"Intel's Gaudi2 Chip Challenges NVIDIA GPUs in MLPerf Benchmarks"
TL;DR
Intel's Gaudi2 chip has emerged as the only alternative to NVIDIA GPUs for training large language models (LLMs) according to MLPerf benchmarks. Intel claims that Gaudi2 offers similar price/performance to NVIDIA's A100 in FP16 workloads and aims to surpass NVIDIA's H100 by September in FP8 workloads. The benchmarks show impressive time-to-train results for GPT-3 and other models, indicating the growing software maturity of Gaudi2. Additionally, Intel's Xeon Platinum CPUs also demonstrated strong performance in LLM training, achieving training times of less than 50 minutes for BERT and ResNet-50 models.
- Intel’s Gaudi2 Chip Is The Only Alternative To NVIDIA GPUs For LLM Training As Per MLPerf Benchmarks Wccftech
- NVIDIA H100 GPUs Set Standard for Generative AI in Debut MLPerf Benchmark Nvidia
- New benchmark tests speed of systems training ChatGPT-like chatbots Yahoo Finance
- NVIDIA And Intel Publish First GPT3 Benchmarks; AMD, AWS, And Google Are MIA Forbes
- NVDA Stock Gains on New Performance Records TipRanks
Want the full story? Read the original reporting
Read on Wccftech