{ "query": { "task": "LLM generation", "model_key": "llama323b" }, "task": "LLM generation", "size_class": "by model", "batch": "per group", "mixed": false, "mixed_warning": null, "groups": [ { "size": "LLM ≤4B", "batch": "1", "platforms": 35, "rows": 43 } ], "rank_groups": [ { "model_key": "llama323b", "label": "Llama 3.2:3b", "platforms": 35, "platforms_with_power": 0 } ], "reference": null, "excluded_count": 0, "efficiency_excluded_count": 26, "notes": [ "edge hardware only", "model_key \"llama323b\"" ], "count": 35, "results": [ { "platform": "Dell XPS 13 (Intel Core 5 320) (as tested by Jeff Geerling)", "platform_url": "/platform/dell-xps-13-intel-core-5-320-as-tested-by-jeff-geerling.html", "perf": 23.88, "perf_unit": "tok/s", "metric": "generation_throughput", "model": "Llama 3.2:3b", "precision": "as in linked issue (Ollama default tags are 4-bit)", "stack": "llama.cpp (Vulkan)", "size_class": "LLM ≤4B", "batch": "1", "like_for_like": true, "differs": [], "confidence": "C", "confidence_label": "C – community/review", "power_w": 14, "power_label": "reported in result source — Measured, Whole system", "power_kind": "Measured, Whole system", "model_key": "llama323b", "pb": { "tag": "Measured", "measured": true, "vendorMeasured": false, "typical": false, "scope": "Whole system", "rmax": false, "rmin": false }, "power_scope": "whole-system", "power_is_range_max": false, "power_is_range_min": false, "is_partial": "false", "partial_note": null, "rank_default": false, "fpsw_outlier": false, "fpsw_note": null, "rank_note": null, "rank_fps": false, "rank_fpsw": false, "fpsw_ratio_median": null, "power_basis": "measured peak system power (Jeff Geerling, as published)", "power_note": "Measured", "power_range": null, "power_src_url": "https://github.com/geerlingguy/ai-benchmarks/issues/52", "perf_per_w": 1.706, "date": "2026-09-22 (page last edited)", "older_than_18_months": false, "superseded": false, "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": true, "date_basis": "page-edited", "form_factor": "independent test system", "vendor": "(independent test)", "source": { "name": "Jeff Geerling ai-benchmarks README (Findings tables)", "url": "https://github.com/geerlingguy/ai-benchmarks/issues/52", "date": null, "doc": null, "version": "ai-benchmarks README @ 2f56f19", "vendor_page": null }, "power_source": "https://github.com/geerlingguy/ai-benchmarks/issues/52", "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "no_power": false, "group": "llama323b", "excluded_reason": null, "excluded": false, "fits": true, "speed_eligible": true, "eff_ok": false, "soft_reason": "HWBench Data has not verified tok/s/W rows for ranking yet", "eligible": false, "headroom": null, "unknown_badges": [ "price unknown" ], "group_size": 0, "group_rank": null }, { "platform": "Apple M4 (10-core GPU, 120 GB/s)", "platform_url": "/platform/apple-m4-10-core-gpu-120-gb-s.html", "perf": 41.31, "perf_unit": "tok/s", "metric": "generation_throughput", "model": "Llama 3.2:3b", "precision": "as in linked issue (Ollama default tags are 4-bit)", "stack": "Ollama / llama.cpp (see linked issue)", "size_class": "LLM ≤4B", "batch": "1", "like_for_like": true, "differs": [], "confidence": "C", "confidence_label": "C – community/review", "power_w": 30.1, "power_label": "reported in result source — Measured, Whole system", "power_kind": "Measured, Whole system", "model_key": "llama323b", "pb": { "tag": "Measured", "measured": true, "vendorMeasured": false, "typical": false, "scope": "Whole system", "rmax": false, "rmin": false }, "power_scope": "whole-system", "power_is_range_max": false, "power_is_range_min": false, "is_partial": "false", "partial_note": null, "rank_default": false, "fpsw_outlier": false, "fpsw_note": null, "rank_note": null, "rank_fps": false, "rank_fpsw": false, "fpsw_ratio_median": null, "power_basis": "measured