NVIDIA Multi-Instance, the GPU Computing Clustering Technology to Come with Ampere

About two weeks ago NVIDIA introduced the Ampere architecture for the professional segment. This architecture will be the basis for the next gaming solutions, which should take a big leap in performance. The new NVIDIA architecture, among other innovations, will implement Multi-Instance technology. Curiously, this technology is currently under development by AMD, but NVIDIA will be the first to integrate it.

Multi-Instance offers greater release and allocation of the resources available on the graphics card. This technology allows the resources of a graphics card to be divided into clusters of up to 7 instances. It allows clusters to work independently and simultaneously, thus seeking to maximize performance, which will benefit software developers.

NVIDIA will implement Multi-Instance technology in Ampere

This technology at the moment would only be available in NVIDIA graphics, but it aims to be the future in graphics. Parallelizing the work allows to reduce the time of the tasks and thus obtain more performance. Its implementation in the Ampere architecture will allow the company to gain an advantage over its competition, AMD.

It appears that AMD's solution will be more complete, as NVIDIA's solution would still be premature. AMD would offer greater granularity with respect to the allocation of available resources. Basically there will be no limitation in the allocation of the physical resources of the GPU. AMD's solution allows a container to increase or decrease the number of Shaders as needed, operating dynamically. NVIDIA in this sense is more strict and will not offer as much flexibility.

AMD's solution enables full use of GPU resources, maximizing efficiency and power usage. In addition, the advantage over the NVIDIA solution is that it supports queues from different operating systems. Each container in this case will be assigned an ID, either physically or virtually, allowing the generation of multiple queues.

While NVIDIA seems to be implementing it in Ampere for the professional segment and possibly for the gaming segment, we don't know much about AMD's solution. We do not know if it was integrated into RDNA2, which would improve performance in games, especially in Ray Tracing or if he stayed at CDNA for HPC. It could be a great solution for integrating Ray Tracing, a resource-intensive technology.

Source: CT