Next-generation rack-scale platform for agentic AI and reasoning, unifying 72 Rubin GPUs and 36 Vera CPUs in a single NVLink domain.
NVIDIA Vera Rubin NVL72 is a rack-scale AI supercomputer that joins 72 Rubin GPUs and 36 Vera CPUs, together with ConnectX-9 SuperNICs and BlueField-4 DPUs, in a single sixth-generation NVLink domain. It is designed for AI factories that train and serve trillion-parameter, mixture-of-experts and long-context reasoning models.
The platform is built on the third-generation NVIDIA MGX NVL72 rack with cable-free modular trays and is supported by more than 80 ecosystem partners, so the same architecture is available as NVIDIA DGX and from server makers such as Supermicro and ASUS.
Rubin GPUs, Vera CPUs, NVLink switches, ConnectX-9 SuperNICs and BlueField-4 DPUs are engineered to operate as one rack-scale computer.
NVIDIA quotes up to one-tenth the cost per million tokens of GB200 NVL72 for deep-reasoning, agentic inference.
Up to 10x more tokens per megawatt than the previous generation within the same power envelope.
Trains mixture-of-experts models with one-quarter of the GPUs the prior generation needed.
1,400 TB/s of GPU memory bandwidth per rack, plus up to 54 TB of LPDDR5X attached to the Vera CPUs.
Third-generation MGX NVL72 rack with cable-free trays for faster assembly and serviceability.
| GPUs | 72× NVIDIA Rubin |
|---|---|
| CPUs | 36× NVIDIA Vera |
| Total NVIDIA + HBM4 chips | 1,296 |
| NVFP4 inference | 3,600 PFLOPS |
|---|---|
| NVFP4 training | 2,520 PFLOPS |
| FP8/FP6 training | 1,260 PFLOPS |
| FP16/BF16 | 288 PFLOPS |
| TF32 | 144 PFLOPS |
| FP32 | 9,360 TFLOPS |
| FP64 | 2,400 TFLOPS |
| GPU memory | 20.7 TB HBM4 |
|---|---|
| GPU memory bandwidth | 1,400 TB/s |
| CPU memory | Up to 54 TB LPDDR5X |
| NVLink bandwidth | 216 TB/s |
| NVLink-C2C bandwidth | 65 TB/s |
| Networking bandwidth | 32.4 TB/s (bi-directional) |
| Rack | Third-generation NVIDIA MGX NVL72, liquid-cooled |
|---|---|
| Inlet temperature | 45 °C |
| Networking | ConnectX-9 SuperNICs, BlueField-4 DPUs |
| Scale-out fabric | Quantum-X800 InfiniBand or Spectrum-X Ethernet |
Turnkey, ready-to-deploy enterprise AI infrastructure built on the Vera Rubin platform and managed with NVIDIA Mission Control.
Blackwell Ultra rack-scale system built for AI reasoning, training and data processing at scale.
Liquid-cooled 72-GPU NVLink domain that acts as a single massive GPU for trillion-parameter LLM inference and training.
Tell us your workload and timeline for the NVIDIA Vera Rubin NVL72. We reply within one business day with a configuration, pricing and lead time.