[mcraiha]

At least Mistral 7B for its 128 token text generation is 58% faster with 5090 compared to 4090. https://www.phoronix.com/review/nvidia-rtx5090-llama-cpp/3