🔍 Read the full analysis: Nvidia’s Rubin HPC Puts More Than 50,000 CUDA Cores On One GPU For Science. Due In 2028 on ThorstenMeyerAI.com
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
TL;DR
Nvidia announced Rubin HPC at SC26 on Nov. 16, describing a GPU for scientific simulation, data processing and engineering workloads with more than 50,000 CUDA cores. The chip is expected in 2028, but Nvidia has not disclosed performance benchmarks, memory specifications, power use or price.
Nvidia announced Rubin HPC, a GPU designed for scientific simulation and data processing, at the SC26 supercomputing conference in Chicago on Nov. 16. The company says the chip will have more than 50,000 CUDA cores and is expected to be available in 2028, but has not yet published benchmarks or several specifications needed to judge its performance and cost.
The announcement appeared in a Nvidia blog post by Dan Ernst, senior director of supercomputing products. Nvidia describes Rubin HPC as a new GPU in its Vera Rubin platform, aimed at science, engineering and analytics rather than being presented primarily as a general-purpose AI training chip. The company calls its core count a record, but did not provide an exact number.
Nvidia says the design combines a sparse data acceleration capability with a “balanced design” intended to scale data movement alongside computing capacity. It also specifies seventh-generation NVLink for connections between GPUs. The company has not explained how the sparse-data feature works or provided enough implementation detail to assess those claims independently.
The stated use cases include climate modeling, engineering simulation, genomic analysis and financial risk. Nvidia’s description focuses on applications that repeatedly retrieve scattered data, perform calculations and repeat the process at large scale. The company argues that reducing time spent waiting for data could make these workloads run more efficiently. That is Nvidia’s explanation of the design goal, not a result established by independent testing.
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)
Six steps, as Nvidia describes them
- Agent ideaAn AI agent proposes candidate materials, molecules or designsNvidia
- TestEach idea is checked in a simulation before lab work or a prototypeNvidia
- GatherThe code reaches into an immense dataset for a handful of scattered valuesNvidia
- ComputeIt calculates a result and repeats this on an enormous scaleNvidia
- 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
- RefineFaster results shorten the loop of hypothesis, simulation and refinementNvidia
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
A GPU Aimed at Data-Heavy Science
For researchers and organizations running large simulations, the announcement points to a distinct design priority: improving how quickly a GPU can move and process data in workloads that may not behave like dense AI calculations. Nvidia says Rubin HPC is intended to keep computing cores supplied with data for tasks such as sparse matrix operations, stencil computations, sorting and reductions. If the design delivers measurable gains on those workloads, it could affect the time and computing resources needed to test scientific or engineering hypotheses.
Nvidia also links the GPU to workflows in which AI agents propose a material, molecule or design, followed by simulation and refinement. That framing makes simulation speed relevant to AI-assisted research, but the company has not released evidence quantifying how much faster such a workflow would become. The practical value will depend on application performance, system availability, energy requirements and price—not the CUDA-core count alone.
The announcement may matter to supercomputing planners because it signals that Nvidia intends to address workloads with different bottlenecks from AI model training. Yet buyers cannot compare projected performance or operating costs from the information released so far. The central question is whether the specialized data-handling features translate into better results on real scientific software.
Products Worth Considering

Apple Watch Series 12 GPS 46mm Smartwatch

Anker SOLIX C1000 Gen 2 Portable Power Station with 200W Solar Panel

Sony WH-1000XM5 Wireless Noise-Canceling Over-Ear Headphones, Black, International Version
As an affiliate, we earn on qualifying purchases.
How Rubin HPC Fits Nvidia’s Lineup
Nvidia presented Rubin HPC during SC26, held in Chicago from Nov. 15 to 20. The announcement is part of the Vera Rubin platform, but the available information does not identify a specific server or supercomputer that will use the new GPU. Nvidia says the product is expected in 2028, leaving a substantial period before customers can evaluate shipped systems.
The source material compares the headline core count with two existing Nvidia products. Tom’s Hardware reports that the current Rubin GPU has 28,672 CUDA cores; Nvidia’s technical blog confirms that GPU’s 224 streaming multiprocessors. Nvidia lists 21,760 CUDA cores for the GeForce RTX 5090. Using the lower bound in Nvidia’s “more than 50,000” claim, the source calculates that Rubin HPC would have at least 1.7 times the current Rubin count and at least 2.3 times the RTX 5090 count. These are arithmetic comparisons of core counts, not measures of performance.
Nvidia’s earlier Vera Rubin platform material also shows why raw core totals do not settle questions about scientific computing. Its figures list native FP64 vector performance at 33 TFLOPS for Rubin, compared with 34 TFLOPS for Hopper and 40 TFLOPS for Blackwell. Higher FP64 matrix figures for newer chips are described as peak performance using Tensor Core-based emulation algorithms. Those platform figures do not establish Rubin HPC’s own FP64 performance, which Nvidia has not supplied.
“Designed to accelerate simulation and data processing for science, engineering and analytics workflows.”
— Nvidia, describing Rubin HPC
Performance Details Still Missing
Nvidia has not released an exact CUDA-core count, FP64 performance figures, memory capacity or bandwidth for Rubin HPC. It also has not announced power consumption, price, named systems or independent benchmarks. Without those details, researchers and procurement teams cannot determine how the GPU will compare on particular scientific applications or estimate the cost of operating it.
The chip’s physical layout is also undisclosed. Tom’s Hardware suggested that the core count might indicate a configuration similar to a four-die Rubin Ultra GPU, but that is the outlet’s inference, not a confirmed description from Nvidia. Nvidia has not explained the sparse-data acceleration capability beyond saying it is intended to keep cores supplied with data.
The “record” wording is Nvidia’s characterization, and the comparisons available in the source material concern core counts only. More cores do not by themselves establish faster application performance. Whether Rubin HPC improves native double-precision calculations or data movement for specific codes remains unknown.
Benchmarks Before the 2028 Launch
Nvidia says Rubin HPC is expected to be available in 2028. The next useful disclosures would include detailed specifications, system configurations, power and pricing information, and performance results for representative scientific workloads. Those details would allow researchers to compare the GPU with existing accelerators using more than core counts.
Independent testing will be needed to check Nvidia’s claims across applications such as sparse matrix calculations and simulation codes. Until those results and deployment details emerge, the announcement establishes Nvidia’s intended direction for the product, but not its real-world performance or likely availability in particular supercomputers.
Key Questions
What did Nvidia announce?
Nvidia announced Rubin HPC, a GPU in its Vera Rubin platform aimed at scientific simulation, data processing, engineering and analytics workloads.
How many CUDA cores will Rubin HPC have?
Nvidia says it will have more than 50,000 CUDA cores. The company has not provided an exact count.
When is Rubin HPC expected to be available?
Nvidia says the GPU is expected in 2028. It has not named specific systems or customers in the supplied announcement.
Does the core count show how fast the GPU will be?
No. Core count alone does not establish performance for scientific applications. Nvidia has not released Rubin HPC benchmarks, FP64 figures, memory specifications or power data.
What key information has Nvidia not disclosed?
The company has not provided an exact core count, memory capacity or bandwidth, power use, price, detailed chip layout, named systems or independent benchmark results.
Source: ThorstenMeyerAI.com
