---
title: "Inside LM TEK 4U8GPU10KW"
description: "The chassis is built around the cooling loop, the PCIe slot layout follows the loop’s physical envelope, and power distribution is platform-level. It is not a board dropped into a case; it is a platform designed as a single engineering problem."
url: "https://www.lmtek.com/ai-hpc-servers/4u8gpu10kw/inside"
locale: "en"
---

# Inside LM TEK 4U8GPU10KW

The chassis is built around the cooling loop, the PCIe slot layout follows the loop’s physical envelope, and power distribution is platform-level. It is not a board dropped into a case; it is a platform designed as a single engineering problem.

## Engineered as one thermal and power system.

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.

## Density

| Specification | Value |
| --- | --- |
| GPU slots | Up to 8 × full-height PCIe Gen 5 |
| Form factor | 4U · 19-inch rack chassis |

Liquid cooling collapses each GPU’s thermal solution into a single-slot water block. That, not a bigger box, is how eight full-height PCIe Gen 5 cards, up to 700 W per slot sustained, fit in a 4U envelope.

The chassis is engineered around standard EATX server motherboards rather than a proprietary board. PCIe slot positions and clearance accommodate the cooling loop without needing custom riser cards. Mounting hardware is rack-standard. The platform takes ordinary enterprise components and arranges them at extraordinary density.

Top view, cover removed. Shown populated; the platform ships as a barebone.

## Upgrade in place

| Specification | Value |
| --- | --- |
| Entry build | Two GPUs |
| End state | Eight accelerators at up to 700 W per slot (platform envelope) |
| CPU | Single-socket to dual-socket is a board choice |

[Supported boards, sockets and GPUs](/ai-hpc-servers/4u8gpu10kw/specifications#datasheet)

The chassis, loop, and power are sized for the full eight-GPU end state on day one. A two-GPU entry build grows inside the same chassis and on the same loop, in the same four rack units.

Adding a supported accelerator also requires its water block and a check against the published validation specifications.

## The loop

| Specification | Value |
| --- | --- |
| Loop type | Closed integrated loop, fully in-chassis |
| Radiator placement | Monolith triple-width pass-through radiator · front intake; airflow paths engineered alongside the loop |
| Radiator core | 130 mm thick core · 120 × 360 mm standard cooling surface · fits 3 × 120 mm standard fans per side |
| Fans | 6 × industrial-grade fans |

The chassis is built around the closed cooling loop, and the PCIe slot layout follows the loop’s physical envelope rather than a board form factor. Pump, manifold, GPU and CPU water blocks, radiator, and reservoir all live inside the chassis, with the radiator on the front intake so airflow paths are engineered alongside the loop.

Working pressure is 2500 Pa; maximum pressure is 30 000 Pa with an EK-Pro PRM and 75 000 Pa otherwise. There is no external CDU or facility plumbing.

A test unit with the cover removed: the whole loop sits inside the chassis.

## The blocks

| Specification | Value |
| --- | --- |
| Coolant materials stack | Nickel-plated copper · stainless steel · POM Acetal · EPDM · brass · EK-CryoFuel |
| GPU and CPU water blocks | Per chosen configuration; separately specified |

[H200 NVL](/components/gpu-water-blocks/h200-nvl)

[RTX PRO 6000 Blackwell](/components/gpu-water-blocks/rtx-pro-6000-blackwell)

[EK-Pro CPU WB SP5 Rack](/components/cpu-water-blocks/ek-pro-cpu-wb-sp5-rack-ni-acetal)

[EK-Pro CPU WB 4677 Rack Ni + Acetal](/components/cpu-water-blocks/ek-pro-cpu-wb-4677-rack-ni-acetal)

[Compatible components](/components/for/4u8gpu10kw)

The loop uses the materials discipline, pressure envelope, and manifold geometry of the EK liquid-cooling reference, transferred to enterprise rack scale. Cooling designed by the EK engineering team, twenty-five years deep in high-performance liquid cooling.

Block availability is part of the published validation specs.

## Platform power

| Specification | Value |
| --- | --- |
| Power envelope | 10 kW installed (5 × 2 kW); 8 kW sustained, N+1 |
| Power supply | 5 × 2000 W CRPS · 80 PLUS Titanium · with redundancy |
| Power distribution | Platform-level PDU · independent of motherboard VRM |
| PSU redundancy | N+1 at maximum spec: four units carry the 8000 W sustained platform maximum, the fifth unit provides redundancy · implemented by the platform PDU, so lower install specs increase the redundancy margin to the capacity available above current draw |

Five 2000 W 80 PLUS Titanium PSUs with redundancy and a platform-level PDU size the GPU power budget at the chassis, not the board. Headroom for the full 8-GPU configuration at up to 8000 W sustained.

The customer’s motherboard chooses CPU and memory; the platform chooses how power reaches the GPUs.

## Control and protection

| Specification | Value |
| --- | --- |
| Coolant temperature target | Server preset universal at 50°C at the hottest point in the loop |
| Coolant temperature | System shutdown at 60°C coolant temperature |
| Pump failure | Hard shutdown on failure of either pump |
| Boot interlock | System boots only when at least one pump reports function |
| Fan failure | Audible alarm on any fan failure |

The server preset holds the coolant at 50°C at the hottest point in the loop, which favours sustained performance. For the controller revision and monitoring integration in a given deployment, talk to LM TEK engineering.

Protective functions are integral to safe operation: the system shuts down at 60°C coolant temperature, performs a hard shutdown if either pump fails, and boots only when at least one pump reports function. Do not bypass or disable these protections.

An audible alarm signals fan failure. Investigate and resolve the alarm before continued operation.

## Service and access

| Specification | Value |
| --- | --- |
| Serviceability | Dry-break front quick-disconnect couplings · built-in coolant level indicator (reservoir window) |
| Storage bay | Single mounted drive bay for 5 × 2.5-inch drives · in chassis airflow path · positioned for straightforward access during service |

[Available integration resources](/ai-hpc-servers/4u8gpu10kw/resources)

Dry-break quick disconnects support block replacement with the server racked. Integrated IN/OUT service ports provide access for an automated drain, flush and fill cycle without opening the loop. Front fill access and the reservoir window provide access and visibility.

## Specified per deployment, within a published validation envelope.

[Modification projects](/custom-engineering?platform=4u8gpu10kw#project-brief)

The platform ships with published validation specs: supported board formats and sockets, the per-slot GPU envelope, block availability, and the operating limits. A configuration inside them is quotable as is. For a board, GPU, or cooling requirement outside them, send LM TEK engineering the intended configuration.