peak system power (Jeff Geerling, as published)", "power_note": "Measured", "power_range": null, "power_src_url": "https://github.com/geerlingguy/ai-benchmarks/issues/2", "perf_per_w": 1.372, "date": "2026-09-22 (page last edited)", "older_than_18_months": false, "superseded": false, "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": true, "date_basis": "page-edited", "form_factor": "independent test system", "vendor": "(independent test)", "source": { "name": "Jeff Geerling ai-benchmarks README (Findings tables)", "url": "https://github.com/geerlingguy/ai-benchmarks/issues/2", "date": null, "doc": null, "version": "ai-benchmarks README @ 2f56f19", "vendor_page": null }, "power_source": "https://github.com/geerlingguy/ai-benchmarks/issues/2", "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "no_power": false, "group": "llama323b", "excluded_reason": null, "excluded": false, "fits": true, "speed_eligible": true, "eff_ok": false, "soft_reason": "HWBench Data has not verified tok/s/W rows for ranking yet", "eligible": false, "headroom": null, "unknown_badges": [ "price unknown" ], "group_size": 0, "group_rank": null }, { "platform": "M4 Mac mini (10 core CPU) / 32GB (as tested by Jeff Geerling)", "platform_url": "/platform/m4-mac-mini-10-core-cpu-32gb-as-tested-by-jeff-geerling.html", "perf": 41.31, "perf_unit": "tok/s", "metric": "generation_throughput", "model": "llama3.2:3b", "precision": "Ollama default tag (4-bit unless tag says otherwise)", "stack": "Ollama (geerlingguy/ollama-benchmark)", "size_class": "LLM ≤4B", "batch": "1", "like_for_like": true, "differs": [], "confidence": "C", "confidence_label": "C – community/review", "power_w": 30.1, "power_label": "reported in result source — Measured, Whole system", "power_kind": "Measured, Whole system", "model_key": "llama323b", "pb": { "tag": "Measured", "measured": true, "vendorMeasured": false, "typical": false, "scope": "Whole system", "rmax": false, "rmin": false }, "power_scope": "whole-system", "power_is_range_max": false, "power_is_range_min": false, "is_partial": "false", "partial_note": null, "rank_default": false, "fpsw_outlier": false, "fpsw_note": null, "rank_note": null, "rank_fps": false, "rank_fpsw": false, "fpsw_ratio_median": null, "power_basis": "measured peak system power (Jeff Geerling)", "power_note": "Measured", "power_range": null, "power_src_url": "https://github.com/geerlingguy/sbc-reviews/issues/57", "perf_per_w": 1.372, "date": "2024-11-11 (page last edited)", "older_than_18_months": true, "superseded": false, "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": true, "date_basis": "page-edited", "form_factor": "independent test system", "vendor": "(independent test)", "source": { "name": "Jeff Geerling sbc-reviews (AI / LLM inference tables)", "url": "https://github.com/geerlingguy/sbc-reviews/issues/57", "date": null, "doc": null, "version": "ai-benchmarks README @ 2f56f19", "vendor_page": null }, "power_source": "https://github.com/geerlingguy/sbc-reviews/issues/57", "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "no_power": false, "group": "llama323b", "excluded_reason": null, "excluded": false, "fits": true, "speed_eligible": true, "eff_ok": false, "soft_reason": "HWBench Data has not verified tok/s/W rows for ranking yet", "eligible": false, "headroom": null, "unknown_badges": [ "price unknown" ], "group_size": 0, "group_rank": null }, { "platform": "Dell Pro Max with GB10", "platform_url": "/platform/dell-pro-max-with-gb10.html", "perf": 97.93, "perf_unit": "tok/s", "metric": "generation_throughput", "model": "Llama 3.2:3b", "precision": "as in linked issue (Ollama default tags are 4-bit)", "stack": "Ollama / llama.cpp (see linked issue)", "size_class": "LLM ≤4B", "batch": "1", "like_for_like": true, "differs": [], "confidence": "C", "confidence_label": "C – community/review", "power_w": 128.5, "power_label": "reported in result source — Measured, Whole system", "power_kind": "Measured, Whole system", "model_key": "llama323b", "pb": { "tag": "Measured", "measured": true, "vendorMeasured": false, "typical": false, "scope": "Whole system", "rmax": false, "rmin": false }, "power_scope": "whole-system", "power_is_range_max": false, "power_is_range_min": false, "is_partial": "false", "partial_note": null, "rank_default": false, "fpsw_outlier": false, "fpsw_note": null, "rank_note": null, "rank_fps": false, "rank_fpsw": false, "fpsw_ratio_median": null, "power_basis": "measured peak system power (Jeff Geerling, as published)", "power_note": "Measured", "power_range": null, "power_src_url": "https://github.com/geerlingguy/ai-benchmarks/issues/34", "perf_per_w": 0.7621, "date": "2026-09-22 (page last edited)", "older_than_18_months": false, "superseded": false, "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": true, "date_basis": "page-edited", "form_factor": "independent test system", "vendor": "(independent test)", "source": { "name": "Jeff Geerling ai-benchmarks README (Findings tables)", "url": "https://github.com/geerlingguy/ai-benchmarks/issues/34", "date": null, "doc": null, "version": "ai-benchmarks README @ 2f56f19", "vendor_page": null }, "power_source": "https://github.com/geerlingguy/ai-benchmarks/issues/34", "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "no_power": false, "group": "llama323b", "excluded_reason": null, "excluded": false, "fits": true, "speed_eligible": true, "eff_ok": false, "soft_reason": "HWBench Data has not verified tok/s/W rows for ranking yet", "eligible": false, "headroom": null, "unknown_badges": [ "price unknown" ], "group_size": 0, "group_rank": null }, { "platform": "Mac Studio (M3 Ultra 512GB) (as tested by Jeff Geerling)", "platform_url": "/platform/mac-studio-m3-ultra-512gb-as-tested-by-jeff-geerling.html", "perf": 154.6, "perf_unit": "tok/s", "metric": "generation_throughput", "model": "Llama 3.2:3b", "precision": "as in linked issue (Ollama default tags are 4-bit)", "stack": "Ollama / llama.cpp (see linked issue)", "size_class": "LLM ≤4B", "batch": "1", "like_for_like": true, "differs": [], "confidence": "C", "confidence_label": "C – community/review", "power_w": 223, "power_label": "reported in result source — Measured, Whole system", "power_kind": "Measured, Whole system", "model_key": "llama323b", "pb": { "tag": "Measured", "measured": true, "vendorMeasured": false, "typical": false, "scope": "Whole system", "rmax": false, "rmin": false }, "power_scope": "whole-system", "power_is_range_max": false, "power_is_range_min": false, "is_partial": "false", "partial_note": null, "rank_default": false, "fpsw_outlier": false, "fpsw_note": null, "rank_note": null, "rank_fps": false, "rank_fpsw": false, "fpsw_ratio_median": null, "power_basis": "measured peak system power (Jeff Geerling, as published)", "power_note": "Measured", "power_range": null, "power_src_url": "https://github.com/geerlingguy/ai-benchmarks/issues/42", "perf_per_w": 0.6933, "date": "2026-09-22 (page last edited)", "older_than_18_months": false, "superseded": false, "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": true, "date_basis": "page-edited", "form_factor": "independent test system", "vendor": "(independent test)", "source": { "name": "Jeff Geerling ai-benchmarks README (Findings tables)", "url": "https://github.com/geerlingguy/ai-benchmarks/issues/42", "date": null, "doc": null, "version": "ai-benchmarks README @ 2f56f19", "vendor_page": null }, "power_source": "https://github.com/geerlingguy/ai-benchmarks/issues/42", "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "no_power": false, "group": "llama323b", "excluded_reason": null, "excluded": false, "fits": true, "speed_eligible": true, "eff_ok": false, "soft_reason": "HWBench