Skip to content
NVIDIA HGX B300 8-GPU Server

NVIDIA HGX B300 8-GPU Server

Eight Blackwell Ultra SXM GPUs with 2.1 TB HBM3e on NVLink 5; chassis and configuration agreed per order

Model
HGX B300 8-GPU Server
Condition
New, original packaging
Quantity
Ask on WhatsApp

Price on request. Tell us the quantity and destination and we reply within one working day.

How we supply this item

Supplied through licensed partners. Server, networking and data-centre hardware is supplied through licensed partners of InTechnologies FZCO. The partner issues the contract and invoice; we handle sourcing, consolidation in Dubai and delivery coordination. Export compliance

Export compliance. This product is subject to US export regulations. How we handle it is described on the export compliance page. Export compliance

An HGX B300 server is built on NVIDIA’s Blackwell Ultra eight-GPU HGX baseboard. Its eight Blackwell Ultra SXM GPUs share 2.1 TB of HBM3e over fifth-generation NVLink with 1.8 TB/s per GPU and 14.4 TB/s in total, and networking bandwidth reaches 1.6 TB/s. NVIDIA lists 108 PFLOPS of dense FP4 (144 PFLOPS sparse), 72 PFLOPS of FP8 and 36 PFLOPS of FP16/BF16 Tensor Core throughput for the board, with attention performance doubled compared with Blackwell. FP64 is deliberately low at 10 TFLOPS, so this is a platform for AI training and reasoning-model inference rather than classical HPC.

OEMs ship HGX B300 in air-cooled and direct-liquid-cooled chassis, typically 8U to 10U, and the processors, memory, storage, power and cooling vary between them. We therefore leave those rows open: send us the model sizes you plan to run, the number of nodes, rack power per cabinet and whether the site has liquid cooling, and we return configurations from the vendors that can deliver in your window.

InTechnologies FZCO arranges HGX B300 servers for business customers in the UAE, the GCC and Central Asia. Contracts and invoices are issued by our licensed partners; we handle sourcing, consolidation in Dubai Airport Freezone and DDP or DDU delivery. Blackwell Ultra GPUs are controlled under US export regulations, and every order goes through end-user verification.

Specifications

NVIDIA HGX B300 8-GPU Server: Specifications
GPUs8× NVIDIA Blackwell Ultra SXM on an HGX B300 baseboard
Total GPU memory2.1 TB HBM3e
GPU-to-GPU interconnectNVLink 5, 1.8 TB/s per GPU, 14.4 TB/s total
FP4 Tensor Core108 PFLOPS dense, 144 PFLOPS sparse (8 GPUs)
FP8 / FP6 Tensor Core72 PFLOPS sparse (8 GPUs)
Networking bandwidth1.6 TB/s
Server vendor and chassis—
Processors—
System memory—
Storage—
Power supplies and cooling—

— the vendor, chassis and these components are agreed per order. Message us your workload, quantity and rack power, and we will send the configurations we can supply. Ask on WhatsApp

Questions buyers ask

Why are the vendor, CPU and memory shown as a dash?

The configuration is agreed per order. Several OEMs build HGX B300 servers, with different chassis, cooling and component options, and allocation changes by quarter. Send us a message with your requirements and we will propose the vendors and builds we can supply.

How does HGX B300 differ from HGX B200?

B300 uses Blackwell Ultra GPUs with 2.1 TB of total HBM3e instead of 1.4 TB, higher dense FP4 throughput and twice the attention performance, while FP64 is much lower. For FP64-heavy HPC, B200 or H200 may fit better.

Do I need liquid cooling?

Not always. Air-cooled HGX B300 chassis exist, but they need high power per rack and strong airflow; liquid-cooled versions are denser. Tell us your rack power and cooling and we will match the chassis.

Is HGX B300 export-controlled?

Yes. Our licensed partner reviews an end-user statement and, where required, applies for a US export licence before the order is accepted.

NVIDIA

NVIDIA HGX H200 8-GPU Server

Eight H200 SXM GPUs with 141 GB HBM3e each on NVLink; chassis and configuration agreed per order