How Nvidia’s 2028 Rubin HPC Could Change Scientific GPU Workloads
AIThis post was created with the assistance of artificial intelligence (AI).

🔍 Read the full analysis: How Nvidia’s 2028 Rubin HPC Could Change Scientific GPU Workloads 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 SC26 on Nov. 16, describing it as a GPU for science, engineering and analytics workloads, with more than 50,000 CUDA cores and seventh-generation NVLink. The chip is expected in 2028, but Nvidia has not released detailed performance, memory, power or pricing figures, so its real-world effect on scientific computing is not yet known.

Nvidia announced Rubin HPC on Nov. 16 at the SC26 supercomputing conference in Chicago, pitching the planned 2028 GPU at scientific simulation and data processing rather than primarily AI training. Nvidia says the chip will have more than 50,000 CUDA cores and seventh-generation NVLink, but has not yet disclosed the performance and system details needed to judge how much faster research workloads will run.

The announcement appeared in a Nvidia blog post by Dan Ernst, the company’s senior director of supercomputing products. Nvidia describes Rubin HPC as a new GPU in its Vera Rubin platform, intended for science, engineering and analytics. It says the design pairs computing capacity with data movement, aiming to keep cores busy on workloads that repeatedly process scattered values across large datasets.

Nvidia identifies sparse matrix operations, stencil computations, sorting and reductions as target workloads. These patterns appear in areas such as climate modeling, engineering simulation, genomic analysis and financial risk calculations. The company also points to simulations used to test materials, molecules and designs proposed by AI agents, arguing that faster modeling could shorten the cycle of proposing and testing ideas.

The company says Rubin HPC will feature a sparse data acceleration capability and seventh-generation NVLink connections between GPUs. Nvidia has not explained the acceleration mechanism or provided an exact core count beyond “more than 50,000.” The announcement gives an expected availability year of 2028, but does not name a system or provide independent benchmark results.

At a glance
announcementWhen: Announced Nov. 16, 2026, at SC26; expec…
The developmentNvidia announced Rubin HPC, a planned 2028 GPU focused on scientific simulation and data-processing workloads.

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

Simulation Speed Shapes Research Cycles

If Rubin HPC improves performance on the data-heavy tasks Nvidia identifies, research teams could complete more simulation runs within a fixed computing budget or reach results sooner. That could matter in fields where experiments and prototypes are expensive, and where simulations help researchers decide which ideas are worth testing in the physical world. Those are potential consequences, not demonstrated results: Nvidia has not published workload benchmarks for the chip.

The announcement also signals a distinction within GPU computing. Many recent GPU launches have emphasized training and operating AI models, while Nvidia is presenting Rubin HPC around scientific and engineering applications that may be limited by data access as well as arithmetic throughput. Whether a specialized design offers a meaningful advantage will depend on the code, memory system, precision support and software available when systems ship.

For institutions planning supercomputer purchases, the missing specifications matter as much as the core count. A large number of CUDA cores alone does not establish performance on double-precision scientific codes, energy use, total system cost or the number of GPUs required. Procurement decisions will need comparable measurements on relevant applications.

Products Worth Considering

Logitech MX Mechanical Wireless Illuminated Keyboard, Tactile Quiet, Graphite

Logitech MX Mechanical Wireless Illuminated Keyboard, Tactile Quiet, Graphite

Apple Watch Series 8 GPS 41mm Smartwatch with Midnight Aluminum Case and Midnight Sport Band (S/M)

Apple Watch Series 8 GPS 41mm Smartwatch with Midnight Aluminum Case and Midnight Sport Band (S/M)

As an affiliate, we earn on qualifying purchases.

Core Counts Leave Key Questions

Nvidia’s headline figure is large relative to other reported counts, but comparisons should be treated carefully. Tom’s Hardware reports that the current Rubin GPU has 28,672 CUDA cores; Nvidia’s own platform material confirms 224 streaming multiprocessors. Nvidia lists 21,760 CUDA cores for the GeForce RTX 5090. Comparing those figures with Nvidia’s “more than 50,000” yields a lower-bound arithmetic comparison of about 1.7 times the reported Rubin count and 2.3 times the RTX 5090 count. These are core-count ratios, not performance results, and Nvidia did not make those comparisons in its announcement.