Data has not verified tok/s/W rows for ranking yet", "eligible": false, "headroom": null, "unknown_badges": [ "price unknown" ], "group_size": 0, "group_rank": null }, { "platform": "AMD Ryzen AI Max+ 395 / PRO 395 (Strix Halo, Radeon 8060S)", "platform_url": "/platform/amd-ryzen-ai-max-395-pro-395-strix-halo-radeon-8060s.html", "perf": 88.14, "perf_unit": "tok/s", "metric": "generation_throughput", "model": "Llama 3.2:3b", "precision": "as in linked issue (Ollama default tags are 4-bit)", "stack": "Ollama / llama.cpp (see linked issue)", "size_class": "LLM ≤4B", "batch": "1", "like_for_like": true, "differs": [], "confidence": "C", "confidence_label": "C – community/review", "power_w": 133, "power_label": "reported in result source — Measured, Whole system", "power_kind": "Measured, Whole system", "model_key": "llama323b", "pb": { "tag": "Measured", "measured": true, "vendorMeasured": false, "typical": false, "scope": "Whole system", "rmax": false, "rmin": false }, "power_scope": "whole-system", "power_is_range_max": false, "power_is_range_min": false, "is_partial": "false", "partial_note": null, "rank_default": false, "fpsw_outlier": false, "fpsw_note": null, "rank_note": null, "rank_fps": false, "rank_fpsw": false, "fpsw_ratio_median": null, "power_basis": "measured peak system power (Jeff Geerling, as published)", "power_note": "Measured", "power_range": null, "power_src_url": "https://github.com/geerlingguy/ai-benchmarks/issues/21#issuecomment-3164568218", "perf_per_w": 0.6627, "date": "2026-09-22 (page last edited)", "older_than_18_months": false, "superseded": false, "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": true, "date_basis": "page-edited", "form_factor": "independent test system", "vendor": "(independent test)", "source": { "name": "Jeff Geerling ai-benchmarks README (Findings tables)", "url": "https://github.com/geerlingguy/ai-benchmarks/issues/21#issuecomment-3164568218", "date": null, "doc": null, "version": "ai-benchmarks README @ 2f56f19", "vendor_page": null }, "power_source": "https://github.com/geerlingguy/ai-benchmarks/issues/21#issuecomment-3164568218", "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "no_power": false, "group": "llama323b", "excluded_reason": null, "excluded": false, "fits": true, "speed_eligible": true, "eff_ok": false, "soft_reason": "HWBench Data has not verified tok/s/W rows for ranking yet", "eligible": false, "headroom": null, "unknown_badges": [ "price unknown" ], "group_size": 0, "group_rank": null }, { "platform": "Radxa Dragon Q8B (as tested by Jeff Geerling)", "platform_url": "/platform/radxa-dragon-q8b-as-tested-by-jeff-geerling.html", "perf": 12.3, "perf_unit": "tok/s", "metric": "generation_throughput", "model": "Llama 3.2:3b", "precision": "as in linked issue (Ollama default tags are 4-bit)", "stack": "Ollama / llama.cpp (see linked issue)", "size_class": "LLM ≤4B", "batch": "1", "like_for_like": true, "differs": [], "confidence": "C", "confidence_label": "C – community/review", "power_w": 24.3, "power_label": "reported in result source — Measured, Whole system", "power_kind": "Measured, Whole system", "model_key": "llama323b", "pb": { "tag": "Measured", "measured": true, "vendorMeasured": false, "typical": false, "scope": "Whole system", "rmax": false, "rmin": false }, "power_scope": "whole-system", "power_is_range_max": false, "power_is_range_min": false, "is_partial": "false", "partial_note": null, "rank_default": false, "fpsw_outlier": false, "fpsw_note": null, "rank_note": null, "rank_fps": false, "rank_fpsw": false, "fpsw_ratio_median": null, "power_basis": "measured peak system power (Jeff Geerling, as published)", "power_note": "Measured", "power_range": null, "power_src_url": "https://github.com/geerlingguy/ai-benchmarks/issues/54", "perf_per_w": 