Zyphra’s unveiling of ZAYA1 marks a significant moment in the AI hardware landscape, proving that large-scale model training can succeed outside NVIDIA’s well-established ecosystem.  After a year of collaboration with AMD and IBM, Zyphra trained ZAYA1 entirely on AMD Instinct MI300X GPUs, Pensando networking, and the ROCm software stack, demonstrating that AMD’s platform can support major foundation models without exotic configurations or performance compromises (Artificial

Artificial intelligence companies, including OpenAI, are navigating unexpected delays and challenges in developing ever-larger language...