GIGABYTE G293-Z43 4U GPU Server with 8x MI300X

The GIGABYTE G293-Z43 (part number G293-Z43-AAX1) is a 4U rack-mounted GPU server engineered around eight AMD Instinct MI300X accelerators, each integrating a unified CPU-GPU architecture with 192 GB of HBM3 memory per module — delivering an aggregate of 1.536 TB of high-bandwidth memory across the full system. This exceptional memory capacity eliminates the data-transfer bottlenecks that constrain traditional discrete GPU configurations, enabling trillion-parameter large language models and other memory-intensive AI workloads to reside entirely within on-package memory at sustained throughput. The platform is built on AMD's CDNA 3 compute architecture, providing double-precision, single-precision, and mixed-precision (FP8/FP16/BF16) compute paths optimized for both training and inference pipelines.

The G293-Z43 is designed around a dual AMD EPYC (Genoa-class) processor foundation that provides the PCIe and fabric connectivity required to feed eight MI300X accelerators with sufficient host-side bandwidth. The system supports high-speed NVMe storage, enterprise-grade networking provisions, and a fully redundant power subsystem with hot-swap capability, reflecting the operational availability requirements of production datacenter environments. Liquid-cooling-ready thermal design accommodates the substantial thermal design power of the MI300X array, and the 4U chassis balances compute density with serviceability in standard 19-inch racks.

As a purpose-built AI and HPC infrastructure platform, the G293-Z43 targets organizations deploying large-scale generative AI services, scientific simulation, financial modeling, and seismic analysis at production scale. Its compatibility with AMD ROCm software stack, OpenCL, HIP, and popular ML frameworks including PyTorch and TensorFlow makes it readily integrable into existing MLOps pipelines. Omnixon Global supplies the G293-Z43-AAX1 to enterprise buyers across the UAE, GCC, EMEA, and APAC regions.

Ideal for

• Large language model (LLM) training and fine-tuning for generative AI applications requiring multi-hundred-billion to trillion-parameter model support within a single server node
• High-throughput LLM inference serving, leveraging 1.536 TB aggregate HBM3 capacity to host multiple large models concurrently and reduce inter-node communication overhead
• High-performance computing simulations including computational fluid dynamics, molecular dynamics, and climate modeling that benefit from massive on-device memory and FP64 compute
• Multi-modal AI workloads combining vision, language, and structured data pipelines where unified memory architecture reduces staging and preprocessing latency
• Financial risk modeling, Monte Carlo simulation, and quantitative analytics requiring high-throughput mixed-precision compute at datacenter scale
• Seismic data processing and energy-sector subsurface imaging tasks that demand sustained double-precision throughput and large working-set memory capacity

Technical specifications

Available from Omnixon Global. Submit an RFQ and our team will confirm configuration and availability for your order.