Add RTX 4060 8GB, 5070 Ti, 5080, 5090 sweeps (427 rows, 10 GPUs)

README.md

@@ -19,7 +19,7 @@ configs:
 
 # Local LLM Advisor: measured llama.cpp benchmarks
 
-253 `llama-bench` measurements taken by [Local LLM Advisor](https://localllmadvisor.com) on rented NVIDIA GPUs, covering 6 GPUs (rtx-3060-12, rtx-3090, rtx-4060ti-16, rtx-4070, rtx-4090, rtx-5070), 11 GGUF models and context depths of 0, 4096, 16384 tokens. Every row is our own run; **no third-party benchmark rows are included** in this dataset.
+427 `llama-bench` measurements taken by [Local LLM Advisor](https://localllmadvisor.com) on rented NVIDIA GPUs, covering 10 GPUs (rtx-3060-12, rtx-3090, rtx-4060, rtx-4060ti-16, rtx-4070, rtx-4090, rtx-5070, rtx-5070ti, rtx-5080, rtx-5090), 11 GGUF models and context depths of 0, 4096, 16384 tokens. Every row is our own run; **no third-party benchmark rows are included** in this dataset.
 
 ## Attribution (CC BY 4.0)
 
@@ -34,7 +34,7 @@ The full leaderboard (which also aggregates public community results under their
 - Tool: `llama-bench` from llama.cpp build b11658 (commit 3e4dfa8f3), CUDA 12.8 build, flash attention on, all layers on GPU (`-ngl 99`), 3 repetitions (`-r 3`), single GPU.
 - `pp_tps` = prompt processing tokens/s (`pp_n` prompt tokens, default 512); `tg_tps` = generation tokens/s (`tg_n` tokens, default 128). Rows with `depth` > 0 measure speed with that many tokens already in the context.
 - Rented machines differ in CPU and RAM; per-row details (driver, CPU, GGUF file, standard deviations) are in the `notes` column.
-- Date of all runs: 2026-10-16.
+- Dates of runs: 2026-10-16, 2026-10-17, 2026-10-18.
 
 ## Columns