The 5-step pipeline.
How energy turns into a deployed, financed, offtake-matched cluster, and what SLYD does at every step.
How SLYD works →Platform
SLYD Cloud
New and recovered GPU systems.
NVIDIA and AMD systems through documented manufacturer and qualified channel supply, with financing and deployment coordinated on the same platform.
Explore GPU hardware →By accelerator
Infrastructure and services
Compute, hardware, and power in one book.
Browse available GPU capacity by accelerator, configuration, region, and price, or bring supply to qualified demand.
Open marketplace →Compute
Bring supply
For buyers
Start with an indicative structure.
Tell us deal size, structure, and offtake. Any range is preliminary and subject to underwriting, diligence, and documentation.
Open Configure →For lenders
GPU market trends and deployment playbooks.
Infrastructure best practices, hardware comparisons, and industry analysis from the SLYD team.
Read the blog →Company
NVIDIA Vera Rubin platform
A rack-scale platform is a facility decision before it is a purchasing decision. This page covers what NVIDIA publishes about Vera Rubin, what remains preliminary, and the site questions worth answering while the generation is still being planned for.
What is NVIDIA Vera Rubin?
NVIDIA Vera Rubin is a rack-scale AI platform that combines Rubin-generation accelerated computing with NVIDIA CPUs, networking, and system architecture. Buyers should evaluate the platform as a complete deployment, including power, cooling, networking, software, site readiness, availability, and the exact OEM or system configuration under consideration.
Use the names NVIDIA actually publishes
Next-generation platforms attract unofficial names long before launch, and those names end up in procurement documents where they cause real confusion. These are the names NVIDIA publishes today.
NVIDIA Vera Rubin NVL72
The rack-scale system NVIDIA publishes specifications against, configured as 72 Rubin GPUs and 36 Vera CPUs.
NVIDIA DGX Vera Rubin NVL72
NVIDIA's own turnkey build of the platform, published with its networking configuration and the DGX SuperPOD reference architecture.
NVIDIA Rubin GPU and NVIDIA Vera CPU
The accelerator and CPU that make up the platform. NVIDIA publishes individual specifications for the Rubin GPU alongside the rack-level figures.
SLYD does not publish an unofficial product name for this generation. If a supplier quotes a name NVIDIA does not publish, ask what specific part number and configuration it refers to.
What NVIDIA publishes today
Every figure below is published by NVIDIA and carries NVIDIA's own caveats. NVIDIA labels this specification set as preliminary information that is up to and subject to change, so it should be treated as planning input rather than as a contractual specification.
| Specification | Vera Rubin NVL72 | 2 Rubin GPUs and 1 Vera CPU | 1 Rubin GPU |
|---|---|---|---|
| Configuration | 72 Rubin GPUs, 36 Vera CPUs | 2 Rubin GPUs, 1 Vera CPU | 1 Rubin GPU |
| GPU memory and bandwidth | 20.7 TB HBM4, 1,580 TB/s | 576 GB HBM4, 44 TB/s | 288 GB HBM4, 22 TB/s |
| NVFP4 inference (sparse) | 3,600 PFLOPS | 100 PFLOPS | 50 PFLOPS |
| NVFP4 training (dense) | 2,520 PFLOPS | 70 PFLOPS | 35 PFLOPS |
| FP8 or FP6 training (dense) | 1,260 PFLOPS | 35 PFLOPS | 17.5 PFLOPS |
| FP16 or BF16 (dense) | 288 PFLOPS | 8 PFLOPS | 4 PFLOPS |
| NVLink | 260 TB/s, sixth generation | 7.2 TB/s | 3.6 TB/s |
| CPU cores | 3,168 custom NVIDIA Olympus cores, Arm compatible | 88 custom NVIDIA Olympus cores, Arm compatible | Not applicable |
| CPU memory | 54 TB LPDDR5X | 1.5 TB LPDDR5X | Not applicable |
| Scale-out networking bandwidth | 28.8 TB/s | 0.8 TB/s | 0.4 TB/s |
SLYD does not publish a performance comparison between this platform and earlier generations. NVIDIA publishes some figures with sparsity and others as dense, and mixing the two produces a number that means nothing. If a comparison matters to a decision, it should be run on the actual models and software stack in scope.
Where the platform stands
As of the review date on this page, NVIDIA states that Vera Rubin is ramping into full production, with server manufacturers building and shipping Vera Rubin-based systems.
That is NVIDIA's statement about NVIDIA's platform. SLYD does not hold, claim, or represent allocation for this generation, and no page on this site offers a reservation, pre-order, or delivery position for it. Whether a specific configuration can be obtained, in what quantity, for what geography, and on what terms is confirmed at the transaction level against current records.
Reviewed 18 August 2026. Platform status changes; confirm against NVIDIA's current materials before relying on it.
