CoreWeave Achieves Record MLPerf Benchmark with NVIDIA GB200 Superchips

CoreWeave, in collaboration with NVIDIA and IBM, has set a groundbreaking record on the MLPerf Training v5.0 benchmark using the NVIDIA GB200 Grace Blackwell Superchips. The achievement represents the largest submission ever recorded for the benchmark, utilizing nearly 2,500 NVIDIA GB200 GPUs, according to PR Newswire.

Breakthrough Performance

The submission leveraged the power of 2,496 NVIDIA Blackwell GPUs on CoreWeave's AI-optimized cloud platform. This setup allowed CoreWeave to achieve a breakthrough in benchmarking, particularly on the complex Llama 3.1 405B model, completing the task in 27.3 minutes. This performance is over twice as fast as comparable submissions, underscoring the capabilities of the GB200 NVL72 architecture and CoreWeave's cutting-edge infrastructure.

Implications for AI Development

The implications of these results are significant for AI development, as they facilitate faster model development cycles and reduce the Total Cost of Ownership for AI projects. By cutting training times in half and optimizing cost-efficiency, CoreWeave provides its clients with a competitive edge in deploying and developing AI models.

CoreWeave's Industry Leadership

Peter Salanki, CoreWeave's Chief Technology Officer and Co-founder, emphasized the company's commitment to delivering a high-performance cloud platform tailored for AI workloads. CoreWeave's advancements are not just theoretical; they are practically applied, as evidenced by their leading positions in both MLPerf Inference v5.0 and Training v5.0 benchmarks.

CoreWeave's recognition extends beyond technical benchmarks. It has been named among the TIME100 most influential companies and featured on Forbes Cloud 100 in 2024, reflecting its growing influence in the tech industry.

About CoreWeave

Founded in 2017, CoreWeave operates a network of data centers across the US and Europe, specializing in AI-optimized cloud solutions. The company is known as the AI Hyperscaler™ and continues to push the envelope in accelerated computing for enterprises and AI labs worldwide.