DATASHEET

H13 8U GPU Systems

Next-Generation Machine Learning Platform

Propel ML Workloads with 4th Gen AMD EPYC™

Processors and NVIDIA® HGX H100 8-GPU Modules
Optimized for machine learning through massive I/O capacity:
• End-to-end nonblocking I/O to keep GPUs hydrated with data
• 2-socket design supporting 4th Gen AMD EPYC™ Processors
• Up to 24 DIMMs for up to 6 TB of DDR5-4800 memory
• Flexible PCIe 5.0 options for I/O and networking
• Dual-zone cooling optimized for performance and lower operating
costs
• Titanium-Level efficiency power supplies
A+ Server 8125GS-TNHR

Imagine your data lake cascading directly into GPU memory at CPU cycles necessary. Once data is loaded into one accelerator’s
an aggregate rate of 3.2 terabits per second, hydrating the new memory, it can share with its peers through the NVIDIA
engine of AI factories, the NVIDIA HGX H100 8-GPU module. If NVSwitch™ that interconnects all eight GPUs on the 8-GPU
your challenges include feeding machine-learning models with module. This helps to speed complex models that pipeline data
massive amounts of data, and processing that data with the through multiple GPUs. With 900 GBps of bandwidth between
most advanced accelerators available today, the Supermicro any two GPUs, your data moves without bottlenecks.
AS -8125GS-TNHR server is designed to meet your challenges.
GPU-Optimized Server
End-to-End Nonblocking I/O The entire server architecture is built to support the power,
The NVIDIA HGX H100 8-GPU module can transform training cooling, and I/O demands of machine learning training,
exercises that once took weeks into ones that take only days. inferencing, and other data-intensive analytics operations.
Keeping utilization high—and total cost of ownership low— Powered by two AMD EPYC 9004 Series processors and enabled
depends on satisfying the eight GPUs’ voracious appetite for data. with up to 6 TB of DDR5-4800 memory, the interconnections on
this server are based on PCIe 5.0 technology, doubling the
Whether data is supplied from main memory, or from a network- bandwidth of prior generations. Eight slots support tightly-
based data lake, the AS -8125GS-TNHR is designed to move coupled NICs while up to 4 additional slots can be used for
massive amounts from its source to GPU memory with low- general I/O purposes. Similarly, 8 NVMe drives are tightly
latency, nonblocking PCIe 5.0 connectivity. Each of the system’s coupled to individual accelerators, while up to 8 optional drives
GPUs is closely coupled with a PCIe 5.0 switch that provides 16 can provide additional storage.
lanes of connectivity to one of the system’s AMD EPYC 9004
Series processors, 4 lanes to an optional NVMe drive, and 16 Titanium-Level power supplies with up to 96% efficiency keep
lanes to a dedicated PCIe slot that can support up to 400 Gpbs of the GPUs accelerating your machine-learning software while
network connectivity. Remote DMA (RDMA) can set up high- dual-zone cooling with 10 counter-rotating fans keep the
speed transfers from network-based storage through the accelerators within their thermal envelopes.
network interface card, and direct to GPU memory with no
intervening

© 2022 Copyright Super Micro Computer, Inc. All rights reserved November 2022
H13 8U GPU Systems DATASHEET

Made Possible by 4th Gen AMD EPYC Processors

This server is made possible by 4th
Gen AMD EPYC processors, with up to
96 cores per CPU and 192 cores per
server. You can choose the number
of cores , cache size, and clock
frequency appropriate for your application and the rest of the
H13 Generation AS -8125GS-TNHR
features are included at no cost.
Form Factor • 8U rackmount

The AMD EPYC 9004 Series supports massive I/O capacity, • Dual SP5 sockets for AMD EPYC™ 9004 Series processors (two CPUs required)1
Processor Support
• Up to 96 cores and up to 400W cTDP† per processor (up to 192 cores per server)
with 160 lanes of PCIe 5.0 connectivity in this 2-socket server. The
Memory Slots & • 12-channel DDR5 memory support
system-on-chip (SoC) design supports built-in functions Capacity • 24 DIMM slots for up to 6 TB ECC DDR5-4800 RDIMM
including IPMI-based management, on-board M.2 drive, and
• System on Chip
built-in SATA controllers for two drives. The SoC-oriented design • Hardware Root of Trust
On-Board Devices
reduces the number of external chip sets, helping to reduce • IMPI 2.0 with virtual-media-over-LAN and KVM-over-LAN support
• ASPEED AST2600 BMC graphics
complexity and power consumption.
GPU Support3 • NVIDIA HGX H100 8 GPU SXM5 board

Open Management • 8 PCIe 5.0 x16 low-profile slots connected to GPU via PCIe switch
Expansion Slots • 2 PCIe 5.0 x16 full-height half-length slots
Regardless of your data center’s management approach, our • Optional 2 PCIe 5.0 x16 full-height half-length slots via expansion kit
open management APIs and tools are ready to support you. • Up to 8 PCIe 5.0 x4 U.2 NVMe drives connected to GPU via PCIe switch
• 4 PCIe 5.0 x4 NVMe U.2 drives
In addition to a dedicated IPMI port, and a Web IPMI interface, Storage • Optional 4 PCIe 5.0 x4 NVMe U.2 drives2
Supermicro® SuperCloud Composer software helps you • 1 M.2 NVMe/SATA boot drive
configure, maintain, and monitor all of your systems using single- • 2 hot-swap 2.5" SATA drives2
pane-of-glass management. If your DevOps teams prefer to use • 1 RJ45 Dedicated IPMI LAN port
I/O Ports • 2 USB 3.0 Ports (rear)
their own tools, industry-standard Redfish® APIs provide access • 1 VGA Connector
to higher-level tools and scripting languages.
BIOS • AMI Code Base 256 Mb (32 MB) SPI EEPROM
• Built-in server management tool (IPMI 2.0, KVM/media over LAN) with dedicated
LAN port
System Management • Redfish APIs
• Supermicro SuperCloud Composer
• Supermicro Server Manager (SSM) and Supermicro Update Manager (SUM)
• Dual-zone cooling optimized for performance and operational costs with 5 front
System Cooling
and 5 rear counter-rotating fans with optimal speed control
Power Supplies • 6x or 8x 3000W N+N redundant Titanium-Level power supplies
1. Certain CPUs with high TDP (320W and higher) air-cooled support is limited to specific conditions. Please contact
Supermicro Technical Support for additional information about specialized system optimization
2. Optional parts are required for NVMe/SAS/SATA configurations
3. GPU support is limited to specific conditions

© 2022 Super Micro Computer, Inc. Specifications subject to change without notice. All other brands and names are the property of their
respective owners. All logos, brand names, campaign statements and product images contained herein are copyrighted and may not be
reprinted and/or reproduced, in whole or in part, without express written permission by Supermicro Corporate Marketing.