NVIDIA Blackwell platform dominates MLPerf Training 6.0 benchmarks

NVIDIA's Blackwell platform has set new records in the MLPerf Training 6.0 benchmarks, achieving the fastest times across all seven tests. The platform demonstrated strong scaling, with clusters of up to 8,192 GPUs showing significant speedups in training large language models. This performance highlights the importance of high-bandwidth interconnects like NVLink and low-precision computation for efficient large-scale AI training. AI

IMPACT Sets new SOTA on training benchmarks, highlighting the critical role of interconnects and low-precision math for large-scale AI model development.

RANK_REASON The item details performance on industry-standard benchmarks (MLPerf Training 6.0) for new hardware (NVIDIA Blackwell), which is a research-focused outcome. [lever_c_demoted from research: ic=1 ai=1.0]

Read on dev.to — LLM tag →

• Microsoft Azure
• Blackwell
• CoreWeave
• DeepSeek-V3 671B
• Fp8
• graphics processing unit
• Llama 3.1 405B
• MLPerf Training 6.0
• NVFP4
• NVIDIA
• NVL72
• NVLink

• infra
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →