SOLUTION GUIDE

SUPERMICRO RACK SCALE LIQUID COOLING SOLUTIONS

LIQUID COOLING SUPERMICRO DIRECT TO CHIP

ADVANTAGES LIQUID COOLING SOLUTIONS

Liquid cooled rack solution that delivers superior performance

UP TO
and efficiency for large scale AI and cloud scale compute
infrastructure

89 % • Full turn-key single-source solution optimized from

proven total solution blueprints of compute, GPU, storage,
networking, and power & cooling reference designs, with
integrated power management tools

REDUCTION • Support highest densities and highest TDP CPUs & GPUs
in Server Cooling with up to 100kW power and cooling per rack
Component Electricity
Costs • Fully validated and tested at system (L10), rack (L11) and
cluster (L12) levels

UP TO • Accelerated lead times based on in-stock inventory with

deployment in weeks not months

40 %
• Enterprise grade redundant cooling pump and power
supplies, leak-proof connectors and leak detection

REDUCTION
in Electricity Costs for
Entire Data Center DATA CENTER SCALE LIQUID COOLING REDUCES COSTS AND
INCREASES PERFORMANCE
UP TO The latest servers with multiple CPUs and GPUs are creating a significant challenge for
data center operators. High-end servers are now generating up to 10kW of heat, which

55
must be removed from the server. Traditional air cooling through CRAC units, even with
hot and cold aisle separation, is expensive and inefficient.
%
Servers that are application-optimized for AI, HPC, and Analytics require the latest in
CPU and GPU technologies, which run hotter than previous generations. Multiple CPUs
and GPUs per server are needed for performance-intensive computing, driving up the
REDUCTION electricity requirements of the server as well as at the rack level.
in Datacenter
Server Noise
MARCH 2025
 COMPONENTS OF AN EFFICIENT LIQUID COOLING SOLUTION
RACK SCALE
Supermicro’s liquid cooled rack solution consists of several components that are designed
in-house to achieve the highest level of performance and reliability. All the components are
integrated as a rack-level solution providing a true one-stop shop customer experience. The
critical components of Supermicro’s liquid cooled rack solution are:

Coolant Distribution Unit (CDU) –

Contains the pumping system that circulates
the coolant to the cold plates cooling down
the CPUs, GPUs, and DIMMs. The Supermicro
CDU integrates 2 hot-swappable and
redundant pumping modules and power
supply modules, guaranteeing nearly a
100% uptime to the operator. The CDU has
cooling capacity of up to 100kW or 250kW
(depending on CDU configuration) enabling
extremely high rack densities. The CDU
also offers an easy-to-use touch screen to
Supermicro CDU
monitor and control the rack operation, with WebUI
access and integrated in Supermicro’s Super Cloud
Composer® data center management software. The CDU control system optimizes power
consumption while ensuring efficient cooling is delivered to all CPUs, GPUs, and DIMMs. An
effective anti-condensation strategy is adopted to prevent any hardware degradation.

Coolant Distribution Manifold (CDM) – The CDM contains the distribution pipes that supply
coolant to each server and return the hot coolant back to the CDU.
There are two types of CDMs:
Vertical – Vertical manifolds are placed at the front or back of the rack and directly connected
to the CDU with hoses. They deliver coolant to the cold plates on systems with inlet and
outlet pipes.

Horizontal – Horizontal manifolds are placed at the front of the rack and occupy 1U of rack
space each. They connect the vertical manifolds at the rear of the rack to cold plates on
systems with inlet and outlet pipes at the chassis of the rack (SuperBlade and 8-GPU servers)

Quick Disonnect Couplings (QDC) – Flexible QDC are used to bring the cold liquid to
the CPU, GPU, and DIMM cold plates and return the hot liquid to the CDMs. Supermicro
connectors are single handed, zero drip and quick-disconnect allowing operators to service
Horizontal CDM
liquid cooled systems safely and efficiently.

Cold Plates – The cold plates are placed on top of the

CPUs, GPUs, and DIMMs and feature micro-sized channels
through which coolant flows to cool down the chips with
high efficiency. Supermicro cold plates are designed to
reduce hot spots on the chip and achieve ultra-low thermal
resistance.