Facility questions to answer early
These questions have long lead times and they do not depend on which generation eventually gets deployed. Answering them now is useful whether the outcome is Vera Rubin, a current platform, or rented capacity.
How much power can the site actually deliver
Not nameplate capacity on paper but firm, permitted capacity at the meter, with a credible path to more if the deployment grows. Utility timelines are frequently the longest item in the plan. See power infrastructure.
Can the site support liquid cooling
Rack-scale platforms at this density are built around liquid cooling. Facility water loops, heat rejection, water quality, CDU placement, redundancy, and serviceability all need to exist before equipment arrives. See cooling infrastructure.
Will the rack physically fit and land safely
Integrated racks arrive heavy and preassembled. Floor loading, door and lift dimensions, delivery routes, and rigging access are checked against the exact rack, not a generic footprint.
What network fabric does the platform require
Scale-up bandwidth is internal to the rack, but scale-out fabric, optics, and cable routes are a site design. See AI cluster networking.
Who operates it once it is installed
Rack-scale systems change the operating model: firmware baselines, liquid-cooling maintenance, monitoring, and support paths need owners before the first job runs.
What to confirm before committing
- Exact platform, part number, and configuration being quoted
- Which specifications are preliminary and which are final
- Rack power, cooling method, and facility water requirements
- Included networking, optics, and cabling
- Software stack, licensing, and support scope
- Delivery and rigging requirements at the destination
- Who performs installation, commissioning, and acceptance
- Seller authority and governing transaction documents
- Export, end-use, and geographic restrictions
- Availability confirmation with its date and expiration
Vera Rubin questions
What is NVIDIA Vera Rubin?
Vera Rubin is NVIDIA's rack-scale platform pairing NVIDIA Rubin GPUs with NVIDIA Vera CPUs, sixth-generation NVLink, and NVIDIA networking. NVIDIA publishes it as a complete system rather than a single accelerator, so the platform, not the chip, is the unit of evaluation.
What are the official Vera Rubin product names?
NVIDIA publishes NVIDIA Vera Rubin NVL72 as the rack-scale system and NVIDIA DGX Vera Rubin NVL72 as the turnkey NVIDIA-built system, built from the NVIDIA Rubin GPU and the NVIDIA Vera CPU. Names circulating outside NVIDIA's own materials should be treated as unofficial until NVIDIA publishes them.
What does NVIDIA publish for Vera Rubin NVL72 specifications?
NVIDIA publishes a configuration of 72 Rubin GPUs and 36 Vera CPUs, 20.7 TB of HBM4 GPU memory at up to 1,580 TB/s, 260 TB/s of NVLink bandwidth, 54 TB of LPDDR5X CPU memory, and 3,168 custom NVIDIA Olympus cores. NVIDIA marks these as preliminary information that is up to and subject to change.
Is Vera Rubin available to order?
NVIDIA states that Vera Rubin is ramping into full production, with server manufacturers building and shipping Vera Rubin-based systems. That is a statement by NVIDIA about its own platform. It is not a statement that SLYD holds allocation, inventory, or a delivery position, and SLYD makes no such claim.
What does a Vera Rubin deployment require from a facility?
Rack-scale platforms at this density are decided by the building. The specific system configuration determines rack power, cooling method and facility water requirements, floor loading, network cabling paths, and delivery and rigging access. Those requirements must be confirmed against the exact OEM or NVIDIA configuration under consideration rather than estimated from the platform name.
Should I plan for Vera Rubin or deploy a current-generation platform now?
It depends on when capacity is actually needed and what the site can support today. Organizations with a live workload and a ready facility generally proceed with a currently shipping platform, while organizations building new capacity may design the facility so it can accept a later generation. Both paths need the same facility analysis.
How should Vera Rubin performance figures be compared to earlier platforms?
Carefully, and with the basis attached. NVIDIA publishes some figures with sparsity and others as dense, and it labels Vera Rubin specifications as preliminary. Comparing a sparse figure on one platform to a dense figure on another produces a meaningless result, so the precision, sparsity basis, and publication date should travel with every number.
Sources for the specifications on this page
- NVIDIA, Vera Rubin NVL72 product page. Reviewed 18 August 2026. nvidia.com/en-us/data-center/vera-rubin-nvl72
- NVIDIA, DGX Vera Rubin NVL72 product page. Reviewed 18 August 2026. nvidia.com/en-us/data-center/dgx-vera-rubin-nvl72
- NVIDIA, Rubin platform technology page. Reviewed 18 August 2026. nvidia.com/en-us/data-center/technologies/rubin
Specifications are NVIDIA's, reproduced with NVIDIA's own caveats. They are not SLYD guarantees of performance, configuration, or availability.
Plan for the platform generation you will actually deploy
Share the workload, timing, site, and constraints. The facility work is the same whether the answer is this generation, a currently shipping platform, or rented capacity.
Page updated: August 18, 2026