ASUS has revealed that they will present their MGX Series AI supercomputing-ready solutions at GTC 2024.
While not the latest Grace x Blackwell products, these 2U servers, powered by NVIDIA MGX’s modular reference architecture with model names ESC NM1-E1 and ESC NM2-E1, still packs a punch.
They feature the GH200 Grace Hopper Superchip, integrating Arm Neoverse V9 CPU cores with Scalable Vector Extensions (SVE2), NVLink-C2C, BlueField-3 DPUs, and ConnectX-7 network adapters, delivering data throughput of up to 400Gbps. NVIDIA’s AI enterprise software stack equips these servers with a variety of tools and platforms for building and deploying enterprise-grade AI applications, enhancing their appeal.
Additionally, ASUS employs a Direct-to-Chip (D2C) cooling solution to rapidly deploy and improve Power Usage Effectiveness (PUE) ratios over time. While ESC N8-E11 and RS720QN-E11-RS24U models support manifolds and cooling plates, all ASUS servers can accommodate a rear-door heat exchanger, designed to fit standard rack-server configurations for industry-standard deployment.
Over at booth no.730, ASUS will be demoing the ESC4000A-E12, showcasing “No-Code AI” capabilities through an integrated software stack facilitating accelerated AI development, including LLM pre-training, fine-tuning, and inference, with token counts ranging from 7B to 180B, tailored to client requirements to maximize the efficiency and ROI game.