CPU Cold Plate GPU Cold Plate DIMM Cold Plate

RACK SCALE ADVANCED ENGINEERING

As the Rack becomes the new unit for scalable computing, engineering a liquid cooled rack requires careful planning and expert assembly.
A liquid cooling solution needs to be flexible enough to handle a wide range of servers, while being able to cool an entire rack of high
performance systems.

2 Supermicro Rack Scale Liquid Cooling Solutions

Supermicro Rack Scale Liquid Cooling Solutions

RACK SAMPLE CONFIGURATIONS

1U Flagship Rackmount Purpose-Built HPC-at-Scale

NVIDIA GB200 NVL72
Architecture Solution
Up to 18 compute nodes / 72 NVIDIA Up to 44 servers in a 48U rack Up to 21 systems / 84 server nodes in
Blackwell GPUs in a 52U rack a 48U rack

Product Qty. Product Qty. Product Qty.

1U Liquid-cooled Server with

44
2 NVIDIA GB200 Grace™ 18 1U Hyper Server FlexTwin™ Server (2U4N) 21
Blackwell Superchips

33kW Power Shelves 8 CDU 1 CDU 1

CDU 1 Vertical CDM 1 Vertical CDM 1

Switch 1 Switch 1
Vertical CDM 2

NVLink Switch 9
In-band & Out-of-band
1
Management Switch

Solutions Guide | March 2025 3

 GPU-Optimized
Supermicro GPU-optimized systems provide maximum acceleration for large-scale AI training,
large language models, and generative AI applications. The GPU-optimized systems have been
completely re-engineered to take advantage of the latest interconnect, memory, storage, and
cooling technologies to ensure significant performance increases over prior generations. The
modular design includes a dedicated GPU tray which houses an SXM5 (H100/H200) or SXM6
(B200) 8-GPU baseboard and is easily accessible from the cold aisle, simplifying installation and
servicing. For maximum GPU density, a liquid-cooled 4U architecture can be integrated using
Supermicro’s complete direct-to-chip liquid cooling solution, allowing up to 8 systems in a stan-
dard 48U rack for a total of 64 GPUs. Supermicro builds upon its proven AI-optimized system
architecture with thermal-design improvement to handle the most powerful AI GPUs.

BigTwin®
Supermicro BigTwin systems provide maximum performance and serviceability in a multi-node
architecture, with dual Intel® Xeon® 6700 or 6500 series processors per node and a hot-swappable
tool-less design. Optimized for density (2U4N) or storage (2U2N), BigTwin systems with shared
components can be up to 8% more power efficient than standard 1U servers, with Supermicro’s
Resource Saving Architecture of shared power and cooling reducing TCO and TCE. All BigTwin
systems can be air cooled, with liquid cooling options available to not only further reduce power
consumption and noise levels, but also allow maximum compute density of up to eight 350W TDP
CPUs in a 2U chassis.

FlexTwin™
Supermicro FlexTwin is a new 2U 4-node platform designed for maximum performance density
and serviceability in a multi-node architecture, featuring support for the latest CPU, memory,
storage, and cooling technologies. Purpose-built to support demanding HPC workloads includ-
ing financial services, manufacturing, scientific research, and complex modeling, FlexTwin can be
customized to suit specific HPC applications and customer requirements thanks to Supermicro’s
modular Building Block architecture. Each hot-swappable FlexTwin node features direct-to-chip
liquid cooling technology which can remove up to 90% of server-generated heat, which not only
reduces data center cooling costs compared to traditional air cooling but also ensures maximum
compute performance by reducing instances of thermal throttling under maximum load.

SuperBlade®
Supermicro’s GPU-enabled high-performance, density-optimized, and energy-efficient Super-
Blade can significantly reduce initial capital and operational expenses for many organizations.
SuperBlade utilizes shared, redundant components, including cooling, networking, power and
chassis management, to deliver the compute performance of an entire server rack in a much
smaller physical footprint. For even greater compute densities, dual-processor and single-width
blade configurations can support 350W TDPs with optional direct-to-chip liquid cooling. The use
of liquid cooling in data centers not only allows components to run at higher performance levels,
but also reduces the need for Computer Room Air Conditioning (CRAC) units and improves overall
efficiency, lowering OPEX, TCO and TCE.

