# Nvidia Lowered Vera Rubin's Published Memory and NVLink Bandwidth, Calling the Original Figures Preliminary Targets

Times of AI Desk · 2026-11-02 · Infrastructure

[https://timesof.ai/2026/11/nvidia-vera-rubin-spec-sheet-hbm4-nvlink-bandwidth-lowered](https://timesof.ai/2026/11/nvidia-vera-rubin-spec-sheet-hbm4-nvlink-bandwidth-lowered)

> Nvidia's Vera Rubin NVL72 page now lists 19.2 TB/s of HBM4 bandwidth and 3 TB/s of NVLink per GPU, down from the 22 TB/s and 3.6 TB/s it advertised earlier. Rack totals and the HGX Rubin NVL8 figures fell too. Nvidia told Wccftech the earlier numbers were preliminary architectural targets and the new ones reflect what the system sustains in volume production, and that published third-party benchmarks already used production specs.

Nvidia has quietly lowered the headline bandwidth numbers for Vera Rubin, its next big AI system, and the change matters most to buyers comparing it on paper with rivals, because memory bandwidth is one of the main limits on how fast large models generate text. Nvidia's **Vera Rubin NVL72** product page now lists each Rubin GPU with **288 GB of HBM4** at **19.2 TB/s** and **3 TB/s** of NVLink bandwidth. When Nvidia [unveiled the platform at GTC in March](https://timesof.ai/2026/03/nvidia-vera-rubin-platform-for-agentic-ai-unveiled-at-gtc), it advertised **22 TB/s** and **3.6 TB/s**.

## What changed

Wccftech compared the old and new specifications on October 29. It reported these changes:

| Specification | Previous | Now listed |
|---------------|----------|------------|
| Rubin GPU HBM4 bandwidth | 22 TB/s | 19.2 TB/s |
| Rubin GPU NVLink 6 bandwidth | 3.6 TB/s | 3 TB/s |
| NVL72 rack HBM4 bandwidth | 1,580 TB/s | 1,400 TB/s |
| NVL72 rack NVLink bandwidth | 260 TB/s | 216 TB/s |
| HGX Rubin NVL8 HBM4 capacity | 2.3 TB | 2 TB |
| HGX Rubin NVL8 HBM4 bandwidth | 176 TB/s | 130 TB/s |

Wccftech also reported lower NVL8 training figures, 270 rather than 280 petaflops in NVFP4 and 130 rather than 140 in FP8. Scale-out networking bandwidth went up, for example to 32.4 TB/s from 28.8 TB/s on the NVL72. Dataconomy, which revisited the change on November 2, reported that archived snapshots place the page edits between September 1 and September 11.

## Nvidia's explanation

Nvidia did not announce the change. Asked by Wccftech, a representative said: "The specifications published during Vera Rubin's development were preliminary architectural targets. Final specifications on nvidia.com and the Vera Rubin datasheet reflect what the system sustains at qualified, volume production."

Nvidia also argued that "individual component specifications do not determine application performance," and that third-party agentic benchmarks, available for months, were measured using production specifications. It pointed buyers to SemiAnalysis's AgentX and Artificial Analysis's AgentPerf. It said Vera Rubin NVL72 is in full production and shipping.

Vera Rubin is central to Nvidia's growth plans. In August, Nvidia [forecast Vera Rubin at about 20% of its third-quarter data centre revenue](https://timesof.ai/2026/08/nvidia-q2-fy2027-96-billion).

## Limits

- The change first surfaced in reports around October 26 to 29, before this edition's window, and the page edits appear to date from September. We had not covered it.
- Nvidia has not said why the bandwidth fell. Dataconomy cites TrendForce reports that Nvidia asked memory suppliers for faster HBM4 than the JEDEC standard, but we have not seen Nvidia link the change to memory supply.
- Dataconomy reported on November 2 that the per-GPU table still showed 22 TB/s. When we checked the page on November 2, it showed 19.2 TB/s per GPU.
- How much the lower peak figures affect real workloads depends on software and system design. We have not seen independent tests comparing the two configurations.

## Sources

- [Nvidia: Vera Rubin NVL72 product page (accessed November 2, 2026)](https://www.nvidia.com/en-us/data-center/vera-rubin-nvl72/)
- [Wccftech: Nvidia's official statement on Vera Rubin spec change (October 29, 2026)](https://wccftech.com/nvidia-official-statement-on-vera-rubin-spec-change/)
- [Dataconomy: Nvidia quietly cuts Vera Rubin HBM4 and NVLink bandwidth specs (November 2, 2026)](https://dataconomy.com/2026/11/02/nvidia-quietly-cuts-vera-rubin-hbm4-and-nvlink-bandwidth/)