GIGABYTE announced the launch of its AI TOP CXL R5X4 memory expansion card, a cutting-edge solution designed to enhance memory capacity for AI training systems. Tailored specifically for GIGABYTE’s TRX50 AI TOP and W790 AI TOP motherboards, this card leverages the Compute Express Link (CXL) standard to deliver up to 512GB of additional DDR5 RDIMM memory, making it a game-changer for professionals working with large-scale AI models and data-intensive workloads.
The AI TOP CXL R5X4 is not a universal accessory it’s exclusively compatible with GIGABYTE’s AI TOP motherboards, which support PCIe x16 interfaces with CXL 1.1 or 2.0 specifications. These include the TRX50 AI TOP for AMD Ryzen Threadripper PRO processors and the W790 AI TOP for Intel Xeon processors. GIGABYTE emphasizes that users should verify compatibility by contacting their support team, as not all PCIe slots on these motherboards support CXL (e.g., the PCIEX16_4 slot on the TRX50 AI TOP is incompatible). This targeted design ensures seamless integration with GIGABYTE’s AI-focused ecosystem, optimized for tasks like local AI model training and fine-tuning.
Source: Gigabyte
The AI TOP CXL R5X4 is built on a 16-layer High-Density Interconnect (HDI) PCB, measuring 120mm x 254mm. It connects via a PCIe 5.0 x16 interface and supports CXL 2.0/1.1 protocols, enabling efficient memory pooling and sharing for AI and compute-intensive workloads. The card features four DDR5 RDIMM slots, each supporting up to 128GB of ECC registered memory, for a total capacity of 512GB. This expands the system’s memory pool, allowing it to handle models with up to 236 billion parameters, according to GIGABYTE.
Unlike other CXL solutions, such as Samsung’s CMM-D module, the R5X4 uses standard DDR5 modules, offering users flexibility in choosing memory capacities. The card requires an 8-pin EXT12V power connector, with an estimated power draw of 70W at full load (20–30W for the Microchip PM8712 controller and 10W per DDR5 module). Cooling is managed by a single-fan heatsink, ensuring thermal stability under continuous operation.
The AI TOP CXL R5X4 is designed for AI developers, researchers, and enthusiasts running large language models (LLMs) or machine learning workloads locally. By expanding memory capacity through the PCIe bus, it alleviates pressure on the host system’s DRAM, acting as a complementary memory pool. This is particularly useful for scenarios requiring high memory bandwidth, such as training or inferencing large AI models. However, GIGABYTE notes that performance depends on software, operating system, and CXL drivers, with potential latency overheads in general-purpose tasks compared to socket-based DRAM.
The card’s integration with GIGABYTE’s AI TOP ecosystem, including the TRX50 AI TOP motherboard and AI TOP Utility software, streamlines AI development by providing a unified platform for monitoring and optimizing hardware performance. This makes it ideal for research labs, AI startups, and content creators seeking on-premises solutions without relying on cloud infrastructure.
Source: Gigabyte
While the AI TOP CXL R5X4 offers significant memory expansion, users must ensure proper case airflow and power supply capacity to support its operation, especially with four RDIMMs installed. GIGABYTE has not yet disclosed pricing or availability.