Hardware Datasheet · GPU Server
Nexus Compute
Lenovo ThinkSystem SR685a V3 — 8x MI300X for Large-Memory Inference
192GB-per-GPU inference node tuned for high-throughput ROCm serving.
Overview
The Lenovo ThinkSystem SR685a V3 hosts eight MI300X accelerators on dual AMD EPYC processors, an 8U platform optimized for serving large models where per-GPU memory capacity is the binding constraint. Nexus Compute configures the ROCm inference stack, GPUDirect networking, and storage, then delivers each tested system warranty-backed through authorized Lenovo channels.
Specifications
| Accelerator | 8x AMD Instinct MI300X, 192GB HBM3 each (1.5TB total) |
| GPU Interconnect | AMD Infinity Fabric, 6.4TB/s GPU-to-GPU |
| CPU | Dual AMD EPYC 9004/9005 (Genoa/Turin) |
| System Memory | Lenovo TruDDR5 up to 6000 MHz (24 DIMMs) |
| Storage | Up to 16x hot-swap PCIe Gen5 NVMe |
| Networking/Fabric | 8x front PCIe Gen5 FHHL slots with GPUDirect |
| Form Factor | 8U rackmount |
| Management | Lenovo XClarity Controller (XCC) |
| Warranty | Lenovo enterprise warranty (configurable term) |
Typical Use Cases
- ·High-concurrency LLM and generative-model serving
- ·Private, on-premises inference for regulated data
- ·Retrieval-augmented generation at scale
- ·Long-context inference workloads
Industries
Warranty & Support
Supplied through authorized channels with full manufacturer warranty. On-site, next-business-day support options available. Every system is configured, tested, and documented before delivery, with asset and warranty records provided for enterprise audit requirements.
Request a tailored quote
Configurations are tailored per engagement — contact us for pricing and lead times.
sales@nexus-compute.com
+1 737 276 1016
nexus-compute.com
Specifications are indicative and configured to each engagement. All product names, logos, and trademarks are the property of their respective owners. Nexus Compute is an independent enterprise hardware supplier.