POV-Ray is one of our two content creation benchmarks, it uses ray tracing for rendering a scene. Ray tracing has gotten a bit more attention recently, as a higher fidelity alternative to the rasterization pipeline. Ray tracing renders by bouncing rays off objects in a scene; by following light rays for a greater number of bounces, a scene can be depicted with ever greater realism. A standard rasterization is equivalent to ray tracing with no higher order rays. The increased fidelity of ray tracing is vital for certain applications. One common use is visualizing automobiles and the interactions between the paint, weather, reflections etc.
POV-Ray 3.7 beta 36 is the latest 64-bit version available for Windows. We use the standard benchmarking scene run both in single threaded and multi-threaded mode.
Figure 1 – POV-Ray Performance
As expected, the single threaded performance for Westmere is almost identical to Nehalem. However, the multi-threaded performance increases by 42% for the extra 2 cores. Not perfect scaling, but pretty good.
The next chart shows the average power consumption for POV-Ray. One modest advantage of the shrink to 32nm for single threaded applications is that the power consumed by each core has decreased slightly. Using one thread, Nehalem runs at 187W, while Westmere uses 175W. As expected, when running with all threads, both Nehalem and Westmere will max out their power consumption at the same level.
Figure 2 – POV-Ray Power Consumption
To calculate power efficiency, we evaluated the performance in POV-Ray (expressed in Pixels Per Second) divided by the average power while ray tracing. This tends to emphasize the importance of active power control and de-emphasize idle power. In single threaded mode, Westmere’s power efficiency advantage is modest – around 7%. On the other hand, the 42% performance advantage for Westmere directly translates into 42% more power efficiency.
Figure 3 – POV-Ray Power Efficiency
Discuss (6 comments)