#How did Azure achieve record-breaking AI training results?
Azure, Microsoft’s cloud platform, recently achieved unparalleled AI training results, showcasing the fastest performance recorded at a large scale. This is the result of a strengthened partnership with Nvidia and focuses on the latest MLPerf Training v4.1 benchmarks, a respected independent measure for evaluating machine learning hardware effectiveness.
The breakthrough was accomplished with a powerful configuration that utilizes 512 Nvidia H200 GPUs. This setup offers a striking 28% boost in performance compared to the prior configurations that used H100 GPUs.
#What do the benchmarks reveal about Azure's capabilities?
The benchmarks from 2023 previously demonstrated that Azure could train a GPT-3 model with 175 billion parameters using 10,752 H100 GPUs in just around 4 minutes. The new configuration using H200 GPUs enhances performance significantly per GPU. This means less hardware is required to achieve similar or even better training speeds.
The success of this setup does not solely stem from the newer GPUs. Microsoft has implemented innovations across hardware, networking, and software, all contributing to the exceptional performance. Utilizing Nvidia's Quantum InfiniBand networking enables efficient data transfers among GPUs during extensive training sessions. Additionally, the integration of Nvidia’s microservices with Azure’s AI suite, including the AI Foundry platform, positions Azure as a powerful contender in the AI space.
#Who is benefiting from Azure's advancements and what is next?
Black Forest Labs, an AI firm specializing in generative image models, is already leveraging Azure’s H200 virtual machines for its innovative projects. Azure’s GPU offerings span various generations, including configurations based on H200, H100, and GB200 series GPUs. Looking forward, Microsoft indicates plans to incorporate Nvidia’s Blackwell Ultra GPUs later in 2025. These new GPUs are expected to improve performance specifically in reasoning and multimodal AI applications, further reinforcing Azure's position as a leader in cloud-based AI capabilities.