Intel has been working on a unique technology that allows you to boost the GPU (read: gaming) performance of your thin laptop using a powerful desktop PC. Dubbed as a “continual compute” demonstration, this is achieved by taking the concept of an eGPU, and pairing it with a remote PC via a high-speed network connection. Presented by Raja Koduri (Head of Intel’s Graphics Division) at the Realtime Conference on Monday, it is being marketed as a concept of “the metaverse” as an opportunity to sell more products.
The demo was running Hitman 3 on a 14″ ultrabook using the GPU of a local PC to offload some of the graphics workload. With the stock configuration, the notebook chokes under load, with both the CPU and GPU hitting their limits. However, with the remote PC (and its GPU) connected, the game runs at a higher visual quality and much smoother, too.
The interesting part about this demo was the software rather than the hardware. Multi-GPU systems have long struggled with evenly balancing the graphics workload between them in order to get maximum performance. How Intel managed to do this between two GPUs physically apart is rather remarkable. The fact that the two are completely different SKUs makes it even more impressive.
This is precisely what Intel’s infrastructure layer delivers. It sends us additional compute resources available within the network and intelligently allocates it to me delivering the best user experience possible. As the game launches at runtime, the infrastructure layer determines that a better experience is possible using ambient computing resources from my gaming rig,” the narrator continues. The improvements are part of what Intel calls system resource abstraction where the game’s file system is being abstracted and delivered over the network.
I reckon it is some form of DirectX 12’s “Explicit Multi-GPU” functionality or something very similar. According to Intel, this is achieved using an abstraction layer that can detect the presence of a more powerful PC, thereby leveraging it for demanding workloads.