The AMD Instinct MI300X 8-GPU AI/LLM Server is a 6U rackmount system built on the AsRock 6U8M-GENOA2 platform, purpose-built for large language model training, generative AI, and high-performance computing workloads, with support for dual AMD EPYC 9004 Series processors, high-capacity DDR5 memory, 8x AMD Instinct MI300X GPUs, and an all-NVMe flash storage backplane.
The server supports 2x AMD EPYC 9004 Series processors using the SP5 socket platform, with dual 96-core AMD EPYC 9654 CPUs included standard. Memory capacity reaches up to 2.3TB of DDR5 ECC Registered memory across 24 DIMM slots, running at 5600MHz. For AI acceleration, the server features 8x AMD Instinct MI300X GPUs with Infinity Fabric Links, delivering up to 1.5TB of total HBM3 capacity. For storage, the server features a 12-bay 2.5" NVMe hot-swap backplane, with 8 bays connected via PCIe switch and 4 bays CPU-direct, all on PCIe 5.0 x4 lanes.
This server is well suited for environments such as:
- Large language model (LLM) training and fine-tuning
- Generative AI and deep learning workloads
- High-performance computing (HPC) and scientific simulation
- AI inference at scale
- Enterprise applications that benefit from massive HBM3 capacity and all-flash NVMe storage
Expansion includes PCIe 5.0 x16 add-on card slots populated with 2x Broadcom BCM57508 100GbE dual-port QSFP56 NICs, with an optional upgrade to 2x Broadcom 400GbE QSFP-DD NICs for maximum cluster interconnect bandwidth. Power is supplied by 8x 3000W CRPS redundant power supplies (4+4) rated 80 PLUS Titanium. Rack rails are included.
Every unit is fully tested before it ships and configured to your exact specs — processors, memory, drives, networking, and available controller options are installed and verified.
Our team is available by phone or chat whenever you need us. Browse our full selection of custom-configured refurbished IT equipment at The Server Store.
