EVENT DOSSIER
On September 8, 2026, Amazon Web Services released a benchmark report on its machine learning blog, comparing the performance of G7, G5, and G6 GPU instances on the SageMaker AI platform. The newly launched G7 instances are equipped with NVIDIA Blackwell GPUs, achieving measurable improvements in throughput, latency, and cost per token. The tests were conducted using two scenarios: one involved using the Qwen3-Coder-30B model to compare the performance and price of ml.g5.12xlarge (A10G), ml.g6.12xlarge (L4), and ml.g7.12xlarge (RTX PRO 4500 Blackwell) instances; the other scenario used the NVIDIA Nemotron-3-Nano-30B-A3B-NVFP4 model to evaluate G6, G6e, and G7 configurations in combination with vLLM to identify the most cost-effective solution. The article also…
亚马逊云科技在 SageMaker AI 上对比 G7、G5 和 G6 GPU 实例，验证新推出的 G7 实例（搭载 NVIDIA Blackwell GPU）在吞吐量、延迟及每 token 成本方面均实现可衡量的提升。文章通过两个用例展示评估方法：一是使用 Qwen3-Coder-30B 模型，对比 ml.g5.12xlarge (A10G)、ml.g6.12xlarge (L4) 和 ml.g7.12xlarge (RTX PRO 4500 Blackwell) 实例的性能与价格；二是使用 NVIDIA Nemotron-3-Nano-30B-A3B-NVFP4 模型，结合 vLLM 评估 G6、G6e 和 G7 配置以识别最佳性价比方案。此外，文章介绍了 SageMaker AI 生成式 AI 推理推荐…