Radiance Benchmark Test

Tabulated below are run times for the rpict step of the process involved in making an accurate rendering with Radiance. This test was designed mainly to exercise the ambient calculation portion of Radiance, but includes other types of commonly-used elements, namely daylighting, mesh primitive, antimatter, alias, brightfunc, and colorpict. It supports parallel computers (SMP only) via rpiece, though it may be exteneded to support workstation clusters in the future.

More benchmarks at LUXAL Go to the old benchmark page.

Instructions

1. You must download and build Radiance, or have the Radiance distribution on the machine you wish to benchmark. Download the Radiance distribution from the "download Radiance" page of the Radiance Online site.

2. Go to the Radiance Benchmark github page and clone the repository locally.

3. The complete instructions are in the README file. Most users just need to run

   make

   and read the results at the end of the output. SMP users using the bash shell can run one of the following:

   NCPU=2; make smp

   export NCPU=2; make smp

   to run the benchmark on two available CPUs.
4. Edit the

   rad_bench.html

   file and either e-mail it to me or submit a pull request. Thank you for contributing!

Parallel tests

Though it appears difficult to create an rpiece benchmark with timing routines that will return comparable timing accuracy on all platforms, I recognize the interest such results may have to the community. Please submit any SMP results that you may have and they will be posted at the bottom of the page.

Notes

• As of Dec 2010, the Makefile has been modified to work correctly and consistently across any version of Radiance from the last 5-10 years.
• CPUs from 2009 and later are likely to run at clock speeds different than their rated speeds. This is because manufacturers are now including dynamic overclocking on the chip, which sets the clock speed according to the thermal load and not the rated and advertised speed. For example, a single-thread job on an otherwise-unloaded 2.5 GHz 4-core machine might run at closer to 3.0 GHz, while a 4-thread job runs at the rated 2.5 GHz. Since we cannot accurately discertain the true clock speed for any given test, the clock speeds listed below are the standard or rated speeds given by the manufacturer, and not the actual internal clock speed used during the test. Naturally, if you purposefully overclock your machine, please report the effective standard clock speed, not the base speed.
• The default SMP job now uses one column and 64 rows. This seems to produce the fastest times for a wide variety of machines.

Serial Benchmark Results

SMP Benchmark Results