IN Brief:
- Aetina has added the rear-access AEX-2UA2 to its compact 2U SuperEdge server portfolio.
- The existing AEX-2UA1 has achieved NVIDIA Certified System status with two RTX PRO 6000 Blackwell Server Edition GPUs.
- Both systems use Xeon processing and configurable GPU acceleration for local computing and inference workloads.
Aetina has expanded its SuperEdge GPU server portfolio with the AEX-2UA2, a short-depth 2U system using rear-access connections, alongside NVIDIA Certified System status for a specified configuration of the existing AEX-2UA1.
The two systems combine Intel Xeon processing with high-performance NVIDIA graphics accelerators in compact enclosures intended for installations where rack depth, access arrangements, and available cooling capacity constrain the use of conventional computing servers.
Aetina’s AEX-2UA1 uses NVIDIA’s MGX modular server architecture and a single Intel Xeon 6700-series processor with performance cores. It supports two double-width graphics cards, providing a platform for accelerated computing workloads that require substantial processing capability within a relatively shallow enclosure.
NVIDIA has certified the AEX-2UA1 when configured with two RTX PRO 6000 Blackwell Server Edition GPUs. The manufacturer also identifies qualified configurations involving NVIDIA L40S and H200 NVL accelerators, offering different combinations of processing resources and accelerator memory.
The certification applies to the validated system configuration, rather than every possible combination of server components. Changes to graphics hardware, firmware, or other elements can affect whether a particular configuration corresponds to the certified system.
NVIDIA’s published certified-system listing independently identifies the AEX-2UA1 with two RTX PRO 6000 Blackwell Server Edition GPUs. The certification establishes a defined reference configuration for system integrators evaluating the platform.
The server supports graphics cards with power ratings of up to 600W each. Accommodating such components within a 2U enclosure requires attention to the server’s internal airflow, electrical power delivery, and the cooling conditions of the equipment installation.
The AEX-2UA1 measures 438 × 420 × 88mm, providing a short-depth rack-mounted platform. Conventional accelerator servers can occupy substantially deeper enclosures, making available cabinet space an important consideration for installations outside purpose-built data centre rooms.
High GPU power consumption also creates substantial thermal loads. Heat generated by the processors, memory, storage, and power conversion circuitry must be removed without allowing component temperatures to exceed their specified operating limits.
Thermal performance depends on the complete installation, including the direction and availability of airflow through the rack. Compact equipment can be particularly sensitive to blocked air paths or the recirculation of heated exhaust air.
For configurations using compatible NVIDIA H200 NVL GPUs, NVLink provides high-bandwidth communication between accelerators. This is relevant to workloads distributed across multiple GPUs, where transferring intermediate data between processors can influence overall application performance.
The availability of NVLink is configuration-dependent and should not be assumed for every graphics card supported by the platform. The software must also be able to use the available interconnect effectively.
Networking expansion includes a dedicated PCIe Gen5 x16 slot for an NVIDIA BlueField-3 data processing unit or ConnectX-7 network interface controller. These options provide additional resources for high-bandwidth network connectivity and processing functions associated with data movement.
Storage provision includes up to four hot-swappable PCIe Gen5 E1.S drive bays and two PCIe Gen5 M.2 connections. The platform supports NVMe RAID arrangements, allowing storage configurations to be selected according to application capacity, performance, and resilience requirements.
Storage throughput can influence how efficiently a computing workload is supplied with data. However, performance also depends on the application, available system memory, access patterns, and the movement of information between storage, processors, and accelerator memory.
The newly introduced AEX-2UA2 provides an alternative enclosure arrangement, with rear-access connections and a single Intel Xeon 6700-series processor with performance cores. Aetina offers configurations using NVIDIA L40S, RTX PRO 6000 Blackwell Server Edition, and H200 NVL graphics hardware.
The AEX-2UA2 retains a 438 × 420 × 88mm chassis but provides a different access arrangement for installation and maintenance. Its published specifications support up to two passive double-width PCIe GPUs or one supported active double-width graphics card.
This distinction matters when selecting a system for an existing rack. The location of connectors, the direction of airflow, and the ability to service storage or cabling can determine whether equipment can be installed without changing the surrounding infrastructure.
Aetina specifies redundant power supplies for the AEX-2UA2, with 2,000W units in a 1+1 arrangement. Actual available input power depends on the supply voltage, and component selection must remain within the thermal and electrical limits of the complete server.
The published AEX-2UA2 operating temperature range is 10°C to 35°C, reduced to 10°C to 30°C for configurations using the RTX PRO 6000 Server Edition. This requirement makes environmental control an important consideration when the system is installed in industrial or enterprise premises rather than a conventional data centre.
Both systems are intended to support locally hosted AI workloads, including language-model inference, machine vision, and other accelerated computing applications. Processing data on premises can reduce dependence on remote infrastructure, although the response time and resource requirements remain specific to the workload.
Large models may require substantial accelerator memory and, in some configurations, communication between multiple GPUs. Their performance cannot be established from parameter count alone because numerical precision, model architecture, available memory, and software optimisation all influence throughput.
The AEX-2UA2 provides a further hardware option within Aetina’s server range, while the AEX-2UA1 now has a separately documented certification milestone. Aetina has not identified the AEX-2UA2 itself as an NVIDIA Certified System in the supplied announcement.
System integrators can therefore compare the two models according to their validated GPU configurations, physical access requirements, cooling conditions, and connectivity needs before selecting hardware for production workloads.
