NVIDIA Vera Rubin NVL72
Rack-scale system pairing 72 Rubin GPUs with 36 Vera CPUs (3,168 Olympus Arm cores total, 54 TB LPDDR5X CPU memory). Headline NVFP4 inference 3,600 PFLOPS dense; NVFP4 training 2,520 PFLOPS dense. Tensor-core-emulated FP32 SGEMM reaches 28.8 PFLOPS, FP64 DGEMM 14.4 PFLOPS. Aggregate memory shown is HBM4 only (CPU LPDDR5X is excluded from GPU memory aggregate). Total chips per rack: 1,296. Specifications preliminary per NVIDIA GTC 2025 footnote.
System Details
GPU Configuration
System Specifications
Precision Performance Breakdown
| Precision | System Performance | Per GPU | Efficiency |
|---|---|---|---|
| FP64 | 2.400 PFLOPS | 33.3 TFLOPS | — |
| FP32 | 9.360 PFLOPS | 130.0 TFLOPS | — |
| TF32 | 144.000 PFLOPS | 2000.0 TFLOPS | — |
| FP16 | 288.000 PFLOPS | 4000.0 TFLOPS | — |
| BF16 | 288.000 PFLOPS | 4000.0 TFLOPS | — |
| FP8 | 1260.000 PFLOPS | 17500.0 TFLOPS | — |
| INT8 | 18.000 PFLOPS | 250.0 TFLOPS | — |
| NVFP4 | 3600.000 PFLOPS | 50000.0 TFLOPS | — |
| FP6 | 1260.000 PFLOPS | 17500.0 TFLOPS | — |
| FP32_TC | 28.800 PFLOPS | 400.0 TFLOPS | — |
| FP64_TC | 14.400 PFLOPS | 200.0 TFLOPS | — |
Powered by NVIDIA Rubin
This system utilizes 72 × NVIDIA Rubin SXM GPUs, each delivering exceptional performance for AI and HPC workloads.
Per GPU TDP
—
Per GPU Memory
288 GB
Process Node
—
Architecture
Rubin
Documentation & Resources
Official Datasheet
NVIDIA Vera Rubin NVL72 technical specifications
workstation-datasheet-dgx-spark-gtc25-spring-nvidia-us-3716899-web.pdf
NVIDIA • 2025-10-16
NVIDIA DGX Documentation
System guides, deployment resources
Typical Use Cases
The NVIDIA Vera Rubin NVL72 with 72× Rubin GPUs is designed for enterprise-scale workloads requiring 2.4 PFLOPS of compute power with NVLink 6 / NVSwitch interconnect for efficient multi-GPU communication.