---
title: "RM-4U8G - Integrated 8-GPU Server Platform · 4U Liquid-Cooled Barebone"
description: "A barebone AI server platform: eight full-height PCIe Gen 5 GPU slots in a 4U chassis, engineered around a closed liquid cooling loop with an in-chassis radiator, so it deploys in a standard server room without facility plumbing or an external CDU. Platform-level power, no motherboard lock-in. Cooling designed by the EK engineering team; engineered and assembled in Komenda, Slovenia."
url: "https://www.lmtek.com/ai-hpc-servers/rm-4u8g"
locale: "en"
---

# RM-4U8G

_ai-server-platform_

A barebone AI server platform: eight full-height PCIe Gen 5 GPU slots in a 4U chassis, engineered around a closed liquid cooling loop with an in-chassis radiator, so it deploys in a standard server room without facility plumbing or an external CDU. Platform-level power, no motherboard lock-in. Cooling designed by the EK engineering team; engineered and assembled in Komenda, Slovenia.

A barebone platform: LM TEK supplies the chassis, the integrated cooling loop, and platform-level power. You specify the board, CPUs, GPUs, memory, and storage.

## Specifications

| Spec | Value |
| --- | --- |
| Chassis | 4U |
| GPU slots | 8 |
| Cooling | Integrated liquid · in-chassis |
| Max chassis power | Up to 8000 W sustained |
| CPU sockets | AMD SP5, Intel LGA 4677 / LGA 4710 / LGA 4710-2 |

## Proof and deployment

### The honest envelope

- Deploys in standard server-room conditions: 5 to 35°C ambient, 5 to 60°C coolant.
- No facility plumbing. No external CDU.
- Five redundant 2000 W 80 PLUS Titanium PSUs, up to 8000 W sustained.
- You still provision rack power and rack cooling headroom.

### Plan for the heat

A dense GPU server converts nearly all of its electrical draw into heat in the room. In extended thermal torture testing, a fully populated RM-4U8G measured 6800 to 7000 W sustained draw. That heat load is defined by the GPUs and CPUs in the configuration, the same as any other max-spec 8-GPU server. Without added room cooling, ambient will rise under sustained load. The RM-4U8G’s operating envelope extends to 35°C ambient, so a room that warms under load stays inside specification, but scope rack cooling for the configured load before deployment.

In LM TEK torture testing at 30°C ambient, GPU package temperatures held at 76 to 81°C, inside the envelope a real server room can offer.

Acoustic measurements under sustained load are available on request.

### Efficiency, and the water question

Power conversion uses 80 PLUS Titanium supplies, the highest rating in the certification; at multi-kilowatt sustained draw, conversion losses are a real operating cost and an increasingly common bid and tender criterion. And the water objection to AI infrastructure does not apply here: the loop is closed, filled once at assembly with engineered coolant, and consumes no water in operation. There is no facility water connection to consume from.

## Reference configurations

| Tier | GPU | CPU | RAM | Storage | Use case |
| --- | --- | --- | --- | --- | --- |
| Modest start | 2× NVIDIA RTX 6000 Ada | 1× AMD EPYC 9354 (32-core) | 256 GB DDR5 ECC | 2× 3.84 TB NVMe | Workstation crossover · departmental AI · single-team inference |
| Mid-tier | 4× NVIDIA RTX A6000 | 2× AMD EPYC 9474F (48-core) | 512 GB DDR5 ECC | 4× 3.84 TB NVMe | Lab / research group · multi-user inference · light fine-tuning |
| Production maximum | 8× NVIDIA H200 NVL | 2× Intel Xeon Platinum 8580 | 1 TB DDR5 ECC | 8× 7.68 TB NVMe | Enterprise inference · production fine-tuning · multi-tenant |

