Specifications
Model
ASRock Rack 8U16X-GNR2 B300
Form Factor
8U rackmount — 930 × 448 × 353.6 mm (36.6″ × 17.6″ × 13.9″)
GPU Baseboard
NVIDIA HGX™ B300 NVL16 with NVIDIA NVLink™
GPU Count
16 × NVIDIA Blackwell Ultra SXM
GPU Memory
NVIDIA HBM3e per Blackwell Ultra SXM module
CPU
Dual Socket E2 (LGA 4710) — Intel® Xeon® 6700P-series, 6500P-series and 6700E-series
Cooling
Air-cooled — designed for existing air-cooled data centre infrastructure
Power Supply
6+6 redundant 3000W CRPS, 80‑PLUS Titanium efficiency
Networking
Integrated high-speed GPU-to-GPU and scale-out fabric via the HGX baseboard NVLink domain
Use Case
AI factory training, fine-tuning and reasoning inference at rack scale
Deployment Target
Data centres with conventional air-cooled power and cooling envelopes
Availability
Quote-based; QS Compute supplies globally with volume pricing
Overview
The ASRock Rack 8U16X-GNR2 B300 is an 8U air-cooled AI server built around the NVIDIA HGX B300 NVL16 baseboard, populated with 16 NVIDIA Blackwell Ultra SXM GPUs. It is the air-cooled counterpart to the liquid-cooled and waterless-cooled B300 platforms in ASRock Rack’s portfolio — the 4U16X-GNR2/DLC direct-to-chip system and the 4U16X-GNR2/ZC waterless variant — and its purpose is to let operators deploy Blackwell Ultra capacity into data halls that are not plumbed for liquid.
The HGX B300 NVL16 architecture extends the NVLink domain to sixteen GPUs on a single baseboard, so each accelerator sees a much larger coherent GPU pool than an 8-GPU HGX system. That matters for mixture-of-experts training and long-context inference, where the working set does not fit in one GPU’s HBM and cross-GPU communication volume dominates step time. The air-cooled 8U chassis gives ASRock Rack the vertical volume needed to move that heat without liquid infrastructure.
Host compute is provided by dual Socket E2 (LGA 4710) processors from Intel® Xeon® 6700P-series, 6500P-series or 6700E-series families, keeping system I/O, memory capacity and PCIe lane count aligned with the GPU subsystem. Power is delivered by 6+6 redundant 3000W CRPS units at 80‑PLUS Titanium efficiency, giving both the headroom and the redundancy that dense AI racks require.
QS Compute sources the 8U16X-GNR2 B300 and the wider ASRock Rack HGX B300 family for data centre build-outs, including matched CPU, memory, NVMe and networking configurations. Both air-cooled and liquid-cooled deployment paths are available; contact us with your rack power and cooling envelope and we will return a configured quotation.
Key Benefits
16 NVIDIA Blackwell Ultra GPUs on a single HGX B300 NVL16 baseboard. Air-cooled 8U chassis — deploys into existing data centres without liquid infrastructure. Dual Intel® Xeon® 6700P-series host compute on Socket E2 (LGA 4710). 6+6 3000W Titanium CRPS for redundant high-density power delivery. Expanded NVLink domain suits mixture-of-experts training and long-context inference. Global B2B sourcing with matched memory, storage and fabric options.
Applications
Rack-scale AI factory training and fine-tuning; long-context and reasoning inference clusters; mixture-of-experts (MoE) model training; sovereign and enterprise AI cloud build-outs; scientific computing and large-scale simulation; air-cooled retrofits of existing GPU data halls.
Request a Quote — ASROCK RACK 8U16X-GNR2 B300 — 8U AIR‑COOLED NVIDIA HGX B300 NVL16 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 →