Chassis/Form Factor
- 10U Rackmount
- Dimensions: 447 mm (W) x 441.75 mm (H) x 800 mm (D)
- Air-cooled design for easy integration into existing data centers
Processor
Dual Intel® Xeon® 6 processors (e.g., 6767P model: 2.4 GHz, 64-core each, up to 350W TDP)
Supports one single 8-GPU accelerator baseboard
GPUs/Accelerators
- 8x NVIDIA HGX B300 GPUs (Blackwell Ultra architecture)
- Performance: 72 PFLOPS FP8 training, 144 PFLOPS FP4 inference)
- Memory: Up to 288 GB HBM3e per GPU (total ~2.3 TB)
Ideal for agentic AI, AI reasoning, and real-time video generation
Memory
- 32x DDR5 RDIMM slots, up to 6,400 MHz
- Prebuilt config: 2TB (32x 64GB DDR5-6400MHz RDIMMs)
Storage
- 8x hot-swappable E1.S PCIe Gen 5 SSDs
- 2x PCIe M.2 2280 SSDs (boot drives)
- Prebuilt config: 2x 1.92TB NVMe M.2 SSD boot drives + 8x 3.84TB PCIe Gen 5 \
- NVMe SSDs (total ~30TB high-speed I/O)
Networking
- 8x OSFP ports (single-port NVIDIA ConnectX-8 VPI: 800G Ethernet or InfiniBand)
- Prebuilt config: NVIDIA ConnectX-7 200Gb/s single-port SFP
- 1x dedicated RJ45 LAN for BMC management + 1x 1GbE LAN (RJ45)
Expansion
- 4x FHHL SW or 2x FHHL DW PCIe Gen 5 x16 slots (CXL 2.0 ready)
- Prebuilt config: 4x FHHL PCIe Gen 5 x16 slots
- PCIe Gen 6 for high-speed east-west data transfer in massive GPU clusters
Management
- QCT BMC with Redfish API (1GbE LAN)
- In-house OpenBMC for comprehensive monitoring
Power Supply
- 3600W Titanium Level, redundant (1+1)
- Supports wide range of PSUs (800W to 3200W) for efficiency
