Supermicro A+ Server 1024US-TRT ReviewJuly 22, 2022
The Supermicro A+ Server 1024US-TRT, a dual-socket 3rd generation AMD EPYC-based system that supports up to 8TB of memory, has a 1U form factor (SHOP HERE). It features 4x 3.5-inch hybrid storage bays on the front of the system and offers powerful compute and memory density. This system is part of Supermicro’s highest-performing Ultra servers. It’s designed for use as a high-end enterprise server, for virtualization, cloud computing and more.
Not only will this system support the 3rd generation AMD EPYC series processors with up to 64 physical cores and 128 virtual threads but also the newest of the new AMD 3rd Gen EPYC processors with AMD 3D V-Cache technology. Wait a minute! 3D V-Cache technology? We’ll get to that. These are only “new” because it’s April of 2022. Looking just a few years, (months?) into the future whoever reads this will be saying “NEW!?” Technology never ages well…
We usually start on the front of the system but let’s do this from the inside out and start with those 3rd Gen 3D V-Cache AMD EPYCs. This “new” processor is called Milan X. As you may have guessed, since it says with 3D v-Cache technology, there’s more L3 cache. AMD is calling it the first CPU with 3D chiplet technology because they stack this L3 cache memory directly on their Zen 3 chiplets for a short stack of silicon pancakes. How much L3 cache? 768MB! Consider the standard Milan CPUs have only 256MB of L3 cache, which while significant in its own right, just pales in comparison. And consider Intel’s 8380 Ice Lake Platinum processor with 40 cores only has 60MB of cache, not to mention a completely new socket design. We will mention that these direct comparisons are often somewhat misleading as the architecture is totally different. TDPs for Milan X in general remain the same at 240W and 280W depending on which of the 4x Milan X CPUs you choose. So far, there are only 4.
Milan X processors are designed for technical workloads like Structural Analysis, Electronic Design Automation, Fluid dynamics, and Finite Element Analysis, to name a few. Larger L3 cache enables better data queuing with more information stored close to the compute cores. We’re talking significant increases in performance for certain applications anywhere from 46% to 106% compared to 40 Core Ice Lake Intel processors and comparable core count 32 core processors, respectively. These Milan X CPUs use the same socket as the regular AMD Milan CPUs and are compatible with other platforms that support 3rd generation EPYCs. However, you will need to update BIOS if you want to use these.
Since this system is only sold complete with both processors installed it supports the full 32x memory module slots on the system board with each CPU in charge of 16x slots. At maximum capacity, it can be outfitted with up to 8TB of memory. Supported memory modules include registered and load-reduced modules including the 3DS variety. Memory will operate at speeds between 2666 MHz to 3200 MHz.
There are 8x internal fans with automatic speed adjustment aligned in front of each of the CPUs and just behind the backplane. The backplane on this system supports SATA3, SAS3, and NVMe drive formats with 3.5-inch hot swap drive bays in front. SATA3 is the default drive format but you can install 4x SAS3 drives with an optional SAS RAID controller. 4x NVMe drives are also supported with an optional kit that pretty much just includes a new tray with orange tabs for the NVMe drives.
We will mention that there is no provision for M.2 drives on the system board but you can install an optional dual NVMe or SATA M.2 drive Add-on-Card that can be used to boot the system. Alternatively, there are ports on the system board for two disk-on-module SATA DOMs, which can be RAIDed for redundancy.
If you do need an NVMe M.2 boot drive, that add-on-card will be installed in one of the 3x PCIe 4.0 x16 expansion slots at the back of the system. Two are full-height card slots, and the other is a low-profile slot perfect for the Supermicro MegaRAID Low Profile M.2 SSD carrier card.
You can also install a single Nvidia T4 GPU, which are great for distributed environments. While it only consumes about 70W of power you will be limited to CPUs with a TDP of 200W from the extra thermal demand. We’re not sure why they haven’t included the Nvidia A2 GPU with Ampere architecture, tensor cores and a 40-60W power draw. Perfect for AI, AI inference workloads, plus edge and industrial environments. It was also released late last year… maybe it’s just a matter of testing and updating the specs and compatible GPUs. Additional networking cards and HBAs are also supported for more seamless interaction with your network.
The ASPEED AST2500 module provides for remote and at-chassis management of the system, which is accessed from a dedicated LAN port. This seems like a good time to step out for a look at the back of the system.
On the left, two dual redundant Titanium-level 1000W PSUs, with two 10Gb Ethernet LAN ports, 2x USB 3.0 ports. 1Gb Ethernet integrated platform management interface, or IPMI port. Then a COM port with a VGA port, and the Unit ID indicator and button right between them, and then the PCIe slots above.
SuperDoctor 5 or SD5 is Supermicro’s proprietary program to monitor system health information like CPU temperatures, system voltage, power consumption, fan speeds and can be configured to send alerts via email. It comes in local and remote management versions and works with Nagios, the industry standard in systems monitoring.
Again, the front of the system offers 4x 3.5-inch drive trays supporting either SATA, SAS or NVMe drive options, which we have populated with NVMe drives.
Then there is a small control panel to the right with LEDs for general system status, LAN 1 and LAN 2, hard drives, power and a few buttons for Unit ID, reset and Power. Next to that is a bay that can be used to support a slimline optical device.
With 3rd generation AMD EPYC processor support, plus 32 memory module slots and NVMe storage options, this system is a leader in IOPS workloads. For you non-acronym people, that would be input/output per second workloads. In other words, for use as a high-performance computing enterprise server, for virtualization, in-memory database, cloud computing, AI, AI training and inference. There are a lot of possibilities for this system. As a small system there’s really not much else to show.
If you’re interested in this Supermicro server, or are looking for something similar through a different manufacturer, contact us today! We have a huge warehouse full of the latest systems, or if you want an older system, we can get you those too!