AIThis post was created with the assistance of artificial intelligence (AI).

🔍 Read the full analysis: Nvidia Rubin HPC: 50,000-Plus CUDA Cores Could Share One GPU on ThorstenMeyerAI.com

Before you orderOffer from Amazon

Get the latest gadgets delivered free with Prime

  • Fast, free delivery on millions of items
  • Prime Video, Amazon Music and more included
  • Member-only deals all year
Start your free Prime trial Free trial for eligible customers · Cancel anytime
As an affiliate, we earn on qualifying purchases.

TL;DR

Nvidia announced Rubin HPC at the SC26 supercomputing conference on Nov. 16, describing a GPU for scientific simulation and data processing with more than 50,000 CUDA cores. The company says it is expected in 2028, but has not disclosed detailed performance, memory, power, pricing or system information.

Nvidia announced Rubin HPC at the SC26 supercomputing conference in Chicago on Nov. 16, pitching the GPU at scientific simulation and data processing and claiming it will carry more than 50,000 CUDA cores on one chip. The company expects it to be available in 2028, but has yet to publish the specifications needed to judge its real-world performance or cost.

The announcement appeared in a Nvidia blog post by Dan Ernst, senior director of supercomputing products. Nvidia called Rubin HPC a new GPU in the Vera Rubin platform, designed to accelerate work in science, engineering and analytics. The company said the chip will combine its large core count with a sparse data acceleration capability and seventh-generation NVLink connections between GPUs.

Nvidia has not provided an exact core count. For comparison, Tom’s Hardware reports that the current Rubin GPU has 28,672 CUDA cores, while Nvidia’s specifications list 21,760 for the GeForce RTX 5090. Using those figures, a count above 50,000 would be at least about 1.7 times the current Rubin figure and 2.3 times the RTX 5090 figure. Those ratios are arithmetic comparisons, not Nvidia performance claims.

The announcement does not explain how the sparse-data feature works or disclose Rubin HPC’s physical layout. Tom’s Hardware suggested the core count might point to a design resembling a four-die Rubin Ultra configuration, but that is the outlet’s inference, not a confirmed description from Nvidia. The company also did not provide FP64 performance, memory capacity or bandwidth, power use, price, named systems or independent benchmark results.

At a glance
announcementWhen: Announced Nov. 16, 2026; expected in 20…
The developmentNvidia announced Rubin HPC, a GPU aimed at scientific and engineering workloads, and said it will have more than 50,000 CUDA cores on a single chip.

Nvidia Rubin HPC · SC26 · Nov 16, 2026

Where a science GPU sits in the AI research loop

An agent suggests ideas, a simulation tests them, the result guides the next step. Nvidia’s pitch: speed up the test.

The loop and one simulation (schematic)

Agentidea Test Refine many candidates next step sooner Inside one simulation: huge dataset GPU cores

Six steps, as Nvidia describes them

  1. Agent ideaAn AI agent proposes candidate materials, molecules or designsNvidia
  2. TestEach idea is checked in a simulation before lab work or a prototypeNvidia
  3. GatherThe code reaches into an immense dataset for a handful of scattered valuesNvidia
  4. ComputeIt calculates a result and repeats this on an enormous scaleNvidia
  5. Feed the coresRubin HPC: sparse data acceleration is meant to keep more than 50,000 CUDA cores supplied with dataNvidia’s claim, not yet measured
  6. RefineFaster results shorten the loop of hypothesis, simulation and refinementNvidia
Expected in 2028. Nvidia has not published FP64 figures, memory, power or a price for Rubin HPC, and there are no independent benchmarks yet.

Source: Nvidia Blog, “NVIDIA Announces Rubin HPC — A New GPU to Accelerate Agents for Science, Engineering and Analytics”, Nov 16, 2026. Schematic, not the real chip or software.

Graphic: Powered by Thorsten Meyer AI

More Cores for Simulation Work

Rubin HPC signals a specific focus within Nvidia’s data-center roadmap: workloads where performance can depend on moving scattered data as well as performing calculations. Nvidia says climate models, engineering simulations, genomic analysis and financial risk calculations can repeatedly retrieve small amounts of data from large datasets. In those tasks, it argues, processors can spend time waiting for data rather than computing.

The company’s pitch is that a balanced design will pair compute capacity with data movement, helping applications such as sparse matrix operations, stencil computations, sorting and reductions. If the design delivers on that aim, it could matter to research institutions and companies that use simulation to evaluate designs, materials or molecules. Faster simulation could shorten the time needed to test and refine ideas, though Nvidia has not yet supplied independent results showing how much faster Rubin HPC will be.

