NVIDIA has quietly unlocked a feature that has been hidden since the Turing architecture: the GPU System Processor (GSP), which improves performance and efficiency by offloading drivers from the CPU to the GPU. This feature was originally only open to enterprise GPUs, but new notebook GPUs (GeForce RTX 3080 Ti / 3070 Ti) will start to support in 2022.
The GPU system processor is a physical block on the die that can be thought of as a co-processor paired with an NVIDIA GPU. It can communicate with the CPU through a low-level API, and transfer the work originally responsible for the initialization and management tasks of the CPU to the GPU, which not only reduces the overall system delay and improves system performance, but also improves execution efficiency.
NVIDIA explained how to maximize the efficiency between the CPU and GPU when it disclosed the technical details of the fourth-generation Max-Q earlier. NVIDIA refers to this function as “Command Pocessor” here, but foreign media Tom’s Hardware believes that it is essentially a GPU System Processor.
NVIDIA said that the functions of the GPU system processor have not been fully liberated, and will be gradually expanded through driver updates in the future. Since the notebook version of GeForce RTX 3080 Ti / 3070 Ti GPU adopts the fourth-generation Max-Q technology and has been confirmed to support GPU system processors, we will continue to pay attention to whether this feature will be implemented in other existing Ampere architecture or Turing architecture desktop GPUs Unlock.
.