Hyper
The Hyper series brings flagship performance to Supermicro’s range of rackmount servers,
built to take on the most demanding workloads along with the storage & I/O flexibility that
provide a custom fit for a wide range of application needs. With up to 8 PCIe 5.0 expansion
slots in a 2U chassis that can accommodate up to 4 double-width GPUs, enterprises have the
flexibility to add GPUs, DPUs, and other accelerators to increase workload performance. With
Supermicro direct-to-chip liquid cooling solution, Hyper can support dual processors in 1U
chassis with CPU TDP up to 500W and 12 hot-swappable NVMe drive bays.

4 Supermicro Rack Scale Liquid Cooling Solutions

Supermicro Rack Scale Liquid Cooling Solutions

X14/H14 Direct-to-Chip Liquid Cooling Systems

Product Family Server Description
Dual Intel® Xeon® 6900 Series Processors with P-cores
SYS-422GA-NBRT-LCC 4U, 24 DIMMs
NVIDIA HGX™ B200 8-GPU

Dual AMD EPYC™ 9005/9004 Series Processors

AS -4126GS-NBR-LCC 4U, 24 DIMMs
NVIDIA HGX™ B200 8-GPU

GPU Dual AMD EPYC™ 9005/9004 Series Processors

SAS -4126GS-NMR-LCC 4U, 24 DIMMs
AMD Instinct™ MI325X Accelerators

NVIDIA 72-core Grace CPU on GH200 Hopper™ Superchip

ARS-111GL-NHR-LCC
Onboard Memory and GPU

NVIDIA 72-core Grace CPU on GH200 Hopper™ Superchip

ARS-111GL-DNHR-LCC
Onboard Memory and GPU

Dual Intel® Xeon® 6900 Series Processors with P-cores

SYS-222FT-HEA-LCC
2U 4-Node, 24 DIMMs
FlexTwin™
Dual AMD EPYC™ 9005/9004 Series Processors
AS -2126FT-HE-LCC
2U 4-Node, 24 DIMMs

Dual Intel® Xeon® 6900 Series Processors with P-cores

Hyper SYS-122HA-TN-LCC
1U, 24 DIMMs

X14 Optional Liquid Cooling Support Systems

Product Family Server Description
Dual Intel® Xeon® 6900 Series Processors with P-cores
4U, 24 DIMMs
SYS-222HA-TN
Support up to 4 double-width GPUs

Dual Intel® Xeon® 6700/6500 Series Processors with P-cores, or 6700

Series Processors with E-cores
Hyper SYS-122H-TN
1U, 32 DIMMs
Support up to 3 single-width GPUs

Dual Intel® Xeon® 6700/6500 Series Processors with P-cores, or 6700

SYS-222H-TN Series Processors with E-cores
2U, 32 DIMMs
Support up to 4 double-width GPUs

Solutions Guide | March 2025 5

 Product Family Server Description
Dual Intel® Xeon® 6700/6500 Series Processors with P-cores, or 6700
SYS-222BT-DNR Series Processors with E-cores
2U 2-Node, 16 DIMMs

