Supermicro Hyper A+ 2025HS-TNR Server Review

We have the 2U, 12-bay Supermicro Hyper A+ 2025HS-TNR Rack Server, and with that A+ in the name, you know it was the teacher’s pet. Actually, that means it has an AMD EPYC processor, two of them to be exact. This one is compatible with both the 9004 and 9005 series processors, but it does require a software update to support the 9005 series CPUs.

As part of Supermicro’s Hyper systems, the Supermicro Hyper A+ 2025HS-TNR is also one of Supermicro’s flagship servers boasting energy efficiency, performance, and flexibility for the data center. This system is designed for Cloud Computing, Virtualization, AI inference and machine learning, and you know what that means, but before we go there, Software-defined storage, and use as an Enterprise server. That last is a pretty broad category. And yes, this system can be configured with a few GPUs to support the all-important AI inference and machine learning capabilities.

On the front of the system there are 12x 3.5-inch hot-swap storage bays for NVMe, SAS, and SATA storage options. You can have 12x SAS, 12x SATA or 12x NVMe drives or a combination depending on your needs. With all of those options, you will need additional cables, and definitely a controller if you go with SAS.

To the left, a small control panel with a power ON button at the top, combo Unit ID/Reset button, and then a few LEDs for Power, HDDs, NIC1, NIC 2, Power Fail, and information, which is kind of a general operational status indicator.

On the back of the Supermicro 2025HS-TNR are dual redundant power supplies, in this case, 1600W each.

Then a dedicated management port outlined in green, 2x USB 3.2 Gen 1 ports, a VGA port, and an Advanced I/O Module that is OCP 3.0 compatible to support a number of optional network connections and ports. If you don’t want to use any of your PCIe slots, that AIOM connector is a good thing. PCIe slots above, and there are several options for those.

For management of the system, Supermicro has a few software suites up its proverbial sleeve. Super Doctor 5, or SD5, can be used in the command line interface or through a nice graphical web-based interface for both Windows and Linux operating systems. SD5 has two versions, one for local management and one for remote management of the system. For remote management you will need the Out-of-Band (OOB) Management Package. It can also be used with Nagios, which is an opensource program that provides more tools for monitoring and managing the health of the network. With SD5 Management Server you get even more tools and can remotely power on the system. The Intelligent Platform management interface (IPMI) can also provide remote access, monitoring, and management. Remote management is accessed through the integrated ASPEED AST 2600 module on the system board via the dedicated RJ45 management port on the back of the system. There are a few more features like SuperCloud Composer and the Supermicro Update Manager.

Inside the system starting at the back, you can see the PCIe slots and the x16 connector for the Advanced IO Module. All PCIe slots are PCIe 5.0. The system can be configured with up to 8x PCIe 5.0 x8 slots, 4x PCIe 5.0 x16 slots or you can mix and match to get a combination that will compliment your workload. With support for the AMD EPYC 9004 and 9005 series CPUs, this system is also CXL 2.0 compatible.

Now more about those PCIe slots. Supermicro has definitely provided options here like the basic 4x PCIe x16 slots or 8x PCIe 5.0 x8 slots and then combos in between. There are no integrated LAN ports as you will be using either the OCP 3.0 compatible AIOM connector or one or a few of those PCIe slots for network communications. Also, to support an optional RAID controller or Host Bus Adapter (HBA) if you want a bit more control over your storage with options for data protection. In addition, there are 2x PCIe 3.0 x4 slots on the system board for NVMe M.2 drives. Those can be used to support your OS in a mirror RAID for redundancy and resiliency.

GPUs are supported. You can either install 2x double wide GPUs or 2x single-wide GPUs. There are quite a few that have been validated by Supermicro including the NVIDIA H100 80GB and H100 NVL with 94GB both with Hopper architecture. Those two are your basic go-to for AI workloads, but there’s also the L40 the L40S, plus several Ada Lovelace and Ampere options for not only AI but other workloads.  

With the NVIDIA H100 and the H100 NVL, turns out the H100 NVL has higher bandwidth HBM3 memory as opposed to the HBM2 memory on the standard H100 GPU. Also, more memory at 94GB compared to a top end of 80GB on the non-NVL model. With that, the H100 NVL is also more useful for large language models. It’s also optimized for data centers with a lower power consumption compared to the H100. Two NVIDIA H100 NVLs can also be linked together using 3x NVLink connectors on top for up to 188GB of memory when combined. Each H100 NVL also supports up to 7x 12GB MIGs or multi-instance GPUs.

This system supports dual AMD EPYC 9005-series CPUs with a Thermal Design Power Rating of up to 400W. That does limit the supported processors to 160 cores and 320 virtual threads each, for up to 320 cores and 640 virtual threads with both sockets outfitted with 9845 CPUs featuring a TDP of 390W. Next one above, which is not supported is the AMD EPYC 9965 with up to 192 cores and 384 virtual threads and a TDP of 500W. SKUs for the 9005-series CPUs start at 16 cores and a TDP of only 125W. So, there are a number of choices.

If you go with AMD EPYC 9004-series processors you can still get up to 128 physical cores and 256 virtual threads. That would be with the 9754 EPYC processor, which has a TDP of 360W. If you need to address specific applications that could benefit from more cache at the CPU then there are 9004 series CPUs with 3D V-Cache, and those top out at 400W with the 96 core 9684X and 1152MB of L3 Cache. There are 2 more with an L3 cache of up to 768MB compared to standard AMD EPYC CPUs with L3 cache ranging from 64MB to 384MB.

With 24x DDR5 DIMM slots and dual processors, each with 12-Channel memory architecture, either 9004 or 9005 series will support 1 DIMM per channel for optimal performance and top memory speeds. For 9005-series memory speeds of up to 6000MT/s are supported, while the 9004-series will provide up to 4800MT/s. With 2 DIMMs per channel that speed would be reduced, but that is not the case here. For optimal performance, there would be one DIMM in each memory channel for better memory bandwidth. That’s not to say you can’t install less than 12x DIMMs per CPU because you can. Also, up to 128 PCIe 5.0 lanes. That would be for a single socket system. With dual processors installed, it knocks it down to 160 PCIe 5.0 lanes. Still very respectable.

As part of the Hyper server line, the Supermicro Hyper A+ 2025HS-TNR Server can be used to support a number of high-performance computing workloads. With dual GPUs installed it can be used for AI Inference and machine learning applications. Not to mention virtualized environments and some high end VDI with the Multi-Instance GPU aspect.

If you have any questions on this system or another, contact us today! We are sure we have what you need, including gaming systems, parts, and other components. Check out our website.