MLPerf 3 Upsets the AI Apple Cart - Gestalt IT Rundown (34 m)
Gestalt IT, Wednesday, June 28,2023
Training 3.0 benchmark results show performance gains of up to 1.54x compared to six months ago and 33-49x improvement over the first round, driving innovation and energy efficiency in the industry
Intel's Habana Gaudi2 ML training engine competes with Nvidia's offerings, boasting better performance than A100 and lower pricing than H100. Nvidia, on the other hand, unveils their NeMo model with half a trillion parameters and expands the MLPerf Training suite to include GPT-3 and a new Recommendation engine. Their collaboration with CoreWeave showcases the superior performance of the H100, providing a 3.6x speed increase for GPT-3 compared to Intel Xeon and Gaudi2.