Nvidia also connects the product to AI agents that propose scientific or engineering ideas for testing. That is the company’s intended use case, not evidence that the GPU will improve discovery by a specified amount. For buyers, the practical question will be whether the full system’s performance and cost suit their workloads, not simply how many CUDA cores it contains.

Products Worth Considering

As an affiliate, we earn on qualifying purchases.

How Rubin HPC Fits Nvidia’s Roadmap

SC26 runs in Chicago from Nov. 15 to 20, and Nvidia’s announcement was one of its posts tied to the conference. The company places Rubin HPC within the Vera Rubin platform, alongside its broader GPU and interconnect technologies. Nvidia says the new chip will use seventh-generation NVLink; its Vera Rubin NVL72 product page describes the platform’s NVLink generation as sixth.

Core count alone does not establish how well a GPU handles scientific computing. Nvidia’s January technical blog, updated in August, lists native FP64 vector performance of 33 TFLOPS for Rubin, compared with 40 TFLOPS for Blackwell and 34 TFLOPS for Hopper. The same table lists FP64 matrix figures of 200 TFLOPS for Rubin and 150 TFLOPS for Blackwell, but labels those as peak performance using Tensor Core-based emulation algorithms. Nvidia says those results use multiple passes over lower-precision Tensor Cores; they are not the same measure as native FP64 vector performance.

That distinction is relevant because many scientific applications depend on double-precision calculations, and performance varies with the operations a program uses. Nvidia’s technical blog says dedicated FP64 vector performance remains important for applications not dominated by matrix kernels. The Rubin HPC announcement does not provide corresponding FP64 figures for the new GPU, so it cannot yet show how the chip’s core count relates to that workload.

““Designed to accelerate simulation and data processing for science, engineering and analytics workflows.””

— Nvidia, in its Rubin HPC announcement

Performance Details Still Missing

Rubin HPC remains an announcement rather than a fully specified product. Nvidia has not disclosed the exact CUDA core count, the GPU’s layout, how its sparse-data acceleration works, or its memory capacity and bandwidth. It also has not released power figures, pricing, system configurations or independent benchmarks.

Those omissions make it difficult to assess how the chip will perform across different simulation workloads or how it will compare with existing GPUs on a cost, power or application basis. The published comparisons in core counts do not establish a corresponding increase in useful compute performance. It is also not clear whether Rubin HPC’s FP64 capabilities differ from those described for the standard Rubin GPU.

Specifications Expected Before 2028

Nvidia says Rubin HPC is expected to be available in 2028. The company has not announced a more precise release date, prices, customer systems or a schedule for publishing additional specifications. Further platform announcements or technical material will be needed to establish how the GPU is configured and what performance it delivers.

Researchers and system buyers will be looking for application benchmarks, native FP64 results, memory details and power requirements as the launch approaches. Until Nvidia or independent testers publish those details, the core-count figure describes the scale of the announced chip, not its demonstrated performance.

Key Questions

What did Nvidia announce?

Nvidia announced Rubin HPC, a GPU in the Vera Rubin platform aimed at simulation and data processing for science, engineering and analytics. The company says it will have more than 50,000 CUDA cores on a single GPU.

When is Rubin HPC expected to be available?

Nvidia says Rubin HPC is expected in 2028. It has not provided a more exact availability date.

Does more than 50,000 CUDA cores mean it will be faster?

Not by itself. Core count is one specification, and Nvidia has not released benchmarks or detailed performance figures for Rubin HPC. Results will depend on the workload, data movement, memory and other design details.

What key specifications has Nvidia not released?

Nvidia has not announced an exact core count, FP64 figures, memory capacity or bandwidth, power use, pricing, named systems or independent test results. It also has not explained how its sparse data acceleration works.

Source: ThorstenMeyerAI.com

You May Also Like

Advanced Micro Devices Surges In Global Coverage

AMD experiences a surge in international media mentions, with 25 times more coverage than usual, signaling increased global interest in the company.

LG Faces Legal Action Over Alleged Smart TV Spying

Portugal’s Citizens Voice is asking courts to investigate alleged LG smart TV privacy breaches, amid wider questions about ACR opt-outs.

What Changed When Sophos Used OpenAI Daybreak For Threat Investigations?

OpenAI reports Sophos cut threat investigation time by 96% with Daybreak, but the available details do not explain how the figure was measured.

A More Personal Way To Hear The City On Foot

An IdeaNavigator AI concept proposes on-demand, route-aware walking narration, with a two-city test to assess retention and accuracy.