3 ms·
For LLMs and other pure memory-bound workloads, but for eg. diffusion models their FPU SIMD performance is lacking.
by tannhaeuser 6mo ago
For LLMs and other pure memory-bound workloads, but for eg. diffusion models their FPU SIMD performance is lacking.