One of the most popular questions regarding GPUBox Artist refers to performance and scalability. We prepared a short but substantial clip that presents you those two aspects of using Octane Render within the GPUBox-powered infrastructure.
For the purpose of this video we used Octane Render 1.55 Standalone running on CentOS 6.4. The scene that was used for the test was well-known OctaneBenchmark scene on default settings (except samples, which were set to 6000) and Path Tracing kernel.
In the first part of the video we compared two rendering sequences. The left side presents rendering the scene on native 4 GPUs (2 x GeForce GTX 690). Simultaneously, on the right side of the screen you can see rendering the very same scene on the same settings, on identical GPUs, but mounted in a different PC and virtualized with GPUBox Artist.
Native rendering was finished after 4:05, while rendering using the GPUs remotely with GPUBox took 4:07.
In the second part we virtualized and engaged another six GeForce GTX 690 cards (12 GPUs) and launched again the same scene on a total number of 16 GPUs comparing it to native rendering on 4 GPUs. Theoretically, rendering on 4 times as much GPUs should result in 4 times faster rendering and this is what happened here – the rendering took 1 minute and 1 second.
During the test we were using a 20Gb/s InfiniBand network, but additionally we launched the same scene on 1 Gb Ethernet which is not presented in the video. The results are shown in the following table: Continue reading