Specifications
Model
G894-SD3-AAX7
Form Factor
8U rackmount, air-cooled
GPU Platform
NVIDIA HGX B300
GPU Count
8x NVIDIA Blackwell Ultra
GPU Interconnect
NVIDIA NVLink + NVLink Switch
GPU-GPU Bandwidth
1.8 TB/s
Total GPU Bandwidth
14.4 TB/s
Total GPU Memory
Over 2 TB HBM3E
AI Performance
7x over the Hopper platform
CPU
Dual Intel Xeon 6700 / 6500 series
Memory Slots
32x DDR5 DIMM
Networking
8x 800 Gb/s OSFP InfiniBand or 16x 400 Gb/s Ethernet
NIC
NVIDIA ConnectX-8 SuperNIC
DPU Support
NVIDIA BlueField-3
Expansion
4x PCIe Gen5 x16 slots
Power Supply
12x 3000 W 80 PLUS Titanium redundant
Cluster Platform
GIGAPOD AI & HPC infrastructure
Launched
18 September 2025
Overview
The GIGABYTE G894-SD3-AAX7 is Giga Computing's flagship 8U server, built around the NVIDIA HGX B300 platform. Eight NVIDIA Blackwell Ultra GPUs are integrated with NVIDIA NVLink and NVLink Switch to deliver 1.8 TB/s of GPU-to-GPU bandwidth and 14.4 TB/s of total bandwidth from more than 2 TB of HBM3E memory — quoted at seven times the AI performance of the previous Hopper generation.
Dual Intel Xeon 6700 or 6500 series processors drive 32 DDR5 DIMM slots. Networking is delivered through NVIDIA ConnectX-8 SuperNICs as either eight 800 Gb/s OSFP InfiniBand ports or sixteen 400 Gb/s Ethernet ports, with NVIDIA BlueField-3 DPUs supported for offloaded infrastructure workloads.
Four PCIe Gen5 x16 slots provide expansion, and power comes from twelve 3000 W 80 PLUS Titanium redundant supplies inside a compact 8U air-cooled chassis. The system extends the GIGAPOD high-performance AI cluster portfolio, letting a deployment scale from a single server to an ultra-large cluster on the same platform.
Key Benefits
Eight Blackwell Ultra GPUs in 8U with full NVLink Switch fabric for scale-out training. Over 2 TB HBM3E total GPU memory for large-model inference and long-context workloads. Flexible fabric — 800 Gb/s InfiniBand or 400 Gb/s Ethernet on ConnectX-8. 4x PCIe Gen5 x16 slots plus BlueField-3 DPU support for storage and infrastructure offload. Titanium-class efficiency from twelve 3000 W redundant supplies. Air-cooled for halls without liquid infrastructure.
Applications
Large language model training and fine-tuning, AI agent and reasoning inference services, real-time multimedia generation, scientific simulation, financial modelling, and HPC workloads across healthcare, manufacturing, finance and research.
Request a Quote — GIGABYTE G894-SD3-AAX7 — 8U NVIDIA HGX B300 8-GPU AI SERVER
QS Compute — global B2B supply of AI computing hardware, edge AI systems and accelerators. Volume pricing, 15-day sample lead time.
Get Your Quote →