DenseMAX Appliance
Dense, air-cooled AI compute for your own floor
Eight Blackwell GPUs, 24 DDR5 slots and a 400 Gbps fabric in one serviceable chassis.
Delivered with Surogate and our own serving and training engines, built for Blackwell and faster than the open-source stack on the same silicon.
- 8×
- 24×
- 400 Gbps
- 5× 2000 W
Technical Specifications
Everything you need for state-of-the-art generative AI workloads.
Compute & Memory
- CPU
- 2× AMD EPYC™ 9005, 9004 or 97x4 processors
- Memory
- 24× DDR5 ECC RDIMM slots
- Expansion
- 8× PCIe 5.0 ×16 GPU slots
- Accelerators
- Up to 8× NVIDIA RTX 5090 or RTX PRO 6000 Blackwell GPUs
Storage
- Internal
- 2× Internal M.2 NVMe (PCIe 3.0 ×4)
- Bays
- 4× U.2 NVMe bays (PCIe 5.0 ×4)
Networking
- Slots
- 2× PCIe 5.0 ×8 slots for networking (mergeable to 1× PCIe 5.0 ×16)
- Fabric
- 1× 400 Gbps InfiniBand network card
- Front I/O
- IPMI port and 2× 1 GbE RJ45
Power & Environment
- PSUs
- 5× 2000 W industrial-grade PSUs
- Operating temperature
- 10 °C – 35 °C
- Non-operating temperature
- −40 °C – 70 °C
- Non-operating humidity
- 20% – 90% (non-condensing)
I/O & Manageability
- Video
- 1× DB15 VGA port
- USB
- 4× Type-A (USB 3.2 Gen1)
- Management
- Dedicated IPMI for out-of-band management
Form Factor
- Chassis
- Datacenter-ready chassis with optimized airflow and serviceability
- Service
- Tool-assisted access for GPUs, storage, and power supplies
Surogate, pre-installed
The appliance ships with Surogate installed — build agents, run them on your own floor, and train experts on what they do.
Agent Studio
Models, knowledge bases, skills, MCP tools and guardrails — assembled without code.
Managed Runtime
Agents deployed on your own cluster and your own channels, escalating what needs a person.
Serving Engine
Our own inference engine, tuned for Blackwell — ahead of vLLM and llama.cpp on this hardware.
Train Experts
Fine-tune, align or distil a small model on your own traces — on the GPUs in this chassis.
Ready to deploy enterprise-grade AI on-prem?
Request a guided demo or talk to our team about configurations, pricing, and delivery.