Dual Intel® Xeon® 6700/6500 Series Processors with P-cores, or 6700

SYS-622BT-DNC8R Series Processors with E-cores
2U 2-Node, 16 DIMMs

Dual Intel® Xeon® 6700/6500 Series Processors with P-cores, or 6700

SYS-622BT-HNC8R Series Processors with E-cores
2U 4-Node, 16 DIMMs

Dual Intel® Xeon® 6700/6500 Series Processors with P-cores, or 6700

BigTwin® SYS-222BT-HNC8R Series Processors with E-cores
2U 4-Node, 16 DIMMs

Dual Intel® Xeon® 6700/6500 Series Processors with P-cores, or 6700

SYS-222BT-HNC9R Series Processors with E-cores
2U 4-Node, 16 DIMMs

Dual Intel® Xeon® 6700/6500 Series Processors with P-cores, or 6700

SYS-222BT-HNR Series Processors with E-cores
2U 4-Node, 16 DIMMs

Dual Intel® Xeon® 6700/6500 Series Processors with P-cores, or 6700

SYS-222BT-HER Series Processors with E-cores
2U 4-Node, 16 DIMMs

Single Intel® Xeon® 6900 Series Processor with P-cores

SBI-612BA-1NE34
With 6U Enclosure
12 DIMMs

Dual Intel® Xeon® 6700/6500 Series Processors with P-cores or E-cores

SBI-422B-1NE14 With 8U Enclosure
16 DIMMs

Single Intel® Xeon® 6700/6500 Series Processor with P-cores or E-cores

SuperBlade® SBI-612B-1NE34 With 6U Enclosure
16 DIMMs

Dual Intel® Xeon® 6700/6500 Series Processors with P-cores or E-cores

SBI-622B-1NE34 With 6U Enclosure
32 DIMMs

Dual Intel® Xeon® 6700/6500 Series Processors with P-cores or E-cores

SBI-622B-1NE38 With 6U Enclosure
32 DIMMs

6 Supermicro Rack Scale Liquid Cooling Solutions

Supermicro Rack Scale Liquid Cooling Solutions

X13/H13 Direct-to-Chip Liquid Cooling Systems

Product Family Server Description
Dual 5th/4th Intel® Xeon® Scalable Processors
SYS-421GE-NBRT-LCC 4U, 32 DIMMs
NVIDIA HGX™ B200 8-GPU

Dual 5th/4th Intel® Xeon® Scalable Processors

SYS -421GE-TNHR2-LCC 4U, 32 DIMMs
NVIDIA HGX™ H200/H100 8-GPU

Dual AMD EPYC™ 9004/9005 Series Processors

GPU AS -4125GS-TNHR2-LCC 4U, 24 DIMMs
NVIDIA HGX™ H200/H100 8-GPU

Dual AMD EPYC™ 9004/9005 Series Processors

AS -4125GS-TNMR2-LCC 4U, 24 DIMMs
AMD Instinct™ MI300X Accelerators

Dual 5th/4th Intel® Xeon® Scalable Processors

SYS-221GE-TNHT-LCC 2U, 32 DIMMs
NVIDIA HGX™ H100 8-GPU

Quad AMD Instinct™ MI300A Accelerated Processing Units (APUs)

APU AS -2145GH-TNMR-LCC 2U, Onboard 512GB HBM3 stacks memory
4 Onboard GPUs

Dual 5th/4th Intel® Xeon® Scalable Processors

GPU Workstation SYS-751GE-TNRT Rackmount tower, 16 DIMMs
Support up to 4 double-width GPUs

X13 Optional Liquid Cooling Support Systems

Product Family Server Description
Dual 5th/4th Intel® Xeon® Scalable Processors
SYS-421GU-TNXR 4U, 32 DIMMs
NVIDIA HGX™ H200/H100 4-GPU

Dual 5th/4th Intel® Xeon® Scalable Processors

SYS-521GE-TNRT 5U, 32 DIMMs
Support up to 10 double-width GPUs
GPU
Dual 5th/4th Intel® Xeon® Scalable Processors
SYS-421GE-TNRT 4U, 32 DIMMs
Support up to 10 double-width GPUs (Dual-Root PCIe GPU)

Dual 5th/4th Intel® Xeon® Scalable Processors

SYS-421GE-TNRT3 4U, 32 DIMMs
Support up to 8 double-width GPUs (Direct-connect PCIe GPU)

Solutions Guide | March 2025 7

 Product Family Server Description
Dual 5th/4th Intel® Xeon® Scalable Processors
GPU Workstation SYS-751GE-TNRT-NV1 Rackmount tower, 16 DIMMs
Support up to 4 double-width GPUs

Dual 5th/4th Intel® Xeon® Scalable Processors

SYS-121H-TNR
2U, 32 DIMMs
Support up to 3 single-width GPUs

Dual 5th/4th Intel® Xeon® Scalable Processors

SYS-221H-TNR 1U, 32 DIMMs
Support up to 4 double-width GPUs
Hyper
Dual 5th/4th Intel® Xeon® Scalable Processors
SYS-221H-TN24R 2U, 32 DIMMs
Support up to 2 double-width GPUs

