Nvidia is hyping up its new Vera Rubin chip system this week, revealing new efficiency benchmarks for the GPU and CPU combo forward of rival AMD’s annual product occasion in San Francisco on Thursday.
Throughout a prolonged technical workshop final week on the firm’s headquarters in Santa Clara, California, Nvidia executives boasted to a small group of journalists concerning the chip system’s elevated energy and effectivity capabilities. The largest takeaway: Nvidia, which has lengthy specialised in making GPUs, is more and more making an attempt to place itself as a provider of CPUs that may energy AI brokers.
Whereas GPUs are nonetheless the main hardware that firms use to coach and run their AI fashions, the trade’s shift towards extra complicated, agentic techniques has elevated demand for CPUs, which might orchestrate knowledge flows, networking, and different software program duties. That is doubtless one motive Nvidia has been keen to advertise itself as a provider of full AI techniques reasonably than simply AI chips.
Vera Rubin is Nvidia’s successor to its hybrid superchip system Grace Blackwell and represents the linchpin of its near-term future powering the AI trade. It’s designed to supply one CPU for each two GPUs. In a single Vera Rubin NVL 72 tremendous chip system, there are 36 Vera CPUs for each 72 Rubin GPUs. Nvidia can also be promoting the Vera CPU as a stand-alone product, and it has reportedly told Chinese customers these may very well be prepared as quickly as August.
.jpg)
Nvidia executives emphasised that its new Vera Rubin NVL72 racks—a stack of chips packed right into a single liquid-cooled platform—are rather more “plug-and-play” than a few of its earlier merchandise. Throughout a quick tour of a Nvidia knowledge middle lab in Silicon Valley, Nvidia executives shared that OpenAI already has one Vera Rubin rack in use.
Nvidia CEO Jensen Huang didn’t make an look on the workshop in Santa Clara final week; he was in Japan asserting the chipmaker’s new partnerships with a number of Japanese firms to develop AI for robotics. The briefings had been as an alternative led by Ian Buck, Nvidia’s longtime vp of accelerated computing and the architect behind the corporate’s CUDA software program.
“We’re on a street map to crank out new architectures, not simply GPUs however CPUs,” Buck informed reporters. “We’re going to maintain innovating, as a result of it’s do that or die in Silicon Valley.”
The conferences had been held in Huang’s govt briefing middle, the place a number of desks close by had been piled with baggage of Taiwanese snacks that the CEO introduced again from his latest journey to Computex, an enormous annual semiconductor commerce present in Taipei, an Nvidia spokesperson informed WIRED.
