NVIDIA RTX Ada GPUs and Financial Overview

Full-Stack & Data Center Scale Acceleration Drive Significant Cost Savings and Workload Scaling Classical Computing―960 CPU-only servers $10M (including switches, cables, racks) Application CPU server racks LLM Workload: Bert-Large Training and Inference | CPU Server: Dual-EYPC 7763 | GPU Server: Dual-EPYC 7763 + 8X H100 PCle GPUs Accelerated Computing-2 GPU servers $400K Application Re-Engineered for Acceleration CUDA-X Acceleration Libraries Magnum 10 25X lower cost 84X better energy-efficiency NVIDIA

View entire presentation