🔍 Read the full analysis: Nvidia Rubin HPC: A Science GPU With More Than 50,000 CUDA Cores on ThorstenMeyerAI.com
Get business pricing on monitors, keyboards and dev gear
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Nvidia announced Rubin HPC at SC26, describing it as a GPU for science, engineering and analytics workloads, with more than 50,000 CUDA cores and seventh-generation NVLink. The company expects it to be available in 2028, but has not disclosed its exact core count, memory, power, price or performance benchmarks.
Nvidia announced Rubin HPC at the SC26 supercomputing conference in Chicago on November 16, describing the GPU as designed for science, engineering and analytics workloads. The company says one GPU will have more than 50,000 CUDA cores and expects the product to be available in 2028, but has not released several specifications needed to judge its performance or cost.
The announcement appeared in an Nvidia blog post by Dan Ernst, senior director of supercomputing products. Nvidia says Rubin HPC is part of its Vera Rubin platform and is intended to accelerate simulation and data processing. The company called the core count a record, but did not provide an exact number.
Nvidia also cited a sparse data acceleration capability intended to keep the GPU’s cores supplied with data, plus seventh-generation NVLink for connections between GPUs. The company did not explain the sparse-data feature’s implementation or provide measured results showing how it affects specific applications.
For comparison, Tom’s Hardware reports that Nvidia’s current Rubin GPU has 28,672 CUDA cores, while Nvidia lists 21,760 for the GeForce RTX 5090. Based on those figures, the announced floor of more than 50,000 would be at least 1.7 times the Rubin count and 2.3 times the RTX 5090 count. Those ratios compare core counts only; they do not establish a corresponding increase in performance.
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
Why Simulation Workloads Matter
Rubin HPC signals Nvidia’s effort to address computing tasks in which processors can spend substantial time waiting for data rather than performing calculations. Nvidia describes workloads that repeatedly retrieve scattered values from large datasets, calculate results and repeat the process. It names sparse matrix operations, stencil computations, sorting and reductions as patterns the design aims to accelerate.
That focus matters to researchers and engineering teams using supercomputers for tasks such as climate modeling, genomic analysis and financial risk calculations. Faster simulations could help shorten the cycle of testing and revising a scientific hypothesis or design. Nvidia also connects this work to AI agents that propose materials, molecules or designs, arguing that their output must be tested in simulation. That is the company’s rationale, not an independently demonstrated result for Rubin HPC.
The core count alone cannot show whether the GPU will improve those workloads. Performance will depend on details including memory capacity and bandwidth, data movement, precision support, software and system configuration. Those details are especially relevant for scientific applications that rely on double-precision (FP64) calculations.
Popular right now

TCL 55-Inch R7D Series QLED 4K UHD HDR Roku Smart TV (55R7D, 2026)

Apple AirPods Pro 3 Wireless Earbuds with Active Noise Cancellation

Shark AI Ultra Robot Vacuum with Matrix Clean and Self-Empty Base
As an affiliate, we earn on qualifying purchases.
Nvidia’s Scientific Computing Pitch
The Rubin HPC announcement is distinct from Nvidia’s broader Vera Rubin platform pitch: this GPU is described specifically around simulation and data processing, rather than being presented only as a general AI accelerator. Nvidia says the design aims to scale data movement and compute together, reflecting the challenge of keeping processing units busy on data-intensive tasks.
Nvidia’s earlier technical material on the Vera Rubin platform offers a reminder that specifications for other Rubin products should not be assumed to apply to Rubin HPC. In a table published in January and updated in August 2026, Nvidia listed native FP64 vector performance of 33 TFLOPS for Rubin, compared with 34 TFLOPS for Hopper and 40 TFLOPS for Blackwell. The same table listed higher FP64 matrix figures for newer products, while identifying Blackwell and Rubin figures as peak performance using Tensor Core-based emulation algorithms. Those figures describe the platform information Nvidia published, not Rubin HPC’s yet-undisclosed performance.
Nvidia’s announcement was a blog post published during SC26, held in Chicago from November 15 to 20. The available source material reports no separate Nvidia newsroom press release for the chip. Comparisons with existing products are useful for scale, but they do not confirm Rubin HPC’s physical layout or its behavior in real-world scientific applications.
“A record number of CUDA cores — more than 50,000 on a single GPU.”
— Nvidia, in its announcement
Specs Still Missing for Buyers
Nvidia has not released an exact CUDA core count, FP64 performance figures, memory capacity or bandwidth for Rubin HPC. It has also not disclosed power consumption, pricing, named systems or independent benchmarks. Without those details, customers cannot compare expected performance, energy use or purchase cost against existing accelerators.
The company has not explained how the sparse data acceleration capability works or quantified its effect on the workloads it names. The announcement also does not confirm the GPU’s physical layout. Tom’s Hardware suggested the core count might indicate a configuration similar to the four-die Rubin Ultra GPU, but that is an outlet’s inference, not a Nvidia specification. It remains unclear whether Rubin HPC will offer particular advantages for codes that depend on native FP64 math rather than matrix operations.
Details Expected Before 2028
Nvidia says Rubin HPC is expected to be available in 2028. The announcement does not give a more specific release date or identify the systems that will use it. Before researchers and computing centers can evaluate the chip for procurement, they will need fuller specifications, including precision performance, memory, power and system-level details.
Further Nvidia disclosures and independent testing will show whether the GPU’s data-movement features translate into gains for real scientific workloads. Until then, the confirmed information is limited to the product’s stated purpose, its reported core-count floor, seventh-generation NVLink and expected 2028 availability.
Key Questions
What is Nvidia Rubin HPC?
Rubin HPC is a GPU Nvidia announced for simulation and data processing in science, engineering and analytics. The company says it is part of the Vera Rubin platform.
How many CUDA cores will Rubin HPC have?
Nvidia says the GPU will have more than 50,000 CUDA cores, but has not published an exact count. Core totals alone do not establish application performance.
When is Rubin HPC expected to be available?
Nvidia says Rubin HPC is expected in 2028. It has not announced a more precise date or named systems that will use the GPU.
What key specifications has Nvidia not disclosed?
The announcement does not give Rubin HPC’s FP64 performance, memory capacity or bandwidth, power consumption, price or independent benchmark results. Nvidia also has not explained how its sparse data acceleration capability works.
Does 50,000-plus cores mean it will be faster than current GPUs?
Not by itself. The announced core count is a specification, not a benchmark. Actual results will depend on memory, data movement, precision support, software and the workload being run.
Source: ThorstenMeyerAI.com
