The era of Vera Rubin is upon us, as not only NVIDIA's but the world's fastest AI platform is now being delivered to top cloud providers.
Coreweave & Oracle Receive The World's First NVIDIA Vera Rubin NVL72 Systems For Validation, Bringing A Mammoth Leap To Agentic AI
NVIDIA's Vera Rubin is in volume production, but the top AI cloud providers are already getting their hands on the first systems to test and validate. NVIDIA also sent out its first Vera CPUs to major AI firms last month.
Oracle and Coreweave are among the first cloud providers to bring up NVIDIA's Vera Rubin NVL72 platform. These platforms will be tested closely with NVIDIA to deliver next-gen AI performance at the Frontier.
EVP of Oracle Cloud Infrastructure, Mahesh Thiagarajan, was the first to share the pictures of the massive NVL72 rack featuring 72 Rubin GPUs and 36 Vera CPUs. These are world-class hardware components (GPU & CPU) for AI, and aim to become the next big success story for NVIDIA after Grace Blackwell, which continues to shatter all AI records as demonstrated yesterday in MLPerf 6.0 results.
As Jensen mentions, AI is a five-layer cake, and this Vera Rubin system is just one component. The system is backed by precision-made infrastructure to drive its power, cooling, interconnects, net-working, and more. Then there's the software stack, which NVIDIA has built over several years to the point where no other company has yet replicated the same level of adoption or success. It's called CUDA, and for AI, it's CUDA-X.
The second system was installed at CoreWeave and was presented by the Senior Director of Compute Architecture, Jacob Yundt. CoreWave shows the full video of how the system was offloaded from the truck, and required 3-4 persons to get the unit into the facility. Just goes to show how massive a single rack is, now imagine hundreds of thousands of these racks inside AI datacenters.
CoreWeave has made some interesting name choices for its software-defined liquid-cooling stack and Unified Rack Control, Valvey, and Racky, respectively. As per the cloud provider, they are the first to bring up and validate the NVIDIA Vera Rubin NVL72 platform.
As said earlier, NVIDIA's Vera Rubin platform is already off to a grand start, offering the MoE training speed using just 1/4 the GPUs and inferencing with 1/10th the cost-per-token versus Blackwell. The full volume production is now underway, and Q3 will mark the beginning of the very first operational routines for this Agentic AI powerhouse.
About the author: A Software Engineer by training and a PC enthusiast by passion, Hassan Mujtaba serves as Wccftech's Senior Editor for hardware section. With years of experience in the industry, he specializes in deep-dive technical analysis of next-generation CPU and GPU architectures, motherboards, and cooling solutions. His work involves not only breaking news on upcoming technologies but also extensive hands-on reviews and benchmarking.
Follow Wccftech on Google to get more of our news coverage in your feeds.