0.5062, "date": "2026-09-22 (page last edited)", "older_than_18_months": false, "superseded": false, "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": true, "date_basis": "page-edited", "form_factor": "independent test system", "vendor": "(independent test)", "source": { "name": "Jeff Geerling ai-benchmarks README (Findings tables)", "url": "https://github.com/geerlingguy/ai-benchmarks/issues/54", "date": null, "doc": null, "version": "ai-benchmarks README @ 2f56f19", "vendor_page": null }, "power_source": "https://github.com/geerlingguy/ai-benchmarks/issues/54", "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "no_power": false, "group": "llama323b", "excluded_reason": null, "excluded": false, "fits": true, "speed_eligible": true, "eff_ok": false, "soft_reason": "HWBench Data has not verified tok/s/W rows for ranking yet", "eligible": false, "headroom": null, "unknown_badges": [ "price unknown" ], "group_size": 0, "group_rank": null }, { "platform": "Framework 13 (Ryzen AI 5 340) (as tested by Jeff Geerling)", "platform_url": "/platform/framework-13-ryzen-ai-5-340-as-tested-by-jeff-geerling.html", "perf": 23.81, "perf_unit": "tok/s", "metric": "generation_throughput", "model": "Llama 3.2:3b", "precision": "as in linked issue (Ollama default tags are 4-bit)", "stack": "Ollama / llama.cpp (see linked issue)", "size_class": "LLM ≤4B", "batch": "1", "like_for_like": true, "differs": [], "confidence": "C", "confidence_label": "C – community/review", "power_w": 51.1, "power_label": "reported in result source — Measured, Whole system", "power_kind": "Measured, Whole system", "model_key": "llama323b", "pb": { "tag": "Measured", "measured": true, "vendorMeasured": false, "typical": false, "scope": "Whole system", "rmax": false, "rmin": false }, "power_scope": "whole-system", "power_is_range_max": false, "power_is_range_min": false, "is_partial": "false", "partial_note": null, "rank_default": false, "fpsw_outlier": false, "fpsw_note": null, "rank_note": null, "rank_fps": false, "rank_fpsw": false, "fpsw_ratio_median": null, "power_basis": "measured peak system power (Jeff Geerling, as published)", "power_note": "Measured", "power_range": null, "power_src_url": "https://github.com/geerlingguy/ai-benchmarks/issues/33", "perf_per_w": 0.4659, "date": "2026-09-22 (page last edited)", "older_than_18_months": false, "superseded": false, "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": true, "date_basis": "page-edited", "form_factor": "independent test system", "vendor": "(independent test)", "source": { "name": "Jeff Geerling ai-benchmarks README (Findings tables)", "url": "https://github.com/geerlingguy/ai-benchmarks/issues/33", "date": null, "doc": null, "version": "ai-benchmarks README @ 2f56f19", "vendor_page": null }, "power_source": "https://github.com/geerlingguy/ai-benchmarks/issues/33", "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "no_power": false, "group": "llama323b", "excluded_reason": null, "excluded": false, "fits": true, "speed_eligible": true, "eff_ok": false, "soft_reason": "HWBench Data has not verified tok/s/W rows for ranking yet", "eligible": false, "headroom": null, "unknown_badges": [ "price unknown" ], "group_size": 0, "group_rank": null }, { "platform": "MetaComputing AI PC Framework 13 Mainboard (as tested by Jeff Geerling)", "platform_url": "/platform/metacomputing-ai-pc-framework-13-mainboard-as-tested-by-jeff-geerling.html", "perf": 13.1, "perf_unit": "tok/s", "metric": "generation_throughput", "model": "Llama 3.2:3b", "precision": "as in linked issue (Ollama default tags are 4-bit)", "stack": "Ollama / llama.cpp (see linked issue)", "size_class": "LLM ≤4B", "batch": "1", "like_for_like": true, "differs": [], "confidence": "C", "confidence_label": "C – community/review", "power_w": 29, "power_label": "reported