NVIDIA GSP: good for servers, yes, but also for our laptops

Installed on NVIDIA chips since the Turing generation, but unused until now, the GSP (or GPU System Processor) has just been activated by NVIDIA. This small element of the GPU should help improve performance, mainly on laptops
and servers.

In Linux documentation, NVIDIA discreetly announces that it has unlocked a new feature for its professional graphics cards: the GPU System Processor. As explained Tom’s Hardware, this small element present on NVIDIA chips had not yet been exploited. It functions as a coprocessor and makes it possible to offload the CPU from the tasks relating to the initialization and management of the GPU, to manage them directly on… the GPU.

The GSP unlocked by NVIDIA

We learn that this novelty has been unlocked through the latest NVIDIA drivers (version 510.39.01 or later) on all models ranging from the Tesla T4, based on Turing, to the NVIDIA A series, such as the A100 , A2 and A40.

The use of GSP should in particular result in reducing the overall latency of the system, and therefore improving performance, promises NVIDIA. The firm nevertheless admits that this functionality is still limited at this stage, but it should become more important in the future. It will also help to reduce CPU load.

A feature reserved (for the moment) for professional cards…

As indicated above, NVIDIA currently reserves the GSP functionality for professional graphics cards, used on certain laptops, workstations and servers. It would be that said surprising that the firm deprives its consumer references of the GSP and the optimization of the performances that it allows. NVIDIA also ensures that this function will be central for Max-Q laptops, planned for 2022 and beyond.

Still, the chameleon brand does not yet give precise data as to the performance gains expected with the activation of the GSP. Reducing the CPU workload, however, should have a relatively noticeable impact on computing and gaming, especially since recent games, thanks to their low-level APIs, points out Tom’s Hardware, are increasingly taking advantage of CPU power.

Source: Tom’s Hardware