NVIDIA has announced a new chip called Rubin HPC. The company showed it at the SC26 conference on November 16, 2026. It has more than 50,000 CUDA cores on one chip and will be available in 2028.
The GPU is built to speed up simulations and data work in science, engineering and analytics. It is part of NVIDIA's Vera Rubin platform. It also uses sparse data acceleration and the seventh-generation NVLink, which helps GPUs share data faster.
Many scientific workloads process huge datasets while repeatedly accessing small, scattered pieces of information. This slows down GPUs, since data often cannot move as fast as the chip can compute.
NVIDIA says Rubin HPC addresses this bottleneck with a more balanced design. It improves how data moves, not just how fast the chip computes. That way, the CUDA cores stay busy and do not sit idle waiting for data.
The chip speeds up sparse matrix work, stencil computing, sorting and reductions. Scientists use these tasks in simulations, engineering tools and large data analysis.
Also Read: Trump's NVIDIA Stake Sees Sharp Rise Amid US Chip Policy Debate: Reports
NVIDIA wants agents to help in research. They can test hypotheses, run simulations, analyze the results and refine their next attempts.They could analyze large numbers of materials, molecules and designs.
In order to do this, NVIDIA will employ its CUDA-X libraries. The libraries have pre-built, optimized tools for simulations and analysis. The agents can use them to receive results more quickly.
One of such tools is called cuPhoton.xBeam. It allows analyzing the data obtained by the Event Horizon Telescope, which studies black holes. According to NVIDIA, a job that used to take a month now only takes eight minutes.
Another is cuSPARSE, which is open-source. The chipmaker says the system calculated cell shapes 900 times faster than CPUs in a test at Memorial Sloan Kettering Cancer Center. It also introduced Nemotron for Science, an open collection of AI models and datasets to create science agents.
Rubin HPC shows that NVIDIA is targeting more than AI systems that simply respond to questions. The company is developing chips and software that allow AI agents to run complex simulations and support scientific research. Should the technology work, researchers would be able to find their answers much quicker.