Intel oneAPi software supports multi-GPUs
In recent years, Nvidia and AMD have been talking less and less about SLI and CrossFire, and the new generation of consumer graphics cards will hardly see multi-GPU support. The reason is that these technologies provide limited scalability and frame rate improvement in games, and require the support of game engines and APIs. Developers have to spend more energy and time to complete these tasks for a small user base. At the same time, the cost of player construction will also be greatly increased, and the greater power consumption and heat generation will only result in limited application and game support.
According to Igor’s Lab, at the SIGGRAPH 2022 held recently, Intel showed its ARC graphics cards and Arc Pro series workstation graphics cards. Some staff said that Intel may support multiple GPUs through oneAPI. Intel didn’t show a multi-GPU solution at this event because the systems currently available for demonstration use NUC chassis and can only install one graphics card.
oneAPI is a cross-architecture programming tool designed to simplify programming across GPUs, CPUs, FPGAs, and AI accelerators, and can be used with Intel’s own devices, or chips from other vendors, to optimize workloads. Intel staff said that the current multi-GPU support of oneAPI is based on consumer-grade ARC graphics cards, but Arc Pro series workstation graphics cards will also be supported in the future.
It is easy to understand that the workstation graphics card supports multiple GPUs, and NVIDIA will do so now. For example, two RTX A6000s use NVLink technology to achieve interconnection. It’s unclear how well Intel uses oneAPI to optimize multiple GPUs, and the conditions for implementation, but it’s always a good thing to be able to provide support.
With the further promotion of ARC graphics cards, perhaps Intel will announce more details in the next few months.
