--- title: "Liquid Cooling for High-Density GPU Systems: Follow the Interfaces | Aguasto" description: "Understand facility and technology cooling circuits, residual air paths and the evidence needed to evaluate a proposed GPU rack integration." canonical: "https://www.aguasto.com/learn/liquid-cooling-gpu-rack-interfaces" last-updated: "2026-10-13" --- # Liquid Cooling for High-Density GPU Systems: Follow the Interfaces | Aguasto Canonical HTML: [Liquid Cooling for High-Density GPU Systems: Follow the Interfaces](https://www.aguasto.com/learn/liquid-cooling-gpu-rack-interfaces) > Understand facility and technology cooling circuits, residual air paths and the evidence needed to evaluate a proposed GPU rack integration. [Aguasto](https://www.aguasto.com/) [learn](https://www.aguasto.com/learn) Liquid Cooling for High-Density GPU Systems: Follow the Interfaces [Resources](https://www.aguasto.com/resources) [GPU Library](https://www.aguasto.com/gpus) [Model Index](https://www.aguasto.com/models) [Inference Stack](https://www.aguasto.com/inference-stack) [Learn](https://www.aguasto.com/learn) [Insights](https://www.aguasto.com/insights) [Research](https://www.aguasto.com/research) [Engineering tools](https://www.aguasto.com/tools) Aguasto / learn ## Liquid Cooling for High-Density GPU Systems: Follow the Interfaces Understand facility and technology cooling circuits, residual air paths and the evidence needed to evaluate a proposed GPU rack integration. Original Aguasto conceptual illustration. Not manufacturer geometry, an operating Aguasto asset, or a measured benchmark. By Aguasto Research · Published October 13, 2026 · Revision 1 Aguasto Research documents third-party hardware, models and software. References do not represent Aguasto inventory, evaluated deployments or available inference services. Illustration: Liquid Cooling for High-Density GPU Systems: Follow the Interfaces: original technical study. Facility water and technology coolant stay in separate circuits. A CDU heat exchanger links their thermal paths, while residual equipment heat follows a separate air path. Facility water and technology coolant stay in separate circuits. A CDU heat exchanger links their thermal paths, while residual equipment heat follows a separate air path. Original Aguasto illustration. Read the guide for configuration and measurement limitations. ### Trace heat across distinct circuits OCP’s integration guidance distinguishes the facility water system from the technology cooling system, with a coolant distribution unit providing an interface between them. A heat exchanger can transfer heat while keeping the fluids separated. The circuit labels matter: a statement about facility water cannot automatically be applied to the coolant reaching equipment. Follow the exact proposed system rather than assuming every installation uses the same arrangement. Our diagram shows these as two different paths connected by a thermal interface. Its purpose is to make questions visible: which supplier owns each circuit, which specification applies at the boundary and where acceptance evidence is recorded? It is a generic explanatory drawing. It does not specify fluid chemistry, flow, pressure, temperatures or an Aguasto facility design. Those values need the actual equipment documentation and project engineering. [Liquid Cooling Integration and Logistics White Paper, revision1.0 — Open Compute Project (external source)](https://www.opencompute.org/documents/ocp-liquid-cooling-integration-and-logistics-white-paper-revision-1-0-1-pdf) ### Preserve the residual air path NVIDIA’s GB200 SuperPOD architecture provides an example in which liquid and air cooling remain part of the overall deployment. It is a specific system reference, not a universal split for GPU racks. Calling a system liquid-cooled should therefore lead to a boundary question: which components and heat loads use liquid, and which still rely on the room’s air path? When reviewing a proposed module, maintain separate rows for liquid-side requirements, residual air requirements and the external heat-rejection arrangement. Do not assume the enclosure becomes thermally independent because the GPUs use cold plates. Ask how each supplier documents the intended equipment combination and how future configuration changes will be assessed. Our analysis treats cooling as a facility-to-rack integration problem rather than a single marketing characteristic. [DGX SuperPOD Architecture for DGXGB200 — NVIDIA (external source)](https://docs.nvidia.com/dgx-superpod/reference-architecture-scalable-infrastructure-gb200/latest/dgx-superpod-architecture.html) ### A