## Technical datasheet

| Spec | Value |
| --- | --- |
| Form factor | 4U · 19" rack chassis |
| Dimensions (L × W × H) | 760 × 482 × 178 mm |
| Net mass (barebone) | 38 kg · approximate |
| GPU slots | 8 × full-height PCIe Gen 5 |
| GPU power envelope | Up to 700 W per slot · sustained |
| CPU sockets supported | AMD SP5 · Intel LGA 4677 · Intel LGA 4710 · Intel LGA 4710-2 · per chosen motherboard |
| Motherboard form factors | EATX · EEB · server reference |
| Memory slots | Per chosen motherboard · typically 24 × DDR5 RDIMM |
| Storage bays | 8 × NVMe U.2 · hot-swap · front access |
| Networking | Per chosen motherboard · OCP 3.0 mezzanine supported |
| Cooling type | Integrated liquid · in-chassis closed loop · in-chassis radiator · no external CDU |
| Coolant materials stack | Nickel-plated copper · stainless steel · POM Acetal · EPDM seals |
| Working pressure | 2 500 Pa |
| Max pressure | 30 000 Pa (with EK-Pro PRM) · 75 000 Pa (otherwise) |
| Operating temperature range | 5 – 35 °C ambient · 5 – 60 °C coolant |
| Power supply | 5 × 2000 W · 80 PLUS Titanium · hot-swap · redundant |
| Power distribution | Platform-level PDU · independent of motherboard VRM |
| Country of origin | Slovenia |

## Compatibility

Motherboards: EATX, EEB, and server reference boards from major vendors. Workstation boards in the same envelope are also supported but not the primary use case.

CPU sockets: AMD SP5 (EPYC 9004 / 9005), Intel LGA 4677 (Xeon Scalable 4th / 5th gen) and Intel LGA 4710/Intel LGA 4710-2 verified. Other server sockets supported via available motherboard choices.

GPUs: Full-height PCIe Gen 5, up to 700 W per slot sustained, water-block-equipped. Verified with the GPU water-block catalogue (RTX 6000 Ada, RTX A6000, H100 NVL, H200 NVL, A100, RTX PRO 6000 Blackwell, others on request).

Networking: OCP 3.0 mezzanine and standard PCIe NICs supported.

## In the box

- Chassis with rack rails and front bezel
- Integrated cooling loop · pre-filled, pre-tested with EK-CryoFuel coolant
- Redundant power supply assembly · 5 × 2000 W · 80 PLUS Titanium · hot-swap
- Platform-level power distribution unit
- GPU and CPU water blocks per chosen configuration (separately specified)
- Mounting hardware kit and integration manual
- Quick-disconnect couplings for serviceability
- Not included: motherboard, CPUs, GPUs, RAM, storage, network adapters; sourced separately per the configuration BOM

## For procurement

- No proprietary components; every internal part has a second source.
- Board and silicon are sourced and refreshed independently of the chassis.
- One platform carries a build from two GPUs to eight 700 W GPUs, so the entry configuration and the end state are the same asset.
- Titanium-rated power conversion and zero operational water use are scoreable criteria in energy-weighted tenders.

A full Procurement and Total Cost of Ownership brief is available on request.

## Disclaimers

- The RM-4U8G is a barebone server platform. The motherboard, CPUs, GPUs, memory, storage, and network adapters are not included and are sourced separately per the configuration bill of materials. Reference configurations are examples drawn from real customer builds and do not constitute a fixed specification.
- Component compatibility is verified by LM TEK engineering against the buyer’s chosen board, CPUs, and GPUs before quote. Compatibility lists are current at time of publication and subject to change. Contact LM TEK engineering to confirm compatibility for your configuration before purchase.
- Cooling capacity is validated per deployment for the selected GPU. The 700 W per-slot sustained envelope and the 8000 W sustained chassis figure are platform maxima; achievable sustained power depends on the chosen components and ambient conditions.
- The integrated loop is pre-filled with EK-CryoFuel. Use only engineered coolant specified by LM TEK. Use of incompatible or untreated coolant may damage the loop and void warranty.
- Operating envelope: 5–35°C ambient, 5–60°C coolant. Working pressure 2500 Pa; maximum pressure 30 000 Pa with EK-Pro PRM, 75 000 Pa otherwise. Operation outside these limits is not supported.
- All figures are measured at supplier level. Third-party independent measurement reports are available on request under NDA.
- Engineered and assembled in Slovenia (Komenda). Cooling system designed by the EK engineering team.
- Dimensional and technical drawings are copyrighted material of LM TEK and are provided for rack-integration planning only.