Description
Deliver enterprise AI inference, fine-tuning, rendering, and HPC workloads with the Supermicro SYS-522GA-NRT, a 5U dual-socket GPU server built on Intel’s Granite Rapids platform and armed with 8 NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs. Powered by dual Intel Xeon 6952P processors delivering 192 cores and 384 threads, 1.5TB of DDR5-6400 ECC memory, and 768GB of aggregate GDDR7 GPU memory across all eight GPUs, this system is purpose-built for large-scale AI inference, LLM fine-tuning, generative AI, visual computing, 3D rendering, VDI, and scientific simulation.
Each RTX PRO 6000 Blackwell Server Edition GPU delivers 120 TFLOPS of FP32 performance and 4 PFLOPS of FP4 AI throughput via fifth-generation Tensor Cores, connected over PCIe 5.0 x16 in a dual-root topology for maximum host-to-GPU bandwidth. Networking is handled by an NVIDIA ConnectX-6 DX dual-port 100GbE QSFP28 AIOM card for high-speed storage and cluster interconnect traffic, supplemented by dual onboard 10GbE RJ45 ports for management. The dual-root PCIe 5.0 architecture distributes GPU lanes evenly across both processors, minimizing contention and enabling consistent performance under heavy multi-GPU workloads.
Storage includes a 960GB PCIe Gen4 M.2 NVMe SED boot drive and 4x Samsung PM9D3a 3.84TB PCIe 5.0 x4 U.2 NVMe SEDs for 15.36TB of raw high-speed data storage, with the chassis supporting up to 24 hot-swap 2.5″ NVMe bays for future expansion. Air-cooled with 10 heavy-duty PWM fans and passive GPU cooling, the SYS-522GA-NRT features TPM 2.0, Secure Boot, and Silicon Root of Trust for hardened security. Backed by 3-year on-site next-business-day service and Supermicro DCMS management software, this server is ready to deploy in production AI data centers, neoclouds, and enterprise rendering farms.