coolant interface is a multidimensional agreement OCP’s cooling project and logistics guidance place integration and maintenance considerations alongside the cooling technology itself. A named connector or distribution unit does not fully define compatibility. The proposed equipment and facility documentation must establish the relevant mechanical, fluid and operating boundaries. This guide deliberately leaves numerical settings unspecified because they are configuration-dependent. Our procurement worksheet records specification owner, approved equipment combination, applicable interface document, test evidence and unresolved exception for each connection. Ask suppliers to explain what changes when a rack, coolant or distribution unit is replaced. Keep the change record attached to the accepted design. This turns a broad compatibility claim into an auditable question without pretending that a public generic diagram can establish engineering approval. [Liquid Cooling Integration and Logistics White Paper, revision1.0 — Open Compute Project (external source)](https://www.opencompute.org/documents/ocp-liquid-cooling-integration-and-logistics-white-paper-revision-1-0-1-pdf) · [Cooling Environments — Open Compute Project (external source)](https://www.opencompute.org/projects/cooling-environments/) ### Evaluate the limiting interface Cooling equipment names alone do not identify a facility’s accepted operating capacity. The system reference must connect the rack arrangement to its facility services. The NVIDIA architecture and OCP guidance are useful starting points for identifying those dependencies, but neither provides a certificate for a proposed Aguasto installation. A documented design scope and a commissioned operating scope are different pieces of evidence. As a planning method, review the path from equipment to external heat rejection one boundary at a time. Ask which assumptions govern the proposed load, which conditions remain unverified and where the acceptance record closes those uncertainties. Avoid converting a selected component’s rating into a complete facility capacity claim. If two documents use different measurement conditions, preserve both conditions until qualified engineering reconciles them rather than choosing the larger number. [Liquid Cooling Integration and Logistics White Paper, revision1.0 — Open Compute Project (external source)](https://www.opencompute.org/documents/ocp-liquid-cooling-integration-and-logistics-white-paper-revision-1-0-1-pdf) · [DGX SuperPOD Architecture for DGXGB200 — NVIDIA (external source)](https://docs.nvidia.com/dgx-superpod/reference-architecture-scalable-infrastructure-gb200/latest/dgx-superpod-architecture.html) ### Plan the operating handover The integration guidance covers logistics and site connection as well as factory work. That wider scope is relevant to maintainability: equipment is not evaluated only at the moment it arrives. The operating team needs the accepted configuration and supplier documentation to understand the boundaries it has inherited. Actual maintenance and abnormal-condition procedures must come from qualified project teams. Our handover checklist asks who owns alarms and escalation, where the configuration record is stored, which exceptions are unresolved and how a replacement is authorized. These are governance questions rather than instructions for performing work on a cooling system. A useful review will also state what evidence remains absent. The resulting service readiness discussion should connect facility acceptance with the intended customer workload instead of claiming reliability from liquid cooling alone. [Liquid Cooling Integration and Logistics White Paper, revision1.0 — Open Compute Project (external source)](https://www.opencompute.org/documents/ocp-liquid-cooling-integration-and-logistics-white-paper-revision-1-0-1-pdf) ### Turn cooling claims into evidence requests For an enterprise buyer, request the exact equipment boundary, the supporting design documents, the accepted operating conditions and the configuration covered by testing. For an OEM discussion, request responsibility at rack and facility interfaces. For infrastructure finance, keep engineering readiness, operating handover and customer demand as separate evidence categories. These questions help each audience assess a different decision while using the same underlying source record. Read the GPU Library with the same discipline: individual accelerators and rack systems have different power and