Nvidia unveils Rubin HPC, a science-focused GPU due in 2028
Nvidia says the chip will have more than 50,000 CUDA cores but has not given performance, memory or FP64 figures. ServeTheHome says FP64 is the number to watch.
Key takeaways
- Nvidia announced Rubin HPC at SC26, a GPU for simulations and data work in science, engineering and analytics.
- Nvidia says it will have more than 50,000 CUDA cores, sparse data acceleration and seventh-generation NVLink.
- The chip is expected to be available in 2028.
- Nvidia has not disclosed performance, memory or FP64 figures, according to ServeTheHome and Tom's Hardware.
- ServeTheHome calls it Nvidia's first significant HPC-class part since Hopper; AMD and Intel are also pitching HPC GPUs.
Nvidia announced Rubin HPC at the SC26 supercomputing conference on Monday, Nov. 16, describing it as a new GPU in its Vera Rubin platform meant to speed up simulations and data work in science, engineering and analytics. Nvidia expects the chip to ship in 2028.
According to Nvidia, Rubin HPC will have more than 50,000 CUDA cores on a single GPU, which it calls a record, along with a "sparse data acceleration" feature meant to keep those cores supplied with data. GPUs will be linked using seventh-generation NVLink, Nvidia says.
"For agents tackling science and engineering, faster simulation means faster discovery," said Ian Buck, Nvidia's vice president of hyperscale and high-performance computing. Nvidia pitches the chip at scientists and engineers who use AI agents to explore candidate materials, molecules and designs, and who then rely on simulations to test them.
What is known about the chip?
So far, not much. ServeTheHome calls the announcement in effect a teaser, saying Nvidia shared no expected performance figures or memory details. Tom's Hardware says Nvidia did not disclose HBM capacity or bandwidth, or whether the chip shifts resources toward native FP64, the double-precision maths that many classic scientific workloads depend on.
Tom's Hardware notes that today's Rubin GPU has 28,672 CUDA cores, so reaching more than 50,000 would likely require more dies in one package, possibly in a layout like the four-die Rubin Ultra; Nvidia did not reveal the chip's physical design. ServeTheHome estimates the new count is around twice that of the current Rubin chip and says the specifications look closer to Rubin Ultra, the first chip set to support NVLink 7. Both outlets describe these points as their own inferences.
Nvidia says the design scales data movement and compute together rather than raising compute alone, aimed at workloads such as climate modelling, engineering simulation, genome analysis and financial risk. It also says its CUDA-X software libraries can be used with Rubin HPC.
Why does it matter?
ServeTheHome says Rubin HPC would be Nvidia's first significant chip aimed at high-performance computing since the Hopper generation nearly four years ago, after its Blackwell parts focused heavily on AI and low-precision maths. CUDA cores handle general vector maths, which ServeTheHome says matters for HPC because few such workloads can make heavy use of the tensor cores built for AI.
The chip will face competition. ServeTheHome says AMD plans to launch its HPC-focused Instinct MI430X in 2027 and Intel is preparing a GPU line of its own, while Tom's Hardware reports that AMD and Intel are promoting native FP64 performance in their GPUs at this week's conference. Nvidia says the chip will support a wide range of precisions, from NVFP4 up to FP64, ServeTheHome reports, but it has not given FP64 performance figures. ServeTheHome says that figure will be the key one to watch, noting that Nvidia's FP64 performance has stayed a little under 40 TFLOPS since Hopper.
Questions readers are asking
What is Rubin HPC?
A GPU announced at SC26 as part of Nvidia's Vera Rubin platform, built to speed up simulations and data work in science, engineering and analytics.
When will it be available?
Nvidia expects the chip to ship in 2028.
How fast is it?
Nvidia has not shared performance figures. It says the chip will have more than 50,000 CUDA cores, but it has not disclosed memory details or FP64 performance figures.
Sources
- #nvidia
- #rubin
- #hpc
- #supercomputing
- #gpu
Get the day's top AI & tech stories on WhatsApp: Follow the NextFish channel
Get stories like this every morning.
Related
Chips & Infra2 min
China's LineShine stays No. 1 on TOP500 as new entries hit a record low
The November list keeps the same top 10 as June and adds just 27 new systems, the fewest ever. Tom's Hardware ties the drop to new hardware going to AI data centers.
Chips & Infra2 min
US issues rule to stop remote access workarounds to AI chip export limits
The Commerce Department's BIS says the rule targets data centers in third countries. Reports say it adds license rules for five countries, including Singapore and Malaysia.
Chips & Infra2 min
Nvidia board is not considering Huang's return as CEO, CNBC reports
A person familiar with the matter says the board is concentrating on its inquiry into Huang. Tech figures and a petition have called for his return.
Chips & Infra3 min
MLPerf Training v6.1 adds post-training test as Rubin, TPUv7 debut
MLCommons says 24 organizations submitted 250 results. Nvidia claims the fastest times, Lambda reports its own gains, and the first Rubin and Google TPUv7 entries appear.