Overview
The GodSpeed Nucleon is a performance-tuned AI workstation for researchers, developers, and creators who need serious parallel GPU acceleration without compromising stability. AMD’s Ryzen™ 9 9950X3D (Zen 5 + 3D V-Cache) delivers snappy, cache-rich preprocessing and orchestration, while two NVIDIA® RTX™ 5090 GPUs provide massive Tensor throughput for deep learning, generative AI, and large-scale rendering. With ECC memory, a 4TB Gen4 NVMe, and quiet liquid cooling, Nucleon stays responsive under heavy, continuous loads.
Key Features
Dual RTX 5090 performance: Scales training and inference across two next-gen RTX 50-series GPUs
3D V-Cache compute: 16C/32T Ryzen 9 9950X3D excels at cache-sensitive pipelines and data prep
ECC system memory: Error-corrected stability for long training runs and mission-critical work
4TB Gen4 NVMe: High I/O for datasets, checkpoints, and scratch
Thermals that scale: 360mm liquid cooling + tuned chassis airflow for sustained boost clocks
CUDA-ready: Prepped for CUDA, cuDNN, TensorRT, and NCCL with PyTorch/TensorFlow options
Specifications
CPU: AMD Ryzen™ 9 9950X3D (16 cores / 32 threads, Zen 5, 3D V-Cache)
GPU: 2× NVIDIA® RTX™ 5090 (GeForce RTX 50 Series)
System Memory: 128GB DDR5 ECC (UDIMM)*
Primary Storage: 4TB PCIe Gen4 NVMe SSD
Motherboard: X670E workstation-class board (PCIe Gen5, ECC support)
Cooling: 360mm liquid CPU cooler, performance-tuned fan curves
PSU: 1600W 80+ Platinum, fully modular (dual high-power GPU connectors)
Chassis: Full-tower AI workstation (dual-GPU optimized)
Networking: 2.5GbE onboard (10GbE optional)
OS Options: Windows 11 Pro or Ubuntu LTS (CUDA stack)
Warranty: 3-Year HydraCluster Limited Warranty (Parts & Labor)
* Platform note: AM5/X670E supports ECC UDIMM. RDIMM requires a Threadripper PRO (WRX90) platform. Multi-GPU scaling is achieved via data-parallel frameworks (e.g., PyTorch DDP/NCCL); NVLink availability varies by generation/model.
What’s in the Box
GodSpeed Nucleon Workstation (assembled, burn-in tested)
AC power cable
Accessory kits (motherboard/GPU)
Quick-start & recovery guide
Warranty & support card
Options/Upgrades
Memory: 192GB–256GB DDR5 ECC (kit-dependent)
Storage: Add 2–8TB NVMe; optional RAID1/10; dedicated scratch SSD
Networking: 10GbE/25GbE NICs for NAS-based datasets
Thermals: High-static-pressure fan kit; acoustic damping
Software Image: Preinstall NVIDIA drivers, CUDA Toolkit, cuDNN, TensorRT, PyTorch/TensorFlow, Docker + NVIDIA Container Toolkit, W&B, HF CLI
Platform Upgrade: Threadripper PRO (WRX90) for more PCIe lanes and ECC RDIMM support
Warranty
3-Year HydraCluster Limited Warranty (Parts & Labor) with lifetime U.S.-based technical support. Advanced-replacement options available.
Lead Time
Built to order. Standard lead time 7–10 business days (includes dual-GPU burn-in and thermal profiling). Rush build available.


