AI Chips · Product page

NVIDIA Vera Rubin: The Complete Guide

Seven chips, five rack systems, one supercomputer — NVIDIA’s own description. Shipping now, priced nowhere. Simple to expert, every number links to its source.

Ask an AI about this chip:CAGPX
NVIDIA Vera Rubin 10 sections · 2 levels 8 linked sources Checked September 2026
How to read this page. Each section starts Simple, then goes to a Deep dive: stop wherever you have what you need. The small numbers are sources: click one to open the original document. Where the maker has not published a figure -- a price, a die size, a factory address -- this page says so rather than estimate it.

1.At a glance

SimpleStart here

Vera Rubin is NVIDIA’s full-stack platform pairing its new Vera CPU with its Rubin GPU, sold rack-scale as Vera Rubin NVL72 — 72 Rubin GPUs and 36 Vera CPUs. NVIDIA declared it in full production on 31 May 2026 1.

Deep diveThe technical detail

NVIDIA’s own newsroom describes the platform as “seven breakthrough chips, five racks, one giant supercomputer” 2: Vera CPU, Rubin GPU, NVLink 6 Switch, ConnectX-9 SuperNIC, BlueField-4 DPU, Spectrum-6 Ethernet switch, and NVIDIA Groq 3 LPX, detailed separately in August 2026.

2.Launch and history

SimpleStart here

Vera Rubin moved from a roadmap slide to a shipping platform over roughly 14 months — first named at GTC 2025, expanded at CES and GTC 2026, and declared in full production by the end of May 2026.

Deep diveThe technical detail

First named 18 March 2025 at GTC, honouring astronomer Vera Rubin 3. A specialised inference-focused GPU variant, Rubin CPX, followed 9 September 2025 4. NVIDIA declared the Rubin platform in full production at CES on 5 January 2026 5, then formally launched Vera Rubin as the complete platform at GTC on 16 March 2026 2. The Vera CPU itself was unveiled 31 May 2026, the same day NVIDIA declared Vera Rubin in full production, with system availability from partners “starting this fall” (autumn 2026) 61. A scientific-computing variant followed at ISC High Performance on 22 June 2026 7.

3.What’s inside it

SimpleStart here

Each Vera CPU pairs with Rubin GPUs over a coherent link running at 1.8 TB/s; a full Vera Rubin NVL72 rack connects 72 Rubin GPUs and 36 Vera CPUs over sixth-generation NVLink 62.

Deep diveThe technical detail

The Vera CPU uses 88 custom NVIDIA “Olympus” cores, running 176 threads via NVIDIA Spatial Multithreading, with up to 1.5TB of LPDDR5X memory at up to 1.2 TB/s bandwidth 6. It connects to its paired Rubin GPU over NVLink-C2C at up to 1.8 TB/s 6. NVIDIA states Vera delivers “1.8x faster task completion compared with x86 CPUs” 6 and names it the direct successor to Grace, which NVIDIA says has shipped “nearly 2.5 million” units to date 6. A separate CPU-only configuration, a “Vera CPU Rack” of 256 Vera CPUs, is also referenced on NVIDIA’s platform page 2. The Rubin GPU itself is covered on its own page.

4.Full spec table

SimpleStart here

A Vera Rubin NVL72 rack holds 72 Rubin GPUs and 36 Vera CPUs; a Rubin CPX variant of the rack claims 8 exaflops of AI performance and 1.7 PB/s of memory bandwidth 4.

Deep diveThe technical detail
SpecVera CPUVera Rubin NVL72 (rack)
Cores88 custom “Olympus,” 176 threads72 GPUs + 36 CPUs
MemoryUp to 1.5TB LPDDR5XRubin GPU memory (see Rubin page)
Memory bandwidthUp to 1.2 TB/s—
CPU-GPU linkNVLink-C2C, 1.8 TB/s—
GPU interconnect—NVLink 6
Rubin CPX variant (rack)8 exaflops AI performance, 1.7 PB/s memory bandwidth

Vera CPU figures per NVIDIA’s Vera unveiling 6; Rubin CPX rack figures per NVIDIA’s Rubin CPX announcement 4. NVIDIA has not published a single consolidated Vera Rubin NVL72 spec sheet covering GPU-side FLOPS at the rack level in the sources checked; those figures, once published, will be added here rather than estimated.

5.Where it’s made

SimpleStart here

NVIDIA has not published where Vera or Rubin are manufactured, or on what process node. This page does not guess.

Deep diveThe technical detail

No NVIDIA source found — newsroom, developer blog, or investor material — names a foundry or process node for either the Vera CPU or the Rubin GPU as of September 2026. This matches the pattern for the whole Rubin generation: NVIDIA has disclosed architecture and performance details well ahead of manufacturing details.

