Gigabyte Servers with AMD Architectures, Air and Liquid Cooling, at Computex 2026

In the 2026 edition of Computex in Taipei, many innovations related to the world of datacenters were showcased, with an increasing number of companies presenting their solutions. This trend is evident in a market that in 2026 has shown signs of difficulty in the consumer sector but is witnessing day-by-day growth in demand for datacenter solutions for AI.

Gigabyte, through its internal division Giga Computing, offers server solutions that meet the processing needs of different types of customers. From single-unit rack proposals to more complex systems containing up to 8 GPUs for AI processing, and including workstations suited for the needs of offices and small to medium enterprises, a special space was dedicated at Computex 2026 to systems based on AMD Epyc CPU architectures and AMD Instinct for AI acceleration.

It's notable that Gigabyte's various datacenter proposals based on AMD architectures utilize different cooling systems depending on their intended use and construction complexity. In some cases, the traditional air cooling kits approach is used, while in others, liquid cooling systems are employed, which can either be closed-loop within the server or connected externally to a thermal exchange structure present in the datacenter.

In both cases, the servers are paired with processors and GPUs that stand out for their processing power, but they must be managed efficiently in terms of thermal dissipation. This leads to more or less elaborate choices for heat management within the reduced sizes typical of rack servers.

The 1U Rack Server R143-EG2-AAC2 is a classic entry-level proposal for processing environments that require high CPU computing power while allowing configuration flexibility in the graphics card area. The AMD EPYC 8005/8004 processor is paired with a 6-channel memory controller for DDR5 RDIMM modules, with a maximum of 12 modules that can be mounted on the motherboard. In a compact system like this, the number of storage units stands out, with 4 front bays for NVMe, SAS, or SATA 2.5-inch drives alongside 8 additional slots of the same size for SATA or SAS dedicated storage.

There are 3 PCI Express 16x slots, one of which is dedicated to a graphics card, while the others are reserved for other peripheral types.

The G294-Z22-AAP2 system is designed for HPC and AI processing environments in datacenters, boasting a 2U Rack design and the capacity to host up to 8 PCI Express graphics cards depending on the configuration. Processor support includes the AMD EPYC 9005 or 9004 models, paired with 12 memory slots configured for up to 12 channels. For GPUs, it can support up to 8 AMD Radeon AI PRO R9700S cards, or alternatively, NVIDIA H200 or RTX PRO 6000 cards, with a maximum of 4 that can be installed. Power is handled by a pair of 3000 Watt units, configured for redundancy.

The TD46-ZX1-LA04 system is a 4U rack server with a high processing density, featuring 8 AMD Instinct MI355X GPUs with a liquid cooling system. These solutions are paired with a pair of AMD EPYC 9005 or 9004 series processors sharing a direct liquid cooling system equipped with protection against accidental liquid spills from the circuit. Such a system is, of course, accompanied by a large amount of system memory: there are 24 slots for DDR5 RDIMM modules configured for 12-channel memory to maximize data transfer bandwidth.

W793-ZUD-LA71 is a workstation designed for demanding office and corporate environments, capable of being installed in both rack cabinets and open spaces. Inside, it accommodates an AMD EPYC 9005/9004 processor depending on the chosen configuration, paired with DDR5 RDIMM memory in 12-channel mode. The chassis can house up to 4 AMD Radeon AI PRO R970S graphics cards, equipped with an all-in-one liquid cooling system, responsible for all AI-related processing tasks.

Given the power of the integrated hardware, a redundant power system is in place: the rear features 3 2400 Watt power supplies, cooled by liquid, where one operates as redundancy in case of a failure of one of the two primary units.