https://hongping-zh.github.io/ New measurements and analyses on the energy cost of quantized LLM inference (INT8, NF4, FP16) across NVIDIA GPUs. en-us Mon, 08 Jun 2026 00:00:00 GMT https://hongping-zh.github.io/updates/2026-06-08-rtx-pro6000-fp16-e2e/ https://hongping-zh.github.io/updates/2026-06-08-rtx-pro6000-fp16-e2e/ Mon, 08 Jun 2026 00:00:00 GMT RTX PRO 6000 Blackwell FP16 end-to-end baseline for Qwen2.5-3B at 256 and 512 generated tokens, reported in J/1k tokens with throughput and power. https://hongping-zh.github.io/updates/2026-06-03-rtx-pro6000-blackwell-phase/ https://hongping-zh.github.io/updates/2026-06-03-rtx-pro6000-blackwell-phase/ Wed, 03 Jun 2026 00:00:00 GMT Supplemental RTX PRO 6000 Blackwell phase-separated energy profiling data, highlighting prefill/decode behavior and backend-architecture interaction under bitsandbytes 0.49.2. https://hongping-zh.github.io/updates/2026-04-18-qwen25-3b-t4/ https://hongping-zh.github.io/updates/2026-04-18-qwen25-3b-t4/ Sat, 18 Apr 2026 00:00:00 GMT Supplemental update: Qwen2.5-3B on Tesla T4 measurements (FP16 vs NF4), including Figure 5 summary and Table 8 results.