NVIDIA has officially announced its next-generation Rubin GPU architecture, which includes the Rubin and Blackwell GPUs alongside the latest Vera CPU. The new lineup is inspired by the work of American astronomer Vera Rubin, renowned for her groundbreaking contributions to the study of dark matter and galaxy rotation rates.
NVIDIA’s Future GPU Roadmap: A Yearly Product Refresh
NVIDIA’s GPU roadmap reveals an aggressive strategy, with new launches slated for each year. The Blackwell GPUs, with their B100/B200 iterations, will be introduced to data centers later in the year. The following year, 2025, will see the debut of Blackwell Ultra, which boasts 12Hi memory stacks across eight sites, in contrast to the existing 8Hi memory stacks.
Following Blackwell, the Rubin GPUs will make their entrance. Production of the NVIDIA Rubin R100 GPUs is set to ramp up in the fourth quarter of 2025, with mass production of systems including DGX and HGX solutions expected in early 2026. Moving forward, NVIDIA’s plans include the release of Rubin GPUs by 2026 and the Rubin Ultra version in 2027.
The Evolution of NVIDIA’s GPU Designs
NVIDIA’s Rubin R100 GPUs will push the boundaries of design with a 4x reticle size, utilizing TSMC’s CoWoS-L packaging technology and the N3 process node. TSMC has further plans for even larger chips, up to 5.5x reticle size by 2026, accommodating more HBM sites than current offerings. NVIDIA may eventually adopt a new SoIC design—projections are towards a design greater than 8x reticle size in a bigger 120x120mm package.
Next-Generation HBM4 DRAM and CPU Upgrades
On the memory front, NVIDIA is set to incorporate next-generation HBM4 DRAM into its R100 GPUs, with Samsung and SK Hynix gearing up to start development of the advanced memory solution by 2025. HBM4 will feature up to 16-Hi stacks, facilitating an upgrade from the HBM3E memory currently used in NVIDIA’s B100 GPUs.
The Grace CPU is also due for a significant upgrade. The GR200 Superchip module will host a pair of R100 GPUs and an enhanced Grace CPU leveraging TSMC’s 3nm process node—stepping up from the current 5nm node and 72 core setup of the existing Grace CPU.
Power Efficiency and AI: NVIDIA’s Emphasis for Rubin R100 GPUs
In response to the increasing power demands of data center chips, NVIDIA is focusing heavily on improving power efficiency for the next-gen Rubin R100 GPUs. These advancements are expected to enhance the AI capabilities of NVIDIA’s chips. Although the R100 GPUs are not anticipated to be unveiled until the forthcoming GTC next year, the range of developments NVIDIA has in the pipeline promises to deliver innovative solutions for the AI and data center industry.
Advancing NVIDIA’s Legacy in AI and Data Center Technology
NVIDIA continues to evolve its GPU offerings to meet the needs of a rapidly developing industry. With plans spanning from the enhanced Blackwell Ultra set for release in 2025 to the Rubin Ultra in 2027, NVIDIA is poised to maintain its dominant position in AI and data center applications.
The new Vera CPU, named to honor Vera Rubin’s legacy, rounds out NVIDIA’s introduction of momentous technology updates, signaling a future where their products not only bear the names of astronomical pioneers but possibly reflect the scope of their universe-spanfning work. Keep an eye on future GPU and CPU developments as NVIDIA charts its course in the high-stakes world of tech innovation.
