{"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":"mac-studio-m5-ultra","name":"Mac Studio M5 Ultra (base)","vendor":"Apple","class":"Apple unified","kind":"unified","as_of":"2026-10-12","summary":"1.2 TB/s of unified bandwidth: the fastest way to run 70B dense models at home.","memory_gb":{"value":96,"flag":"sourced","unit":"GB","source":"https://www.apple.com/mac-studio/specs/","note":"Base configuration, apple.com, checked 2026-10-12."},"bandwidth_gbs":{"value":1200,"flag":"sourced","unit":"GB/s","source":"https://www.apple.com/mac-studio/specs/","note":"1.2 TB/s, apple.com, checked 2026-10-12."},"fp16_tflops":{"value":80,"flag":"sample","unit":"TFLOPS","note":"Sample FP16 dense compute; pending vendor-sheet citation."},"host_ram_gb":{"value":0,"flag":"sample","unit":"GB","note":"Assumed host RAM for offload."},"host_bandwidth_gbs":{"value":1200,"flag":"sourced","unit":"GB/s","source":"https://www.apple.com/mac-studio/specs/","note":"Unified memory."},"load_watts":{"value":240,"flag":"sample","unit":"W","note":"Sample wall power under load."},"tdp_w":{"value":480,"flag":"sourced","unit":"W","source":"https://www.apple.com/mac-studio/specs/","note":"Maximum continuous power of the whole Mac Studio (apple.com); Apple publishes no chip TDP."},"price_usd":{"value":5499,"flag":"sourced","unit":"USD","source":"https://www.apple.com/shop/buy-mac/mac-studio","note":"Apple US store, base M5 Ultra configuration. Checked 2026-10-12."},"apple":true,"price_cad":{"value":7799,"flag":"sourced","unit":"CAD","source":"https://www.apple.com/ca/shop/buy-mac/mac-studio","note":"Apple Canada store, base M5 Ultra configuration. Checked 2026-10-12."}},"runs":[{"model":"llama-2-7b","hardware":"mac-studio-m5-ultra","quant":"Q4_0","runtime":"llama.cpp (Metal)","test":"tg128","tok_s":179.1,"flag":"sourced","date":"2026-10-12","source":"https://github.com/ggml-org/llama.cpp/discussions/4167","note":"Llama 2 7B Q4_0, llama-bench tg128, measured on the M5 Ultra 80-core GPU option (1.2 TB/s), commit c1d0e7a. The base 64-core model may be slightly slower."}]}