cooling boundaries, and an unknown published value should stay unknown. Follow the commissioning guide for how interface evidence becomes an accepted installation. This is Aguasto’s original evaluation framework informed by primary documents. It contains no measured cooling result, installed equipment claim or universal capacity formula. ### Scope & limitations - Generic circuits illustrate heat-transfer boundaries, not a certified hydraulic design. - No universal liquid/air heat split or rack capacity is assumed. - Actual settings, materials, commissioning and maintenance require equipment-specific qualified engineering. ### Continue the research [How Modular AI Data Centers Are Built and Commissioned](https://www.aguasto.com/learn/modular-data-center-factory-to-commissioning) [What Is a Modular AI Data Center?](https://www.aguasto.com/learn/what-is-a-modular-ai-data-center) [NVIDIA GB200 NVL72](https://www.aguasto.com/gpus/gb200-nvl72) [NVIDIA GB300 NVL72](https://www.aguasto.com/gpus/gb300-nvl72) [The Aguasto modular thesis](https://www.aguasto.com/why-modular) [Potential OEM and strategic collaboration](https://www.aguasto.com/partners) ### Sources & verification The register preserves each source's publication and verification date. Manufacturer claims and documentation are attributed; Aguasto has not run the cited products in production. - [Liquid Cooling Integration and Logistics White Paper, revision1.0 — Open Compute Project ↗](https://www.opencompute.org/documents/ocp-liquid-cooling-integration-and-logistics-white-paper-revision-1-0-1-pdf) FWS/TCS/CDU terminology and liquid-to-liquid isolation; Manufacturing, transport, inspection and commissioning handoffs; Wetted materials and project-specific tests. Publication: date not stated in the reviewed source. Verified: 2026-10-13. - [DGX SuperPOD Architecture for DGXGB200 — NVIDIA ↗](https://docs.nvidia.com/dgx-superpod/reference-architecture-scalable-infrastructure-gb200/latest/dgx-superpod-architecture.html) DGXGB200 uses both direct-liquid and air cooling; Reference architecture ties systems, networks, management and storage. Last modified: 2025-11-19. Verified: 2026-10-13. - [Cooling Environments — Open Compute Project ↗](https://www.opencompute.org/projects/cooling-environments/) Cold-plate/TCS scope includes tubing/manifolds/couplings/CDUs; Different cooling architectures and standardization work. Publication: date not stated in the reviewed source. Verified: 2026-10-13. ### In this publication [Trace heat across distinct circuits](https://www.aguasto.com/learn/liquid-cooling-gpu-rack-interfaces#section-1) [Preserve the residual air path](https://www.aguasto.com/learn/liquid-cooling-gpu-rack-interfaces#section-2) [A coolant interface is a multidimensional agreement](https://www.aguasto.com/learn/liquid-cooling-gpu-rack-interfaces#section-3) [Evaluate the limiting interface](https://www.aguasto.com/learn/liquid-cooling-gpu-rack-interfaces#section-4) [Plan the operating handover](https://www.aguasto.com/learn/liquid-cooling-gpu-rack-interfaces#section-5) [Turn cooling claims into evidence requests](https://www.aguasto.com/learn/liquid-cooling-gpu-rack-interfaces#section-6) [More learn ↗](https://www.aguasto.com/learn) [Source register ↓](https://www.aguasto.com/learn/liquid-cooling-gpu-rack-interfaces#sources) ## Company navigation - [Home](https://www.aguasto.com/) - [Infrastructure](https://www.aguasto.com/infrastructure) - [GPU Cloud](https://www.aguasto.com/gpu-cloud) - [AI Inference](https://www.aguasto.com/ai-inference) - [Development & Investment](https://www.aguasto.com/development) - [Company](https://www.aguasto.com/company) - [Contact](https://www.aguasto.com/contact) - [Privacy](https://www.aguasto.com/privacy) - [Resources](https://www.aguasto.com/resources) - [GPU Library](https://www.aguasto.com/gpus) - [Model Index](https://www.aguasto.com/models) - [Inference Stack](https://www.aguasto.com/inference-stack) - [Learn](https://www.aguasto.com/learn) - [Insights](https://www.aguasto.com/insights) - [Research](https://www.aguasto.com/research) - [Industry cases](https://www.aguasto.com/research/cases) - [Why Modular](https://www.aguasto.com/why-modular) - [Partners](https://www.aguasto.com/partners) - [GPU comparison](https://www.aguasto.com/gpus/compare) - [Model comparison](https://www.aguasto.com/models/compare) - [Engineering tools](https://www.aguasto.com/tools) - [Development friction](https://www.aguasto.com/research/hyperscale-development-friction) - [NVIDIA H100 SXM](https://www.aguasto.com/gpus/h100) - [NVIDIA H200 