Jim Keller, a renowned processor architect known for his work on various processors, recently critiqued Nvidia’s CUDA architecture, comparing it to a swamp similar to x86. He criticized CUDA for its complex structure and backward compatibility, noting that it may not be as efficient as open-source software frameworks. Keller highlighted Nvidia’s tools like the Triton Inference Server and TensorRT, which offer optimized performance for AI models. While Keller’s views on AMD’s ROCm and Intel’s OneAPI are unclear, it’s apparent that he has reservations about x86’s future. His insights suggest that Nvidia’s platform may not align with his vision, despite his impressive career in the tech industry.
https://www.tomshardware.com/tech-industry/artificial-intelligence/jim-keller-criticizes-nvidias-cuda-and-x86-cudas-a-swamp-not-a-moat-x86-was-a-swamp-too