Groq LPU Inference Engine
Comparison of Groq LPU die fabricated domestically in older 14nm process versus Nvidia GPU fabricated in the latest 4nm process.
Comparison of Groq LPU die fabricated domestically in older 14nm process versus Nvidia GPU fabricated in the latest 4nm process.
It's not that hard to build better processors than Intel/AMD and NVIDIA, it just hard to do it in a way that it's a drop-in replacement. The main problem for the likes of Groq is that they didn't build a general-purpose in-memory machine that can run legacy code, or something that just does a better job with CUDA.