6.Which systems use it

SimpleStart here

Cloud partners named in NVIDIA’s Vera Rubin full-production release include Microsoft Azure, Oracle Cloud, CoreWeave, Lambda, IBM Cloud, Firmus, GMI Cloud, IREN, Nebius, Nscale and Vultr 1; NVIDIA’s broader Vera Rubin platform materials additionally name AWS, Google Cloud, Crusoe and Together AI 2.

Deep diveThe technical detail

System builders and manufacturing partners named in the full-production release include Dell Technologies, HPE, Lenovo, Supermicro, Foxconn, Wistron, Wiwynn, Pegatron, Compal, Inventec, ASUS, GIGABYTE, MSI, Quanta/QCT, NetApp, VAST Data and WEKA 1; Cisco is named as an OEM partner in NVIDIA’s wider Vera Rubin platform materials 2. Science and HPC partners named for the scientific-computing variant include the Leibniz Supercomputing Centre (system named “Blue Lion,” targeted 2027), Lawrence Berkeley National Laboratory/DOE (“Doudna”), and Los Alamos National Laboratory (“Mission,” “Vision,” “Veritas”) 7.

7.Official pricing

SimpleStart here

No official NVIDIA pricing exists for Vera, Rubin or Vera Rubin at any level — chip, rack, or system — as of September 2026.

Deep diveThe technical detail

No NVIDIA source and no partner’s own pricing page publishes a Vera Rubin figure as of this check. Any dollar figures reported elsewhere for Vera Rubin hardware are analyst cost modelling, not an NVIDIA or partner disclosure, and are not repeated on this page.

8.Real-world performance

SimpleStart here

NVIDIA claims Vera Rubin delivers “up to a 10x reduction in inference token cost” versus Blackwell, and needs “one-fourth the number of GPUs” to train certain mixture-of-experts models 52 — company-stated comparisons, not independent benchmarks.

Deep diveThe technical detail

NVIDIA’s full-production release states Vera Rubin delivers “10x agent throughput at scale compared with the previous-generation NVIDIA Grace Blackwell platform” 1 — the closest NVIDIA comes to a direct platform-level comparison against GB300. The Rubin CPX rack variant separately claims “7.5x more AI performance and 3x faster attention” versus GB300 NVL72 specifically 4. As of September 2026, no independent MLPerf submission for the shipping Vera Rubin platform has been published.

9.What came before, what came next

SimpleStart here

Came before: Grace Blackwell (GB300 NVL72). Comes after: an architecture NVIDIA has named “Feynman,” paired with a new CPU codenamed “Rosa” 8.

Deep diveThe technical detail

NVIDIA’s Rubin platform page names Grace Blackwell as the generation Vera Rubin succeeds 5. A larger rack configuration — referred to at GTC 2025 as “Rubin Ultra” and a 144-GPU NVL144 rack — is targeted for the second half of 2027; the exact wording around its timing is not fully consistent across NVIDIA’s own GTC 2025 material, so this page treats it as an H2 2027 roadmap item pending a clearer official statement 3. Beyond that, NVIDIA named its next full architecture generation Feynman at GTC 2026, with no official year yet attached 8.

10.Hidden in plain sight

SimpleStart here

NVIDIA has now named two consecutive generations after women scientists: Vera Rubin, who found the first strong evidence for dark matter, and “Rosa,” the CPU codename honouring Rosalind Franklin.

Deep diveThe technical detail

NVIDIA’s own GTC 2025 keynote coverage states plainly: “Paying tribute to astronomer Vera Rubin, Huang outlined a roadmap…” 3. After years of naming data-center architectures after male scientists (Pascal, Volta, Turing, Ampere, Hopper, Blackwell), Vera Rubin is the first named for a woman — and NVIDIA confirmed at GTC 2026 that the next generation’s CPU is codenamed Rosa, after chemist Rosalind Franklin, whose X-ray diffraction work was critical to discovering the structure of DNA 8. Two in a row looks like a deliberate pattern, not a one-off tribute.

11.Sources

8 sources, checked September 2026. Where NVIDIA or another maker has not published a figure, this page says so rather than estimate it.

  1. NVIDIA Newsroom: NVIDIA Vera Rubin Ramps Into Full Production to Power Agentic AI Factories WorldwideOfficial
  2. NVIDIA Newsroom: NVIDIA Vera Rubin PlatformOfficial
  3. NVIDIA Blog: NVIDIA Keynote at GTC 2025: AI News, Live UpdatesOfficial
  4. NVIDIA Newsroom: NVIDIA Unveils Rubin CPX, a New Class of GPU Designed for Massive-Context InferenceOfficial
  5. NVIDIA Newsroom: NVIDIA Rubin PlatformOfficial
  6. NVIDIA Newsroom: NVIDIA Unveils Vera, the CPU for AgentsOfficial
  7. NVIDIA Newsroom: NVIDIA Vera Rubin Delivers World-Class Supercomputers for ScienceOfficial
  8. NVIDIA Blog: GTC 2026 NewsOfficial