AMD has released the TrueAudio Next 1.2 open source library version upgrade. An update that has several notable performance and feature improvements, and reflects the improvements built into the version of TAN supported in Steam Audio.
First, the audio convolution algorithm includes a significant acceleration option called the “head-to-tail” partition method. This arrangement allows an audio wire to send a real-time audio buffer to the TAN convolution to receive a much faster response than with conventional partitioned convolution.
AMD announces TrueAudio Next 1.2
Most of the computational overload occurs in the background, between buffer sends to TAN, so this method is very friendly for parallel processing. At the same time, it provides a significant reduction in latency and performance acceleration, as the audio wire making the call is not blocked waiting for the full convolution to be calculated.
Other optimizations were also incorporated into the TAN Graphics audio acceleration library to minimize memory usage, buffer transfer, and sync overload. In particular, there is a significant improvement in performance when the IR kernel is dynamically updated while running convolution.
The TAN GPU Utilities Library supports AMD Resource Reservation, where a configurable part of the GPU can be reserved for audio processing, in addition to normal GPU computing resources. Resource pool protects audio and graphics queues and computes resources lock on to each other, allowing them to coexist on the GPU like never before.