NVIDIA HGX B300 8-GPU Baseboard

The NVIDIA HGX B300 8-GPU Baseboard is the Blackwell Ultra platform for AI data centers, carrying eight NVIDIA B300 SXM GPUs with 288 GB of HBM3e each, up to 2.1 TB per baseboard. It combines up to 144 PFLOPS of FP4 Tensor Core performance (108 PFLOPS dense) with 1.8 TB/s of GPU-to-GPU bandwidth over fifth-generation NVLink and NVSwitch, and integrates eight NVIDIA ConnectX-8 SuperNICs at up to 800 Gb/s per GPU for scale-out networking.

SKU: HGX-B300-8GPU Categories: Data Center GPUs, GPU Brand: NVIDIA

Technical Specifications
Overview

Product SKU	HGX-B300-8GPU
Form Factor	8x NVIDIA Blackwell Ultra SXM
GPU	8x NVIDIA B300 SXM
GPU Memory	288 GB HBM3e per GPU
Total Memory	2.1 TB HBM3e per 8-GPU baseboard
Memory Bandwidth	Up to 8 TB/s per GPU (up to 64 TB/s per node)
GPU Architecture	NVIDIA Blackwell Ultra
FP4 Tensor Core	144 PFLOPS sparse / 108 PFLOPS dense
FP8/FP6 Tensor Core	72 PFLOPS sparse / 36 PFLOPS dense
INT8 Tensor Core	3 POPS sparse / 1.5 POPS dense
FP16/BF16 Tensor Core	36 PFLOPS sparse / 18 PFLOPS dense
TF32 Tensor Core	18 PFLOPS sparse / 9 PFLOPS dense
FP32	600 TFLOPS
FP64 / FP64 Tensor Core	10 TFLOPS
NVIDIA NVLink	Fifth generation, 1.8 TB/s GPU-to-GPU, 14.4 TB/s total
NVIDIA NVLink Switch	NVLink 5 Switch
Networking	8x NVIDIA ConnectX-8 SuperNICs, up to 800 Gb/s per GPU (2 x 400 Gb/s)
Networking Bandwidth	1.6 TB/s
Attention Performance	2x vs. HGX B200
PCI Express	8x Gen5 x16 and 1x Gen4 x2 per baseboard
Supported CPU Platform	Dual x86 CPU sockets, minimum 48 physical cores per socket (56 recommended), 2.0 GHz minimum base clock
System Memory	Minimum 2 TB per node, minimum 500 GB/s bandwidth
DPU	1x NVIDIA BlueField-3 DPU per server (north/south)
Local Storage	1-2 TB NVMe per CPU socket plus 1 TB NVMe boot drive (NVIDIA Enterprise RA recommendation)
Cooling	Air-cooled or direct-to-chip liquid, per OEM system design
Security	TPM 2.0 module (secure boot)
System Management	SMBPBI over SMBus (OOB) to BMC, PLDM T5-enabled, SPDM-enabled
Target Workloads	Large language models, traditional deep learning inference, HPC
NVIDIA AI Enterprise	Add-on

Overview

The NVIDIA HGX B300 8-GPU Baseboard is the NVIDIA Blackwell Ultra platform behind NVIDIA-Certified B300 systems that are not DGX systems. It integrates eight NVIDIA B300 SXM GPUs on one baseboard connected by fifth-generation NVLink and an NVLink 5 Switch, giving 288 GB of HBM3e per GPU — 2.1 TB across the board — and up to 64 TB/s of aggregate memory bandwidth. It is built for the largest training, fine-tuning and inference workloads.

At a glance

GPUs: 8x NVIDIA B300 SXM (Blackwell Ultra); 288 GB HBM3e per GPU, 2.1 TB per baseboard
Memory bandwidth: up to 8 TB/s per GPU, up to 64 TB/s per node
FP4 Tensor Core: 144 PFLOPS sparse / 108 PFLOPS dense
FP8/FP6 Tensor Core: 72 PFLOPS sparse / 36 PFLOPS dense
INT8: 3 POPS sparse / 1.5 POPS dense; FP16/BF16: 36 PFLOPS sparse / 18 PFLOPS dense; TF32: 18 PFLOPS sparse / 9 PFLOPS dense
FP32: 600 TFLOPS; FP64 / FP64 Tensor Core: 10 TFLOPS
NVLink: fifth generation, 1.8 TB/s GPU-to-GPU, 14.4 TB/s total; NVLink 5 Switch
Networking: 8x ConnectX-8 SuperNICs, up to 800 Gb/s per GPU (2×400 Gb/s); 1.6 TB/s
PCIe: 8x Gen5 x16 and 1x Gen4 x2; DPU: 1x BlueField-3 per server
Host platform: dual x86 sockets, minimum 48 cores per socket (56 recommended), minimum 2 TB system memory
Cooling: air or direct-to-chip liquid, per OEM system design; TPM 2.0 secure boot

When to choose the HGX B300

Choose this baseboard when the target is frontier-scale training, large fine-tuning or high-throughput inference, and the eight GPUs must behave as a single accelerator with maximum NVLink bandwidth. Per NVIDIA, attention-heavy workloads see up to 2× the performance of the previous HGX B200 generation. Where a smaller footprint or lower power is enough, a four-GPU HGX platform or PCIe-class accelerators are the more economical route.

Platform & compatibility

The baseboard is integrated by OEMs into NVIDIA-Certified B300 server designs (air- or liquid-cooled) and is not a standalone card. NVIDIA specifies a dual x86 platform with at least 48 cores per socket, 2 TB of system memory, 1–2 TB NVMe per socket plus a 1 TB boot drive, and a BlueField-3 DPU. Management uses SMBPBI over SMBus, PLDM T5 and SPDM. NVIDIA AI Enterprise is available as an add-on.

FAQ

Is this a GPU card? No — it is an 8-GPU baseboard that OEMs build into certified servers; end users buy a complete system, not a plug-in card.

How much memory does a node have? 288 GB of HBM3e per GPU, 2.1 TB per eight-GPU baseboard.

Air or liquid cooling? Both are supported and chosen by the OEM system design.

Configuration, availability and lead time depend on the required quantity, server pairing and delivery destination. Send your requirement to receive a quotation.