SXM](https://www.aguasto.com/gpus/h200) - [NVIDIA B200 SXM](https://www.aguasto.com/gpus/b200) - [NVIDIA B300](https://www.aguasto.com/gpus/b300) - [NVIDIA GB200 NVL72](https://www.aguasto.com/gpus/gb200-nvl72) - [NVIDIA GB300 NVL72](https://www.aguasto.com/gpus/gb300-nvl72) - [NVIDIA Vera Rubin NVL72](https://www.aguasto.com/gpus/vera-rubin-nvl72) - [AMD Instinct MI300X](https://www.aguasto.com/gpus/mi300x) - [AMD Instinct MI325X](https://www.aguasto.com/gpus/mi325x) - [AMD Instinct MI355X](https://www.aguasto.com/gpus/mi355x) - [AMD Instinct MI455X](https://www.aguasto.com/gpus/mi455x) - [AMD Helios](https://www.aguasto.com/gpus/helios) - [GLM-4.5](https://www.aguasto.com/models/glm-4-5) - [Qwen3-235B-A22B](https://www.aguasto.com/models/qwen3-235b-a22b) - [DeepSeek-R1](https://www.aguasto.com/models/deepseek-r1) - [Llama 3.3 70B Instruct](https://www.aguasto.com/models/llama-3-3-70b-instruct) - [Gemma 3 27B IT](https://www.aguasto.com/models/gemma-3-27b-it) - [gpt-oss-120b](https://www.aguasto.com/models/gpt-oss-120b) - [Kimi-K2-Instruct](https://www.aguasto.com/models/kimi-k2-instruct) - [GLM-5.3](https://www.aguasto.com/models/glm-5-3) - [Qwen3.8-27B](https://www.aguasto.com/models/qwen3-8-27b) - [DeepSeek-V4.1-Flash](https://www.aguasto.com/models/deepseek-v4-1-flash) - [Llama 4 Scout 17B-16E Instruct](https://www.aguasto.com/models/llama-4-scout-17b-16e-instruct) - [Gemma 4 31B IT](https://www.aguasto.com/models/gemma-4-31b-it) - [Kimi K3](https://www.aguasto.com/models/kimi-k3) - [vLLM](https://www.aguasto.com/inference-stack/vllm) - [SGLang](https://www.aguasto.com/inference-stack/sglang) - [TensorRT-LLM](https://www.aguasto.com/inference-stack/tensorrt-llm) - [NVIDIA Dynamo](https://www.aguasto.com/inference-stack/dynamo) - [NVIDIA Triton Inference Server](https://www.aguasto.com/inference-stack/triton) - [AMD ROCm](https://www.aguasto.com/inference-stack/rocm) - [Ray Serve](https://www.aguasto.com/inference-stack/ray-serve) - [Kubernetes GPU orchestration — NVIDIA device plugin](https://www.aguasto.com/inference-stack/kubernetes) - [Model weights are only the beginning of the memory budget](https://www.aguasto.com/learn/model-weights-and-runtime-memory) - [Why tokens per second is not enough to benchmark inference](https://www.aguasto.com/research/inference-benchmark-methodology) - [Prefill, decode and the cost of separating them](https://www.aguasto.com/learn/prefill-decode-and-disaggregated-serving) - [Choosing between vLLM, SGLang and TensorRT-LLM](https://www.aguasto.com/learn/choosing-an-inference-serving-framework) - [From Power to Inference—Building a Distributed AI Compute Network](https://www.aguasto.com/learn/from-power-to-inference) - [Dedicated compute and shared inference: compare the whole workload](https://www.aguasto.com/insights/dedicated-and-shared-inference-economics) - [Distributed inference reliability begins with the failure boundary](https://www.aguasto.com/learn/distributed-inference-reliability) - [Core Scientific: reading contracted, energized and billed capacity](https://www.aguasto.com/research/cases/core-scientific-capacity-disclosures) - [Why Modular AI Infrastructure Matters Now](https://www.aguasto.com/research/why-modular-ai-infrastructure-matters-now) - [What Is a Modular AI Data Center?](https://www.aguasto.com/learn/what-is-a-modular-ai-data-center) - [How Modular AI Data Centers Are Built and Commissioned](https://www.aguasto.com/learn/modular-data-center-factory-to-commissioning) - [Liquid Cooling for High-Density GPU Systems: Follow the Interfaces](https://www.aguasto.com/learn/liquid-cooling-gpu-rack-interfaces) - [Why Electricity Availability Shapes AI Infrastructure Deployment](https://www.aguasto.com/research/electricity-availability-and-time-to-power) - [What Is KV Cache? Capacity, Reuse and Isolation](https://www.aguasto.com/learn/kv-cache-capacity-and-policy) - [Continuous Batching: Throughput Without Hiding Latency](https://www.aguasto.com/learn/continuous-batching-and-latency) - [Applied Digital: Reading the Polaris Forge 1 Delivery Milestones](https://www.aguasto.com/research/cases/applied-digital-polaris-forge-capacity-milestones) - [Crusoe Abilene: Separate Delivered Infrastructure from Expansion Plans](https://www.aguasto.com/research/cases/crusoe-abilene-delivery-and-expansion) - [Official LinkedIn](https://www.linkedin.com/company/aguasto) This is a generated representation of the public HTML page. Diagrams are conceptual. GPU and inference services are planned and are not currently available to book or call. The contact page describes the available inquiry flow.