156GB VRAM // 128 PCIe LANES // 25GbE FABRIC

STOP RENTING AGENTS.
OWN THE CLUSTER.

Imagine opening your IDE and having Qwen 3.6 ingest your entire multi-repo codebase with zero thinning and near-instant inference. The ultimate dual-node AI Forge for local-first developers.

Dual Node Rig

System 1: The AI Heavyweight

Primary Compute / Inference & Training Node

  • Board: ASUS Pro WS WRX80E-SAGE SE WIFI (128 PCIe Gen4 Lanes)
  • CPU: AMD Threadripper PRO 3955WX (16-Core / 32-Thread)
  • RAM: 256GB G.Skill Trident Z 3600MHz
  • Cooling: Noctua NH-U14S (Dual Push/Pull)
  • Storage: 6TB NVMe (3x 2TB SN850X) + 20TB WD Ultrastar HDD

GPU Array (120GB VRAM)

  • 1x NVIDIA RTX 3090 Founders Edition (w/ Box)
  • 1x EVGA FTW3 ULTRA 3090 (w/ Box)
  • 1x EVGA FTW3 HYBRID 3090 (240mm AIO, w/ Box)
  • 1x EVGA XC HYBRID 3090 (240mm AIO)
  • 1x GIGABYTE AORUS XTREME WATERFORCE HYBRID 3090 (240mm AIO)

Power Delivery

  • Thermaltake Toughpower GF3 1650W (Primary)
  • Thermaltake Toughpower GF3 1350W (Secondary PCIe via Add2PSU)

System 2: The Dev Node

Secondary / RAG / Vector DB Support Node

  • Board: Gigabyte AORUS TRX40 Pro WiFi
  • CPU: AMD Threadripper 3960X (24-Core / 48-Thread)
  • RAM: 128GB G.Skill Trident Z 4000MHz
  • Cooling: Noctua NH-U9 (Dual Fan)
  • Storage: 2TB Samsung 990 PRO + 1TB WD Black SN850X

GPU Array (36GB VRAM)

  • 3x EVGA RTX 3060 XC (12GB models)

Power Delivery

  • Thermaltake Toughpower GF3 1350W
other stuff

Cluster Interconnect

These systems aren't just sitting next to each other. They are linked via a dedicated ultra-low latency backbone for massive multi-node tasks.


  • NICs: 2x Mellanox ConnectX-4 Lx (MCX4121A-ACAT) 25GbE SFP28 Adapters
  • Capabilities: Full support for RoCE v2 (RDMA)
  • The Edge: Lightning-fast model weight distribution and direct memory access between the WRX80 and TRX40 nodes.

Extreme Thermal Management

Housed in the massive Thermaltake W200, featuring a custom 31-fan high-pressure array to ensure zero thermal throttling.


  • 10x Noctua 3000 iPPC (120mm)
  • 7x Noctua 3000 iPPC (140mm)
  • 6x Phanteks T30 (120mm)
  • 8x Arctic P12 Max
  • Control: Lamptron FC5v3 + 2x Arctic 10-port SATA Hubs
  • Mounts: Premium LINKUP & EZ-DIYFAB PCIe 5.0/4.0 Risers

System Stability Verified

This rig is built for production, not just benchmarks. 100% pass on gpu-burn and cuda-memtest across all 8 GPUs. CPUs verified stable with 24-hour stress-ng runs. Flawless VRAM integrity.

other stuff
$14,600 OBO

Location: Clarkson, NY

Local pickup only. This unit is an absolute tank (approx. 200lb+). Please bring a vehicle with appropriate cargo space. Can be demonstrated running local inference upon viewing.

Contact to Purchase