Dual 5th/4th Intel® Xeon® Scalable Processors

SYS-621H-TN12R 2U, 32 DIMMs
Support up to 4 double-width GPUs

Dual 5th/4th Intel® Xeon® Scalable Processors

SYS-221BT-DNTR
2U 2-Node, 16 DIMMs

Dual 5th/4th Intel® Xeon® Scalable Processors

SYS-621BT-DNRT
2U 2-Node, 16 DIMMs

Dual 5th/4th Intel® Xeon® Scalable Processors

SYS-221BT-DNC8R
2U 2-Node, 16 DIMMs

Dual 5th/4th Intel® Xeon® Scalable Processors

SYS-621BT-DNC8R
2U 2-Node, 16 DIMMs

Dual 5th/4th Intel® Xeon® Scalable Processors

SYS-221BT-HNR
BigTwin® 2U 4-Node, 16 DIMMs

Dual 5th/4th Intel® Xeon® Scalable Processors

SYS-221BT-HNTR
2U 4-Node, 16 DIMMs

Dual 5th/4th Intel® Xeon® Scalable Processors

SYS-221BT-HNC8R
2U 4-Node, 16 DIMMs

Dual 5th/4th Intel® Xeon® Scalable Processors

SYS-221BT-HNC9R
2U 4-Node, 16 DIMMs

Dual 5th/4th Intel® Xeon® Scalable Processors

SYS-621BT-HNC8R
2U 4-Node, 16 DIMMs

Dual 5th/4th Intel® Xeon® Scalable Processors

SYS-621BT-HNTR
2U 4-Node, 16 DIMMs

8 Supermicro Rack Scale Liquid Cooling Solutions

Supermicro Rack Scale Liquid Cooling Solutions

Product Family Server Description

Single 5th/4th Intel® Xeon® Scalable Processors
SYS-521E3-RTB
4U 4-Node, 16 DIMMs
FatTwin®
Single 5th/4th Intel® Xeon® Scalable Processors
SYS-F511E2-RT
4U 8-Node, 16 DIMMs

Single 5th/4th Intel® Xeon® Scalable Processors

SBI-411E-1G With 8U Enclosure
8 DIMMs

Dual 5th/4th Intel® Xeon® Scalable Processors

SuperBlade® SBI-621E-1NE34 With 6U Enclosure
32 DIMMs

Dual 5th/4th Intel® Xeon® Scalable Processors

SBI-621E-1NE38 With 6U Enclosure
32 DIMMs

Solutions Guide | March 2025 9

LLC Systems

HGX™ GPU Instinct™ GPU FlexTwin™

X14/H14/X13/H13 H14/H13 X14/H14

Hyper Instinct™ APU GPU Workstation

X14 H13 X13

Optional LCC Systems

HGX GPU PCIe GPU SuperBlade

X13 SYS-421GU-TNXR X14/X13

Hyper BigTwin FatTwin

X14/X13 X14/13 X13

10 Supermicro Rack Scale Liquid Cooling Solutions

 Accelerate
Supermicro Rack Scale Liquid Cooling Solutions
Everything

Supermicro Liquid Cooling Tower

Solution Summary
Liquid Cooling Tower data centers consume is
Worldwide, the amount of energy that
n of this is to cool the servers
increasing yearly. A significant portio dedic ated to AI
The Supermicro Liquid Cooling Tower solution offers a storag
versatile ands. With
e system energy efficient
the increa sed use of serversof
method removing
and
air cooling costs are skyrocketing, requiring liquid
the heat produced by today’s latest servers. The modular
training,design is available
traditional in different sizes with multiple
be the only viable
g to be
coolinand considered. In fact, liquid cooling may
cell configurations up to a maximum of 10MW per unit scalable up to multiple cellt trajec
curren configurations
tory of increasing
technology moving forward with the air systems to
from 5MW to hundreds of MW. Each cell can be usedTDP separately,
at the serverenabling cooling
and rack level. tower
The ability redundancy.
of forced
heat gener ated in the entire data center requires significant
remove the
conditioning infrastructure,
investments in computer room air
cantly increa sing the TCO of a data center. Liquid cooling is
signifi
cantly more efficie nt at the data center level and requires an
signifi
The Supermicro green computing
entire solution to cool the systems.
of-the-art integrated liquid cooling
total solution now includes a state-
re.
tower as part of the overall infrastructu
soluti on is versatile and energy
The Supermicro Liquid Cooling Tower
ced by today’s latest servers. The
efficient at removing the heat produ
remove the heat from today’s
system is optimized to transport and
is pioneering a way to obtain a
most powerful AI servers. Supermicro
s to the cooling infrastructure.
complete AI solution, from the server

