GPUs & AI Accelerators

NVIDIA GB200 NVL72

Name: NVIDIA GB200 NVL72
Brand: NVIDIA
SKU: GB200 NVL72

GB200 NVL72 (NVIDIA)

Architecture

Blackwell

Tdp W

1200

Interconnect

Fifth-generation NVLink: 1.8 TB/s; PCIe Gen5: 128 GB/s

Specifications

Blurred values are subscriber-only. Sign in to start a free trial — 2 full pages a day — or subscribe for unlimited access.

Memory Bandwidth Tb S	8
Interconnect Bandwidth Gb S	130
Interconnect	Fifth-generation NVLink: 1.8 TB/s; PCIe Gen5: 128 GB/s
Tdp W	1200
Memory Type	HBM3E
Architecture	Blackwell
Fp16 Bf16 Tflops Dense	2500
Fp8 Tflops Dense	5000
Fp16 Bf16 Tflops Sparse	5000
Fp32 Tflops	80
Int8 Tops Dense	5000
Fp4 Tflops Dense
Fp4 Tflops Sparse
Fp64 Tflops
Fp8 Tflops Sparse
Int8 Tops Sparse
Media Engines
Product Name

NVIDIA Blackwell GPUs \| Grace CPUs
CPU Cores
Total NVFP4 Tensor Core
Total FP8/FP6 Tensor Core
Total Fast Memory
Total Memory Bandwidth
FP4 Tensor Core (Individual Blackwell GPU)
FP8/FP6 Tensor Core (Individual Blackwell GPU)
INT8 Tensor Core (Individual Blackwell GPU)
FP16/BF16 Tensor Core (Individual Blackwell GPU)
TF32 Tensor Core (Individual Blackwell GPU)
FP32 (Individual Blackwell GPU)
FP64/FP64 Tensor Core (Individual Blackwell GPU)
GPU Memory \| Bandwidth (Individual Blackwell GPU)
Multi-Instance GPU (MIG)
Decompression Engine
Server Options
Key Feature - Grace CPUs
Key Feature - Blackwell GPUs
Key Feature - LPDDR5X memory
Key Feature - HBM3E support
Key Feature - Fast-access memory
Key Feature - NVLink domain
Architecture description
Real-time LLM inference speedup
GPT-MoE-1.8T Real-Time Throughput (Output Tokens per Second per GPU)
NVLink GPU-to-GPU interconnect (fifth-generation)
LLM training speedup
GPT-MoE-1.8T Model Training Speedup
Database query speedup
Database Join Query (Queries per Second)
Energy efficiency vs H100
Energy Efficiency (Relative)