HYPERPC AMPERE XR ULTRA
8 GPUs in 4U. Liquid cooling. No external CDU.
A high-density barebone platform for on-premises AI, model training, and high-performance computing. It fits into a standard server room and scales from 2 to 8 accelerators.
HYPERPC AMPERE XR ULTRA
8 GPUs in 4U. Liquid cooling. No external CDU.
A high-density barebone platform for on-premises AI, model training, and high-performance computing. It fits into a standard server room and scales from 2 to 8 accelerators.
Not an off-the-shelf box, but the foundation of your solution
AMPERE XR ULTRA provides a proven barebone platform in which the chassis, power system, and liquid cooling loop work as a single system. The integrator selects the motherboard, processors, GPUs, memory, storage, and networking to match the workload.
Motherboard agnostic
EATX, EEB, CEB, and TEB. AMD SP5 or Intel LGA 4677 / 4710.
From 2 to 8 GPUs
Start with your current workload and expand the system without replacing the chassis.
Built in Slovenia
Engineering assembly and cooling-loop testing in Komenda, Slovenia.
Control where the cloud falls short
An on-premises platform puts your team back in control of budgets, data, latency, and the experimentation environment.
Data sovereignty
Minimal latency
Secure sandbox
8 GPUs without compromising cooling
Single-slot water blocks free up space inside the 4U chassis, while a unified loop removes heat from the GPUs and CPUs under sustained full load.
Validated with 8 × NVIDIA H200 NVL
Testing at an ambient temperature of 30 °C confirmed stable thermals under full load. Final values depend on the selected components and server-room conditions.
- Up to 700 W of continuous power per slot
- Up to 8000 W of continuous platform power
- No external CDU or connection to the building water supply
All cooling is contained within 4U
The pump, manifold, radiator, reservoir, and water blocks form a pre-filled, tested loop. Quick-disconnect fittings allow individual components to be serviced without dismantling the entire system.
Platform-grade reliability
The power system and cooling loop are engineered at the chassis level for high-density workloads, redundancy, and servicing without prolonged downtime.
5 × 2000 W Titanium
Redundant power supplies and a dedicated PDU deliver power to the platform independently of the motherboard VRM. Maximum continuous chassis power is up to 8000 W.
- 80 PLUS Titanium and redundancy
- Up to 700 W continuous power per GPU slot
- 5 × 2.5-inch drive bays
Replace a GPU without dismantling the entire loop
Quick-disconnect fittings isolate the required component. The service station performs a complete cycle of leak testing, draining, cleaning, flushing, vacuum filling, and inspection.
Only the approved EK-CryoFuel engineered coolant is used.
Start with 2 GPUs. Scale up to 8.
One platform supports every stage of team maturity, from departmental inference to production AI and HPC.
Department
2 × RTX 6000 Ada
1 × AMD EPYC 9354
256 GB ECC
Inference, visualization, and initial on-premises models
Research group
4 workstation/HPC-class GPUs
Dual-socket platform
512 GB ECC
Shared research and fine-tuning environment
Enterprise AI
8 × NVIDIA H200 NVL
2 × Xeon Platinum 8580
1 TB ECC
Training, large models, and HPC workloads
The configurations shown are for guidance only. Compatibility, cooling, and final continuous power are validated for each project before a proposal is issued.
The AMPERE XR ULTRA platform. A HYPERPC solution.
Clearly divided responsibilities preserve component flexibility while giving the customer a single point of contact for design, integration, and support.
LM TEK
Platform engineering
- Chassis, PDU, and redundant power
- Pre-filled liquid cooling loop
- Compatibility and thermal-capacity validation
HYPERPC
Complete server design
- Selection of CPU, GPU, memory, networking, and storage
- Assembly, burn-in testing, and software stack configuration
- Deployment, ongoing support, and a unified SLA
The platform's technical foundation
Barebone platform specifications. The final specification is confirmed after the selected BOM has been validated.
| Parameter | Specification |
|---|---|
| Form factor | 4U, 19-inch rack |
| Dimensions and weight | 760 × 482 × 178 mm; approximately 38 kg in barebone configuration |
| GPU | Up to 8 full-size PCIe Gen 5 GPUs; up to 700 W continuous power per slot |
| Power | 5 × 2000 W 80 PLUS Titanium, redundant; up to 8000 W sustained |
| Motherboards | EATX / EEB / CEB / TEB |
| Processor platforms | AMD SP5; Intel LGA 4677 / 4710 / 4710-2 |
| Memory and storage | Depends on the selected motherboard, typically up to 24 DDR5 RDIMMs; 5 × 2.5-inch drives |
| Cooling | Closed loop within the chassis, EK-CryoFuel, quick-disconnect fittings; no external CDU required |
| Operating conditions | Air: 5–30 °C; coolant: 5–60 °C |
| SKU / manufacturing | 3831129902367; engineered & assembled in Komenda, Slovenia |
Frequently asked questions about the platform
Key engineering requirements are clarified during the configuration design stage.
- Is an external CDU required? No. The LM TEK 4U8GPU10KW uses a closed liquid cooling loop inside the 4U chassis, with the pump, manifold, radiator, reservoir, and water blocks already integrated. No connection to an external CDU or the building water supply is required.
- Can I start with 2 GPUs? Yes. The platform can be configured with 2 GPUs and later expanded to 8 accelerators without replacing the chassis. Compatibility of the selected cards and cooling capacity are validated before a proposal is prepared.
- Which accelerators are supported? The platform supports up to 8 full-size PCIe Gen 5 GPUs rated at up to 700 W per slot. Supported options include NVIDIA RTX 6000 Ada, RTX A6000, A100, H100 NVL, H200 NVL, and RTX PRO 6000 Blackwell; other models are reviewed separately.
- What is included in the barebone platform? The barebone platform includes a 4U chassis, rails, front panel, integrated liquid cooling loop, five 2000 W 80 PLUS Titanium power supplies, a PDU, water blocks, quick-disconnect fittings, and documentation. The motherboard, CPU, GPU, memory, storage drives, and network cards are selected separately.
- How is the liquid cooling loop serviced? Quick-disconnect fittings make it possible to isolate and replace a specific GPU without dismantling the entire loop. Service includes leak testing, draining, cleaning, flushing, and vacuum filling; only approved EK-CryoFuel coolant is used.
- What are the server-room requirements? A standard 19-inch rack, adequate electrical capacity, and room cooling sized for the selected configuration are required. The operating inlet-air temperature is 5–30 °C; no external CDU or building water supply is needed. Maximum continuous platform power is up to 8000 W.
We will configure the platform for your model and infrastructure
We will define the workload, available rack power, and GPU and storage requirements, then provide a validated BOM and deployment plan.