NVIDIA Rubin GPU · Rubin

BF16, dense

4,000 TFLOPS FP16/BF163 288 PFLOPS 8 PFLOPS 16 PFLOPS NVIDIA Vera Rubin datasheet (Oct 2026) (8 PFLOPS on the page; read from a table with a column for each product, whose head reads “Configuration 72 NVIDIA Rubin GPUs | 36 NVIDIA Vera CPUs 2 NVIDIA Rubin GPUs 1 NVIDIA Vera CPU”; divided by the 2 chips the figure is for)

FP8, dense

17,500 TFLOPS FP8/FP6 Training3 1,260 PFLOPS 35 PFLOPS 70 PFLOPS NVIDIA Vera Rubin datasheet (Oct 2026) (35 PFLOPS on the page; read from a table with a column for each product, whose head reads “Configuration 72 NVIDIA Rubin GPUs | 36 NVIDIA Vera CPUs 2 NVIDIA Rubin GPUs 1 NVIDIA Vera CPU”; divided by the 2 chips the figure is for)

FP4, dense

35,000 TFLOPS NVFP4 Training 2,520 PFLOPS 70 PFLOPS 140 PFLOPS NVIDIA Vera Rubin datasheet (Oct 2026) (70 PFLOPS on the page; read from a table with a column for each product, whose head reads “Configuration 72 NVIDIA Rubin GPUs | 36 NVIDIA Vera CPUs 2 NVIDIA Rubin GPUs 1 NVIDIA Vera CPU”; divided by the 2 chips the figure is for)

Memory

288 GB 20.7 TB HBM4 | 1,400 TB/s 576 GB HBM4 | 38.5 TB/s NVIDIA Vera Rubin datasheet (Oct 2026) (read from a table with a column for each product, whose head reads “Configuration 72 NVIDIA Rubin GPUs | 36 NVIDIA Vera CPUs 2 NVIDIA Rubin GPUs 1 NVIDIA Vera CPU”; divided by the 2 chips the figure is for)

Memory bandwidth

19.3 TB/s 20.7 TB HBM4 | 1,400 TB/s 576 GB HBM4 | 38.5 TB/s NVIDIA Vera Rubin datasheet (Oct 2026) (read from a table with a column for each product, whose head reads “Configuration 72 NVIDIA Rubin GPUs | 36 NVIDIA Vera CPUs 2 NVIDIA Rubin GPUs 1 NVIDIA Vera CPU”; divided by the 2 chips the figure is for)

Figures come from the Vera Rubin Superchip column (2 Rubin GPUs and 1 Vera CPU), so each is for 2 GPUs (per: 2). The NVFP4 Training row has no sparse/dense mark; it is taken as dense like the other training rows. The HGX page lists one Rubin GPU at 288 GB and 22 TB/s, more than half the superchip's 38.5 TB/s; that page's text holds HTML markup, so it is not quoted. No power figure is published. NVIDIA said 'in full production' on 2026-01-05 but set shipments for fall 2026; 'racks running at partners' (2026-08-26) is the first statement of systems at customers. No dated 2024 or 2025 NVIDIA page found gives a clear window for the Rubin GPU itself.

1. 5 Jan 2026 Promised Rubin-based products available from partners in the second half of 2026 in the second half of 2026. NVIDIA Rubin is in full production, and Rubin-based products will be available from partners the second half of 2026. NVIDIA Kicks Off the Next Generation of AI With Rubin
2. 31 May 2026 Promised production shipments of Vera Rubin to begin this fall in the second half of 2026. Production shipments of Vera Rubin are set to begin starting this fall. NVIDIA Vera Rubin Ramps Into Full Production to Power Agentic AI Factories Worldwide
3. 26 Aug 2026 Shipped: Vera Rubin ramping into full production, with racks running at partners. Announced the NVIDIA Vera Rubin platform is ramping into full production with racks running at partners including CoreWeave, Google Cloud, Microsoft Azure, Oracle Cloud Infrastructure and Nebius. NVIDIA Announces Financial Results for Second Quarter Fiscal 2027

Epoch AI’s names for this chip: Rubin.

• Tesla V100 SXM2 417 gigaflops of dense FP16 compute per watt, at 300 W
• A100 SXM 80GB 780 gigaflops of dense BF16 compute per watt, at 400 W
• H100 SXM 1,414 gigaflops of dense BF16 compute per watt, at 700 W
• H200 SXM 1,414 gigaflops of dense BF16 compute per watt, at 700 W
• GH200 Grace Hopper Superchip 990 gigaflops of dense BF16 compute per watt, at 1,000 W
• B200 2,200 gigaflops of dense BF16 compute per watt, at 1,000 W
• GB200 Grace Blackwell Superchip 5,000 teraflops of dense BF16 compute a chip; NVIDIA publishes no power figure
• B300 (Blackwell Ultra) 2,000 gigaflops of dense BF16 compute per watt, at 1,100 W