Physical design details are also absent. Tom’s Hardware suggested the number might point to a configuration resembling a multi-die Rubin Ultra GPU, but that is an outlet’s inference, not a confirmed description from Nvidia. The company has not disclosed Rubin HPC’s die layout, memory capacity or bandwidth, power requirements, price, or its native FP64 performance, a key measure for many traditional scientific applications.

That last gap is relevant because Nvidia’s Vera Rubin platform technical material reports 33 TFLOPS of native FP64 vector performance for Rubin, compared with 34 TFLOPS for Hopper and 40 TFLOPS for Blackwell. It separately lists higher FP64 matrix figures for newer chips using Tensor Core-based emulation. Those platform figures do not establish Rubin HPC’s specifications; they show why the new chip’s precision and workload-specific results will be needed before its scientific value can be assessed.

““A record number of CUDA cores — more than 50,000 on a single GPU.””

— Nvidia, in its Nov. 16, 2026, blog announcement

Performance Details Still Missing

Nvidia has not released FP64 figures, memory capacity or bandwidth, power consumption, pricing, named systems, or independent benchmarks for Rubin HPC. It has also not specified the exact CUDA core count or explained how the sparse data acceleration capability works. Without those details, it is not possible to determine whether the chip will outperform existing GPUs on particular scientific applications, or what performance would cost in power and money.

The 2028 availability date is a company expectation, not a delivery guarantee. It remains unclear which supercomputer vendors or research institutions will use Rubin HPC, how its software support will compare with existing platforms, and whether its benefits will apply broadly or mainly to selected data-intensive workloads.

Benchmarks Will Test Nvidia’s Pitch

The next useful evidence will be fuller technical specifications, named systems and tests on scientific applications. Researchers and computing-center operators will need measurements that identify the workload, precision, system configuration and energy use, rather than relying on core counts alone. Nvidia has not announced a date for releasing those details.

Rubin HPC is currently expected in 2028. Until then, the announcement establishes Nvidia’s target workloads and some architectural claims, but not the chip’s measured effect on scientific computing. Comparisons with current hardware will become possible when production systems and independently checkable results are available.

Key Questions

What is Nvidia Rubin HPC?

Rubin HPC is a GPU Nvidia announced for its Vera Rubin platform, aimed at simulation and data processing in science, engineering and analytics.

When is Rubin HPC expected to be available?

Nvidia says Rubin HPC is expected in 2028. The company has not provided a specific release date or named systems.

How many CUDA cores will it have?

Nvidia says the GPU will have more than 50,000 CUDA cores, but has not disclosed the exact count. That specification does not by itself show how fast the chip will perform on scientific workloads.

Has Nvidia shown that Rubin HPC will speed up simulations?

Not yet. Nvidia has described the workloads it is targeting, but has not published Rubin HPC benchmarks or independent performance results.

What important specifications have not been disclosed?

Nvidia has not provided Rubin HPC figures for FP64 performance, memory, power or price, and has not explained the sparse-data acceleration feature in detail.

Source: ThorstenMeyerAI.com

You May Also Like

Nvidia’s First GPU: GeForce 256

Nvidia released the GeForce 256 on Oct. 11, 1999, and marketed it as its first GPU, helping establish a lasting graphics-card category.

Nvidia’s CEO Search: What We Know So Far

Nvidia CFO Colette Kress is interim CEO as the board starts a permanent search. A reported approach and public suggestions remain unconfirmed.

Could Grok Voice 3.0 Change How You Interact With AI?

A headline identifies xAI’s Grok Voice 3.0 announcement, but available material does not confirm its features, release date or availability.

GPT-6.1 Terra May Help Teams Cut Costs On Routine AI Tasks

OpenAI launched GPT-6.1 Terra at lower token prices. Independent tests show its low and medium settings offer lower task costs, with trade-offs in capability.