eliminated from the get-go. This might sound complicated, but it really isn't. So long as you test using typical game settings, the rankings are a good indication of the price/performance ratio for each graphics card.

Power Draw and Temperature Measurement

We fully-loaded our CPU with Prime95 running threads at low priority in the background. Then, we measured the power draw of an entry-level graphics card with well-known power characteristics (first at idle, and then with FurMark running). Subtracting the card’s idle and load power from the total system power yielded, in each case, the total system power draw without a graphics card. These two values nearly matched, which allowed us to simply subtract 131 watts from all subsequent power measurements in order to obtain the power draw of the graphics card by itself. Of course, we also analyze the total power draw as a sanity check.

React To This Article