In this article and video, composer Alessandro Panella tests VSL’s new Vienna Power House – a GPU power extension for Vienna MIR Pro 3D, powered by GPU Audio, which offloads convolution processes to the graphics card in order to reduce the load on the CPU. Over to Alessandro…
CPU power is never enough, we always need some more to be able to perform the increasingly taxing duties of present-day studio workflows. Let’s take the mixing of film soundtracks as a general example: hundreds of tracks that need some kind of processing (from a simple corrective EQ to full-blown processing chains), in various output formats which need lots of channels to recreate the sensation of natural listening. This translates into huge amounts of data to be calculated by our machines… but we all know these things.
To give our CPUs some rest, Vienna has teamed up with GPU Audio to give us Vienna Power House - a GPU power extension for its acclaimed MIR Pro 3D convolution reverb.
The test shown in the video is aimed at determining the percentage of CPU power that can be saved by moving the load of MIR Pro 3D to the GPU of an M1/M2-based Mac computer (for completeness of information, the software also runs on Windows machines with NVIDIA and AMD graphic cards).
The test was conducted on a 2021 MacBook Pro, with M1 Pro chip and 16Gb RAM, running ProTools version 2023.6 and MIR Pro 3D version 7.2.3652.
The first step consisted of loading an orchestral session with 28 tracks and routing them to MIR Pro 3D using an instance of the dedicated plugin on each track. I then loaded the “Salzburg Festspielhaus - Grosses Haus” venue in MIR Pro 3D and moved the 28 orchestra sources to their “usual place” on stage. I assigned a dedicated role to each source so that the software might calculate the correct convolution for the various specific instruments in terms of positioning and frequency response.
In this first test, the comparison between CPU and GPU usage showed a very slightly improved performance when the GPU option was active, but the difference was not huge.
For the second test, I duplicated the initial 28 tracks two times, which caused the difference in CPU usage to decrease a few percent more when the GPU was activated.
For the third test, I duplicated the initial tracks another two times, for a total of 140 tracks. In this case the difference in CPU usage between the two options being investigated was about 10%.
At this point I had to stop adding tracks, because my system started to become unstable. Nevertheless, it was evident that the higher the number of tracks, the better was the performance in terms of CPU usage when the GPU option was activated.
Conclusion
Using the GPU as an extra resource for computing duties is a concept which has been around for some time, but in my experience it has never been taken seriously enough. The main criticism that I hear when talking about this possibility is that modern computers have so much CPU power that there’s no need to resort to GPU use. Well, that might be true, but in the wrong circumstances even a couple of percentage points can make the difference between being able to complete a task or not.