NVIDIA Unveils Next-Generation Vera Rubin Superchip Featuring Two Powerful GPUs Set for Production in 2024

NVIDIA has unveiled its next-generation Vera Rubin Superchip at the GTC event in Washington, marking a significant leap towards the future of artificial intelligence deployment.

NVIDIA Prepares for Mass Production of Vera Rubin Superchip with First Rubin GPUs in Labs

During the GTC event in October 2025, CEO Jensen Huang introduced the much-anticipated Vera Rubin Superchip. This was the inaugural public display of the motherboard, referred to by NVIDIA as a Superchip, which integrates the advanced Vera CPU alongside a pair of high-powered Rubin GPUs. Additionally, the motherboard features ample LPDDR system memory, which pairs effectively with the HBM4 memory utilized in the Rubin GPUs.

Vera Rubin Superchip presented on stage, showcasing specifications including 100 PF AI and 2 TB Rapid Memory.

NVIDIA Vera Rubin NVL144 System Set for Launch in 2H 2026

The NVIDIA Vera Rubin NVL144 platform will feature two innovative chips. The Rubin GPU will incorporate two reticle-sized chips, delivering up to 50 PFLOPs of FP4 performance and boasting a staggering 288 GB of cutting-edge HBM4 memory. Alongside this, the platform will include an 88-core Vera CPU designed on a custom Arm architecture, providing 176 threads and 1.8 TB/s NVLINK-C2C interconnect bandwidth.

The performance metrics of the NVIDIA Vera Rubin NVL144 platform are commendable, featuring an impressive 3.6 Exaflops of FP4 inference and 1.2 Exaflops of FP8 training. This reflects a significant 3.3 times performance enhancement compared to the GB300 NVL72 model. Additionally, it offers 13 TB/s of HBM4 memory and 75 TB of fast memory, marking a 60% improvement over its predecessor, and double the capabilities of NVLINK and CX9, with ratings of 260 TB/s and 28.8 TB/s respectively.

Upcoming NVIDIA Rubin Ultra NVL576 System Expected in 2H 2027

Following the NVL144, NVIDIA is set to launch the Rubin Ultra platform in the latter half of 2027. This system will expand the NVL architecture from 144 to 576. While maintaining the CPU’s architecture, the Rubin Ultra GPU will feature four reticle-sized chips that promise up to 100 PFLOPS of FP4 performance and a substantial HBM4e memory capacity of 1 TB dispersed across 16 HBM sites.

When it comes to performance, the NVIDIA Rubin Ultra NVL576 is poised to deliver 15 Exaflops of FP4 inference and 5 Exaflops of FP8 training, achieving a remarkable 14-fold increase over the GB300 NVL72. Additionally, it will feature an astounding 4.6 PB/s of HBM4 memory and 365 TB of fast memory, representing a significant 8x improvement over the GB300. Moreover, the NVLINK and CX9 specifications will see a 12-fold and 8-fold increase in capabilities, with throughput ratings reaching 1.5 PB/s and 115.2 TB/s respectively.

NVIDIA's roadmap presentation on stage illustrating product iterations including the V100 HBM2 and BlueField-3.

NVIDIA’s Comprehensive AI GPU Roadmap

GPU Codename	Feynman	Rubin (Ultra)	Rubin	Blackwell (Ultra)	Blackwell	Hopper	Ampere	Time	Pascal
GPU Family	GF200?	GR300?	GR200?	GB300	GB200/GB100	GH200/GH100	GA100	GV100	GP100
GPU WeU	F200?	R300?	R200?	B300	B100/B200	H100/H200	A100	V100	P100
Memory	HBM4e/HBM5?	HBM4	HBM4	HBM3e	HBM3e	HBM2e/HBM3/HBM3e	HBM2e	HBM2	HBM2
Launch	2028	2027	2026	2025	2024	2022-2024	2020-2022	2018	2016