ITEN

Home › Articles › Supercomputing 2026: Nvidia unveils Rubin HPC with more than 50,000 CUDA cores; Intel responds with first tests of the D270 GPU

Supercomputing 2026: Nvidia unveils Rubin HPC with more than 50,000 CUDA cores; Intel responds with first tests of the D270 GPU

At the SC26 conference in Chicago Nvidia announced Rubin HPC, a GPU for simulation and scientific computing due in 2028 with a record number of CUDA cores. Intel showed the first benchmarks, its own and on pre-production silicon, of its 350 W D270 GPU.

Statcade graphic: Supercomputing 2026: Nvidia unveils Rubin HPC with more than 50,000 CUDA cores; Intel responds with first tests of the D270 GPU
Image: Statcade graphic based on data from Nvidia and Intel

The Supercomputing 2026 conference (SC26), running in Chicago from 15 to 20 November, brought the first announcements from the big chipmakers on Monday the 16th.

Nvidia presented Rubin HPC, a new GPU in the Vera Rubin platform designed to accelerate simulation and data processing in science, engineering and analytics. According to the company it is the GPU with the most CUDA cores ever made, “more than 50,000” on a single chip, alongside a sparse data acceleration feature (datasets with many zeros) and seventh-generation NVLink interconnect. Delivery is planned for 2028. For comparison, SiliconANGLE notes that the current flagship Rubin GPU has more than 28,000 CUDA cores; Tom's Hardware suggests that such a number implies a four-die configuration like Rubin Ultra, but Nvidia has not confirmed it. At the same event Nvidia also presented the DGX Spark HPC Developer Kit, a desktop cluster of four DGX Spark systems connected in a ring, and new CUDA-X libraries for research. The announcements come at a turbulent time for the company, as Tom's Hardware notes.

Intel, for its part, published the first head-to-head comparisons for the D270 GPU, the project known until October as Crescent Island: an air-cooled 350 W PCIe card with Xe3P architecture and LPDDR5X memory instead of HBM (160 GB on Intel's card, up to 480 GB on partner cards). Against the 600 W RTX PRO 6000 Blackwell Server Edition, Intel claims up to 5.9 times the throughput in double-precision (FP64) financial calculations and a smaller advantage in inference, up to 1.12 times the tokens generated on Qwen3-30B-A3B in FP8. These are, however, internal results on pre-production silicon: Hardware Busters points out that the financial tests list eight Nvidia cards against a single D270 without clarifying whether the data are normalised, that Nvidia's inference figures come from an AWS virtual machine and that the RTX PRO 6000 uses the same GB202 chip as the GeForce RTX 5090, with FP64 capability reduced by product design. There is no price or availability date for the D270 yet; the first customer sample shipments started in October.

Neither chip is meant for gaming PCs, but both show where the two brands are concentrating their design and manufacturing resources.

Sources: Nvidia, official blog on SC26 (16 November 2026); SiliconANGLE (16 November 2026); Tom's Hardware (16 November 2026); Hardware Busters (16 November 2026). Text by Statcade.

Support Statcade

If Statcade's data is useful to you, you can buy us a coffee on Ko-fi.

Buy us a coffee on Ko-fi (opens in a new tab)

The week’s numbers

One email a week, with video game numbers. No third-party promotions.

Sign-up is for people aged at least 14. We send you a confirmation email: you join the list only after you click the link.