#fp8

• [0007]2026-05-13From 429 GB/s to the DRAM wall: writing an FP8 quantizer on an RTX 5080#cuda#quantization#fp8#blackwell#sm120#rtx#5080
• [0006]2026-05-078.5x Faster Speech-to-Text: From 429ms to 50ms on a Single GPU#inference#stt#fp8#hopper#sm90

all posts →