NVIDIA DGX B300

rack
8× B300
NVSwitch + NVLink
2025
FP64
0.0
PFLOPS
FP32
0.6
PFLOPS
Power
14.0
kW Total
Memory
2100
GB Total

Eight Blackwell Ultra SXM GPUs on the HGX B300 baseboard with Intel Xeon 6776P processors, in a 10U chassis. NVIDIA quotes 144 PFLOPS of FP4 inference performance for this machine, which is the sparse figure; the dense equivalent is 108 PFLOPS, and both are recorded. Power is given by NVIDIA as approximately 14 kW.

We will point you at suppliers who have it. Free, and no signup.

FP64
0.01
PFLOPS
FP32
0.60
PFLOPS
TF32
9.00
PFLOPS
FP16
18.00
PFLOPS

System Details

GPU Configuration

GPU Count: 8 GPUs
Architecture: Blackwell Ultra
Interconnect: NVSwitch + NVLink

System Specifications

Form Factor: rack
Total Power: 14.0 kW
Total Memory: 2.1 TB
Memory Bandwidth: 64 TB/s

Precision Performance Breakdown

PrecisionSystem PerformancePer GPUEfficiency
0.010 PFLOPS 1.3 TFLOPS
0.600 PFLOPS 75.0 TFLOPS
9.000 PFLOPS 1125.0 TFLOPS
18.000 PFLOPS 2250.0 TFLOPS
18.000 PFLOPS 2250.0 TFLOPS
36.000 PFLOPS 4500.0 TFLOPS
108.000 PFLOPS 13500.0 TFLOPS
FP6 36.000 PFLOPS 4500.0 TFLOPS

Where a vendor states a basis, the figure shown is the dense one, and any figure it publishes is listed separately. Vendors commonly headline the sparse number instead: for NVIDIA's 2:4 structured sparsity that is exactly twice the dense figure, though other vendors' sparse modes do not all follow that ratio.

Powered by NVIDIA B300

This system utilizes 8 × NVIDIA B300 SXM 288GB GPUs, each delivering exceptional performance for AI and HPC workloads.

Per GPU TDP

1400W

Per GPU Memory

288 GB

Process Node

4nm

Architecture

Blackwell Ultra

Documentation & Resources

Flopper Spec Sheet

NVIDIA DGX B300 specifications, generated from our database. Printable.

Vendor product page

NVIDIA DGX B300 documentation

View ↗

workstation-datasheet-dgx-spark-gtc25-spring-nvidia-us-3716899-web.pdf

NVIDIA • 2025-10-16

View Document ↗

NVIDIA DGX Documentation

System guides, deployment resources

Browse ↗

Typical Use Cases

Large Language Model Training
Distributed Deep Learning
Multi-GPU Inference
HPC Simulations
Scientific Computing

The NVIDIA DGX B300 runs 8× B300 GPUs over NVSwitch + NVLink, delivering 36.0 PFLOPS FP8 dense.