Friday, June 26, 2026
EN·DarkSubscribe
AI Infrastructure · News & Analysis
HomeChips & HardwareReport
Chips & Hardware · Report

NVIDIA announced next-generation Rubin and Blackwell Ultra GPUs along with supercharged Vera CPUs.

New GPU and CPU product lines define the AI-chip roadmap for the next generation of compute infrastructure.
Trade pressSlicast · June 2, 2024 · Global · Source: wccftech.com
importance 85

NVIDIA has confirmed its next-generation Rubin GPU architecture, accelerating its product roadmap to deliver a new GPU offering each year. The Rubin architecture is named after American astronomer Vera Rubin, who made significant contributions to the understanding of dark matter in the universe and pioneered work on galaxy rotation rates. Alongside Rubin, NVIDIA announced upgrades to its Blackwell GPUs and introduced the latest Vera CPU.

Blackwell GPUs, specifically the B100 and B200 iterations, will arrive in data centers later this year. NVIDIA also plans to release a supercharged Blackwell variant featuring 12Hi memory stacks across 8 sites, compared to the 8Hi memory stacks on existing products, expected to launch in 2025. Following this release, NVIDIA's next-generation Rubin R100 GPUs will enter mass production in the fourth quarter of 2025, with DGX and HGX systems expected to begin mass production in the first half of 2026. The Rubin platform will be available by 2026, followed by an Ultra version in 2027.

The Rubin R100 GPUs will utilize a 4x reticle design, compared to 3.3x for Blackwell, and will be manufactured using TSMC's CoWoS-L packaging technology on the N3 process node. NVIDIA is planning future versions using a new SoIC design featuring greater than 8x reticle size in a 120x120mm package configuration with up to 12 HBM sites, versus 8 HBM sites on current 80x80mm packages. However, the R100 GPUs are more realistically expected to utilize a 4x reticle size. The Rubin GPUs will leverage next-generation HBM4 memory, with Samsung and SK Hynix commencing development in 2025 featuring up to 16-Hi stacks, with mass production expected in late 2025.

NVIDIA is upgrading its Grace CPU architecture for the GR200 Superchip module, which will house two R100 GPUs alongside an upgraded Grace CPU based on TSMC's 3nm process node. The current Grace CPU is built on TSMC's 5nm process and packs 72 cores, totaling 144 cores on the Grace Superchip solution. The next-generation ARM CPU is confirmed to be known as Vera. One of NVIDIA's primary focuses for the Rubin R100 GPUs is power efficiency, with the company aiming to provide significant improvements while increasing AI capabilities. The R100 GPUs are expected to be unveiled at next year's GTC conference.

Read the original
NVIDIA announced next-generation Rubin and… · Slicast