{"as_of":"2026-10-12","license":"CC-BY-4.0","license_url":"https://creativecommons.org/licenses/by/4.0/","cite_as":"The Local Bench (thelocalbench.com), data as of 2026-10-12, CC BY 4.0","source_url":"https://thelocalbench.com","hardware":{"id":"rtx-4090-24g","name":"GeForce RTX 4090 24 GB","vendor":"NVIDIA","class":"Discrete GPU","kind":"gpu","as_of":"2026-10-12","summary":"Fastest 24 GB card; the extra compute shows up in prefill more than decode.","memory_gb":{"value":24,"flag":"sourced","unit":"GB","source":"https://www.nvidia.com/en-us/geforce/graphics-cards/40-series/rtx-4090/","note":"Vendor spec page, checked 2026-10-12."},"bandwidth_gbs":{"value":1008,"flag":"sourced","unit":"GB/s","source":"https://en.wikipedia.org/wiki/GeForce_RTX_40_series","note":"NVIDIA's GeForce spec pages list bus width but not bandwidth; figure from the cited Wikipedia spec table (bus width cross-checked on nvidia.com, 2026-10-12)."},"fp16_tflops":{"value":165,"flag":"sample","unit":"TFLOPS","note":"Sample FP16 dense compute; pending vendor-sheet citation."},"host_ram_gb":{"value":64,"flag":"sample","unit":"GB","note":"Assumed host RAM for offload."},"host_bandwidth_gbs":{"value":64,"flag":"sample","unit":"GB/s"},"load_watts":{"value":500,"flag":"sample","unit":"W","note":"Sample wall power under load."},"tdp_w":{"value":450,"flag":"sourced","unit":"W","source":"https://www.nvidia.com/en-us/geforce/graphics-cards/40-series/rtx-4090/","note":"Total Graphics Power, nvidia.com."},"price_usd":{"value":4395,"flag":"sourced","unit":"USD","source":"https://www.newegg.com/p/pl?d=rtx+4090+24gb+graphics+card&N=4131","note":"Lowest new in-stock RTX 4090 listing on Newegg US (incl. marketplace sellers; card discontinued, scarce). Checked 2026-10-12."},"price_cad":{"value":5599.99,"flag":"sourced","unit":"CAD","source":"https://www.newegg.ca/p/pl?d=rtx+4090+24gb+graphics+card&N=4131","note":"Lowest new in-stock listing on Newegg Canada (MSI Gaming Trio, ships from Canada). Checked 2026-10-12."}},"runs":[{"model":"gemma-3-27b","hardware":"rtx-4090-24g","quant":"Q4_K_M","runtime":"llama.cpp (CUDA)","test":"tg128 @ 8k depth","tok_s":33.4,"min":32.8,"max":33.9,"repeats":5,"flag":"sample","date":"2026-10-12","note":"Placeholder figure from the approved mockup; not a measurement."},{"model":"llama-2-7b","hardware":"rtx-4090-24g","quant":"Q4_0","runtime":"llama.cpp (CUDA)","test":"tg128","tok_s":186.21,"flag":"sourced","date":"2026-10-12","source":"https://github.com/ggml-org/llama.cpp/discussions/15013","note":"Llama 2 7B Q4_0, llama-bench tg128, full GPU offload, flash attention off, single GPU. Community scoreboard entry; commit and submitter in the source table.","min":186.08,"max":186.34}]}