THE METAVERSE IS COMING. Whether it revolutionizes the world or struggles to gain traction, there’s one thing experts, especially those at Intel, agree on: Every person in the metaverse will need far more compute power than they have today. How much more? 1,000x. Per person.
How will we get there? By repurposing some of the tools we have today, says Jon Bach, president of Puget Systems, a custom builder of workstations for creators located in Auburn, Wash.
Gaming PCs were developed to render the best possible video and sound environments for game play. Those same tools may soon power virtual reality applications and eventually the metaverse.
Raja Koduri, Intel’s senior vice president and general manager of the accelerated computing systems and graphics group, outlined a plan for powering the metaverse last December in a post and demo video. The goal is to “decouple user experience from the confines of specific hardware,” he said. In other words, share resources across a network.
Intel’s “continual compute” initiative, which aims to let multiple PCs share an external GPU over the network, is part of its strategy for realizing that goal. At the company’s Vision 2022 event in May, Koduri conducted a demo during which a standard laptop played a modern game with the expected results of lower video definition and jerky movements. When the continual compute infrastructure engaged, the laptop used the graphic resources of a gaming PC across the stage, resulting in full 4K graphics and smooth flying and rendering. Koduri claimed this level of performance will be the lowest bar for the metaverse in four or five years.
Jon Bach
The same tools can power artificial intelligence programs and neural networks, he added. “A new neural net needs to double its compute power every three to four months,” Koduri said.
Though the continual compute infrastructure draws on locally networked gaming PCs, he added, it could utilize resources across the cloud in the future.
For some time, vendors have worried that workstations will disappear because of the cloud, but that’s not happened yet, Bach notes. “Lately, the move has been to the edge of the cloud, and this makes workstations and gaming PCs edge cloud engines.” The more you can share their resources, he adds, the better.
“Gaming PCs and graphics workstations sit idle much of the time. If I need resources, and the PC in the next cubicle is idle, I want to share those resources seamlessly,” Bach says. In some ways, this is like Hollywood rendering farms, where dozens or hundreds of clustered graphics workstations with high-powered GPUs control the film rendering process.
Intel is expected to release four new discrete GPUs in the Arch Alchemist line for consumers later this year, each able to share resources. Bach has one in his test lab now and says, so far, “it’s not too shabby for generation one. It’ll be good to have new competition to keep NVIDIA honest.”
Being able to use existing Xeon graphics APIs on the new Arch Alchemist GPUs will speed adoption, according to Intel. But hardware isn’t the issue as much as software written to support clustering and rendering, says Bach.
He needs system agents for every client, and the agent will need to provide auto discovery, automatic connections, and the ability to share resources. “We’ll preinstall [agent software] on every system. Maybe Intel will make it free with each of their GPU boards.”
Intel says its goal is a persistent and immersive internet to power the metaverse. Turning gaming PCs into edge rendering nodes and sharing those resources is the first step.
“I think Intel’s serious,” concludes Bach. “They can drive their GPU sales, and gain market share,” especially with the right continual compute software.
Image: iStock
