NVIDIA publishes no per-GPU board power for the NVL72 bin, only rack power of approximately 120 kW. tdp_watts is deliberately NULL rather than carrying a contradicted third-party figure.

NVIDIA GB200 NVL72 GPU 186GB

Blackwell SXM 2024 4nm Spec confidence: Official
FP8 (dense)
5,000
TFLOPS
FP32
80
TFLOPS
VRAM
186 GB
HBM3e
Bandwidth
8.0 TB/s
memory
Download datasheet

Overview

This is a single Blackwell GPU as deployed inside a GB200 NVL72 rack, which is the unit cloud providers actually rent and quote. It is the same silicon as the HGX B200 but a different bin: NVIDIA rates the liquid-cooled rack part about 11 percent higher on every tensor and shader row, and gives it 186 GB of HBM3e against the air-cooled board's 180 GB. Dense NVFP4 is 10 PFLOPS here against the HGX part's 9. NVLink 5 carries 1.8 TB/s per GPU, which is what makes a 72-GPU coherent domain practical in the first place. Two of these plus one Grace CPU make up the GB200 Grace Blackwell Superchip, so that row's figures are pair totals and sit at roughly double the ones here. If you are comparing rental prices, this is the row to compare against; if you are costing rack hardware, the superchip row is the building block.

Performance

Peak theoretical throughput by precision type

Precision
Dense
2:4 Sparse
FP64
64-bit floating point
40TFLOPS
Structured sparsity is a tensor-core feature; this vector precision has no sparse form
FP32
32-bit floating point
80TFLOPS
Structured sparsity is a tensor-core feature; this vector precision has no sparse form
TF32
TensorFloat-32
1,250TFLOPS
2,500TFLOPS
BF16
Brain Float 16
2,500TFLOPS
5,000TFLOPS
FP16
16-bit floating point
2,500TFLOPS
5,000TFLOPS
FP8
8-bit floating point
5,000TFLOPS
10,000TFLOPS
FP6
5,000TFLOPS
10,000TFLOPS
FP4
4-bit floating point
10,000TFLOPS
20,000TFLOPS
INT8
8-bit integer
5,000TOPS
10,000TOPS

Specifications

Architecture

Blackwell

Form Factor

SXM

Launch Year

2024

Process Node

4nm

Memory

186 GB HBM3e

Bandwidth

8,000 GB/s

TDP

No verified data available

Max Power

No verified data available

Transistors

208.0 billion

Interconnect

1.8 TB/s NVLink

Spec Confidence

Official

Full Specifications

Chip Design
Transistors 208.0 billion
Process Node 4nm
Chiplets 2 (Two reticle-limited dies joined by a 10 TB/s die-to-die link, presented to software as one GPU)
Memory
VRAM 186 GB
Memory Type HBM3e
Bandwidth 8.0 TB/s
Interface Width No verified data available
Interconnect & I/O
GPU-to-GPU NVLink
Interconnect Bandwidth 1.8 TB/s
Power & Thermal
TDP No verified data available
Max Board Power The vendor publishes no maximum board power; the Max Power tile above estimates it from TDP
Cooling Liquid-cooled
Enterprise Features
ECC Memory Yes
Sparsity Yes
General
Form Factor SXM
Architecture Blackwell
Process Node 4nm
Launch Year 2024

Datasheet & Resources

Data Provenance

Every figure traced to a source

Primary Source

Publisher
NVIDIA
Published
2024-03-18

Data Quality

Spec confidence
Official
Clock basis
Boost
Core precisions with figures
9 of 9
Normalization
All values in TFLOPS

Compare cloud providers offering on-demand GPU instances for AI training, inference, and HPC workloads.

Browse GPU Cloud Providers

Similar GPUs

Frequently Asked Questions

How many TFLOPS does the NVIDIA GB200 have?

The NVIDIA GB200 delivers 80 TFLOPS FP32, 2,500 TFLOPS FP16 and 5,000 TFLOPS FP8 at peak.

What is the power consumption of the NVIDIA GB200?

Flopper does not currently have a verified TDP (Thermal Design Power) figure for the NVIDIA GB200.

How much memory does the NVIDIA GB200 have?

The NVIDIA GB200 is equipped with 186 GB of VRAM with 8,000 GB/s of memory bandwidth.

What architecture is the NVIDIA GB200 based on?

The NVIDIA GB200 is based on the Blackwell architecture, launched in 2024.

Stay Updated on GPU Releases

Get notified when new GPUs are added or specifications are updated.

Loading verification...

No spam, unsubscribe anytime.

Back to GPUs