Supermicro SuperCloud Composer® for Liquid Cooled Data Centers

for the modern data center with the
following capabilities:
r is designed
The Supermicro Liquid Cooling Towe
SuperCloud Composer’s LCCM (Liquid Cooling Consult•Module) Closed Loopis aDesig
powerful
n tool to collect vital information on physical assets and sensor dataperfrom
ns up to reach a maximum of 10MW
unit
in differe nt sizes with multiple cell configuratio
a CDU (Cooling Distribution Unit), including pressure, humidity, pump
• Modular Desig and ble
n – availa valve status, and more. CDUdancy data is presented in real-time, enabling users
used separ ately, enabl ing cooling tower redun
to monitor operating efficiency of their liquid cooled racks.
• EachUsing
cell canthese insights, SuperCloud Composer (SCC) helps the user to set up alerts, manage
for liquid coolin g workl oads
firmware updates, and more. • Optimized cooling tower.
rature water can be returned to the
r tempe
• High Temperature Fills (>60C) – Highe
r efficiency
• High Temperature range enabling highe
mptio n – 6.2kW /MW of cooling
• Low power consu
of cooling
• Low Water consumption – 6.2GPM/MW prevention
coated Stainless Steel for corrosion
• Long Lifetime: 15-20years with epoxy
s, not mont hs)
• Quick deployments (week

Tower Solution Work?

How Does the Supermicro Liquid Cooling
pumps through each
1. Warm facility water is circulated by
ing the heat from
Supermicro Liquid Cooled Racks, remov
s. The heat is
the coolant loop and cooling the server
warm water through
transferred from the coolant to the
, the water exits
the CDU heat exchanger, and, as a result
at a higher temperature.
water down to a
2. The cooling tower cools the warm
be returned to the
temperature at which the water can
racks, and the cycle is then repeated.

September 2023
All rights reserved
Copyright Super Micro Computer, Inc.
Supermicro’s Rack Integration Services,
© 2023
Turnkey Cluster Level Liquid Solutions
Supermicro’s Rack Integration Services leverage application-
optimized motherboards, chassis, cooling subsystems, networking
components, cluster management tools, energy-efficient power
supply technologies, and compact enclosures to design and develop
customized and enterprise solutions. Supermicro understands the
requirements of today’s fast-paced business environments and
customer requirements. We offer an end-to-end integration service
that helps customers reduce overhead, maximize efficiency and
quality, and gain a competitive advantage through reduced time-to-
results.

Supermicro works with leading organizations in all geographies

to design, install, and test various liquid cooling solutions. The
Supermicro process involves a rigorous set of phases that ensure the
most optimized and tested solutions for environments where liquid
cooling is required for maximum performance.

Solutions Guide | March 2025 11

Supermicro Rack Scale Liquid Cooling Solutions

ABOUT SUPER MICRO COMPUTER, INC.

Supermicro® (NASDAQ: SMCI), the leading innovator in high-performance, high-efficiency server

technology is a premier provider of advanced server Building Block Solutions® for Data Center, Cloud
Computing, Enterprise IT, Hadoop/Big Data, HPC and Embedded Systems worldwide. Supermicro is
committed to protecting the environment through its “We Keep IT Green®” initiative and provides
customers with the most energy-efficient, environmentally-friendly solutions available on the
market.

[Link]/liquidcooling

No part of this document covered by copyright may be reproduced in any form or by any means
— graphic, electronic, or mechanical, including photocopying, recording, taping, or storage in an
electronic retrieval system — without prior written permission of the copyright owner.

Supermicro, the Supermicro logo, Building Block Solutions, We Keep IT Green, SuperServer, Twin,
BigTwin, TwinPro, TwinPro², SuperDoctor are trademarks and/or registered trademarks of Super Micro
Computer, Inc.

All other brands names and trademarks are the property of their respective owners.

© Copyright 2025 Super Micro Computer, Inc. All rights reserved.

Printed in USA Please Recycle 02_LiquidCooling-205_R16