Nvidia's GB300 Blackwell Ultra "desktop superchip" introduced by Asus for workstations: A standard desktop boasting power surpassing numerous server racks, capable of up to 784GB of coherent memory, offering a staggering 20 PFLOPS AI performance.
Nvidia's GB300 Blackwell Ultra: A Desktop Superchip Revolutionizing AI Compute
Nvidia has unveiled its latest AI-focused powerhouse, the GB300 Blackwell Ultra, a desktop superchip that combines the Nvidia Grace ARM-based CPU with the Blackwell Ultra GPU. This innovative technology promises up to 20 petaflops (PFLOPS) of AI performance and supports up to 784 GB of unified coherent LPDDR5X and HBM3E memory.
The GB300 Blackwell Ultra is not just a step forward for Nvidia; it's a leap into a new era of AI computing. The desktop workstation offers significant expansion with three PCIe x16 slots, three M.2 SSD slots, and powerful GPU power delivery capable of up to 1,800W. It also features next-generation Tensor Cores with enhanced FP4 precision and supports Nvidia DGX OS, alongside high-speed networking with ConnectX-8 SuperNIC (800 Gb/s). Pricing is expected to be in the five-digit USD range, reflecting the cost of comparable SXM-based GPUs that cost tens of thousands of dollars.
In the server space, Nvidia's GB300 NVL72 system includes 72 Blackwell Ultra GPUs, 36 Grace CPUs, and 18 BlueField-3 DPUs, with 21 TB of HBM3e memory per rack and an aggregate NVLink bandwidth of 130 TB/s. This setup supports ultra-fast GPU-to-GPU communication and high-bandwidth 800 Gb/s InfiniBand networking, delivering up to 10x better responsiveness, 5x throughput per watt over prior Hopper architectures, and 50x higher inference output for reasoning models.
Nvidia's strategic move to proliferate advanced AI hardware via partnerships with OEMs like Asus, Dell, and Lambda is also noteworthy. Asus has launched the ExpertCenter Pro ET900N G3 desktop workstation, powered by the GB300 Grace Blackwell Ultra, aiming to bring this cutting-edge AI compute to broader enterprise and pro markets. These collaborations enable AI-focused workstations to reach various customer segments, not limited to Nvidia’s direct offerings.
The GB300 Blackwell Ultra workstation and associated partnerships mark a significant shift in Nvidia's approach to AI hardware, signalling the company's growing ambitions beyond GPUs. The ExpertCenter Pro ET900N G3 is more than just another AI workstation; it signals Nvidia's intent to bring its AI hardware into more accessible form factors. The Ascent GX10, a compact desktop based on the smaller GB10 Grace Blackwell platform, further underscores this intent.
Grace enables unified, high-bandwidth memory and compute performance that traditional setups can't easily match. While Grace isn't designed to replace general-purpose CPUs, it complements Nvidia's GPU dominance, edging into the CPU landscape with the ARM-based Grace processor, built specifically for AI and HPC workloads.
Volume shipments of GB300 servers are expected to ramp up by September 2025. This collaboration could gradually erode the share in AI-focused servers and workstations of AMD and Intel, but for now, it doesn't put Nvidia in a head-to-head battle with these competitors. The GB300 Blackwell Ultra is a testament to Nvidia's commitment to driving the AI revolution forward, making AI compute more accessible and efficient than ever before.
| Specification / Feature | Details | |---------------------------------|----------------------------------------------| | CPU + GPU | Nvidia Grace ARM CPU + Blackwell Ultra GPU | | Max Memory | Up to 784 GB unified LPDDR5X + HBM3E | | AI Performance | Up to 20 PFLOPS AI in desktop form | | Expansion | 3 PCIe x16 slots, 3 M.2 SSD slots | | Power Delivery | Up to 1,800W to GPU | | Networking | ConnectX-8 SuperNIC (800 Gb/s) | | Pricing | Estimated five-digit USD range | | OEM Partners | Asus (ExpertCenter Pro ET900N G3), Dell, Lambda |
[1] TechCrunch: Nvidia's new Grace CPU is designed for AI and HPC workloads
[2] AnandTech: Nvidia's Grace CPU and Blackwell GPUs are here for HPC and AI
[4] Tom's Hardware: Nvidia's Grace CPU and Blackwell GPU: The Future of AI and HPC
- The GB300 Blackwell Ultra, Nvidia's latest AI-focused powerhouse, is not only revolutionizing AI compute but also redefining the role of technology in the field of data-and-cloud-computing, as it incorporates advanced AI technology, cutting-edge Tensor Cores, and high-speed networking.
- With the introduction of AI-focused gadgets like the GB300 Blackwell Ultra desktop superchip, artificial-intelligence is no longer confined to data centers and servers; it's becoming accessible to a broader range of customers, including enterprises and professionals, thanks to Nvidia's strategic partnerships with OEMs like Asus, Dell, and Lambda.