in result source — Measured, Whole system", "power_kind": "Measured, Whole system", "model_key": "llama323b", "pb": { "tag": "Measured", "measured": true, "vendorMeasured": false, "typical": false, "scope": "Whole system", "rmax": false, "rmin": false }, "power_scope": "whole-system", "power_is_range_max": false, "power_is_range_min": false, "is_partial": "false", "partial_note": null, "rank_default": false, "fpsw_outlier": false, "fpsw_note": null, "rank_note": null, "rank_fps": false, "rank_fpsw": false, "fpsw_ratio_median": null, "power_basis": "measured peak system power (Jeff Geerling, as published)", "power_note": "Measured", "power_range": null, "power_src_url": "https://github.com/geerlingguy/ai-benchmarks/issues/51", "perf_per_w": 0.4517, "date": "2026-09-22 (page last edited)", "older_than_18_months": false, "superseded": false, "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": true, "date_basis": "page-edited", "form_factor": "independent test system", "vendor": "(independent test)", "source": { "name": "Jeff Geerling ai-benchmarks README (Findings tables)", "url": "https://github.com/geerlingguy/ai-benchmarks/issues/51", "date": null, "doc": null, "version": "ai-benchmarks README @ 2f56f19", "vendor_page": null }, "power_source": "https://github.com/geerlingguy/ai-benchmarks/issues/51", "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "no_power": false, "group": "llama323b", "excluded_reason": null, "excluded": false, "fits": true, "speed_eligible": true, "eff_ok": false, "soft_reason": "HWBench Data has not verified tok/s/W rows for ranking yet", "eligible": false, "headroom": null, "unknown_badges": [ "price unknown" ], "group_size": 0, "group_rank": null }, { "platform": "Radxa Oryon O6 Mini ITX Board (as tested by Jeff Geerling)", "platform_url": "/platform/radxa-oryon-o6-mini-itx-board-as-tested-by-jeff-geerling.html", "perf": 181.06, "perf_unit": "tok/s", "metric": "generation_throughput", "model": "llama3.2:3b", "precision": "Ollama default tag (4-bit unless tag says otherwise)", "stack": "Ollama (geerlingguy/ollama-benchmark)", "size_class": "LLM ≤4B", "batch": "1", "like_for_like": true, "differs": [], "confidence": "C", "confidence_label": "C – community/review", "power_w": 409, "power_label": "reported in result source — Measured, Whole system", "power_kind": "Measured, Whole system", "model_key": "llama323b", "pb": { "tag": "Measured", "measured": true, "vendorMeasured": false, "typical": false, "scope": "Whole system", "rmax": false, "rmin": false }, "power_scope": "whole-system", "power_is_range_max": false, "power_is_range_min": false, "is_partial": "false", "partial_note": null, "rank_default": false, "fpsw_outlier": false, "fpsw_note": null, "rank_note": null, "rank_fps": false, "rank_fpsw": false, "fpsw_ratio_median": null, "power_basis": "measured system power printed with the run (Jeff Geerling)", "power_note": "Measured", "power_range": null, "power_src_url": "https://github.com/geerlingguy/ai-benchmarks/issues/13#issuecomment-2855552157", "perf_per_w": 0.4427, "date": "2025-05-06 (page last edited)", "older_than_18_months": false, "superseded": false, "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": true, "date_basis": "page-edited", "form_factor": "independent test system", "vendor": "(independent test)", "source": { "name": "Jeff Geerling ai-benchmarks issue results (comments)", "url": "https://github.com/geerlingguy/ai-benchmarks/issues/13#issuecomment-2855552157", "date": null, "doc": null, "version": "ai-benchmarks README @ 2f56f19", "vendor_page": null }, "power_source": "https://github.com/geerlingguy/ai-benchmarks/issues/13#issuecomment-2855552157", "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "no_power": false, "group": "llama323b", "excluded_reason": null, "excluded": false, "fits": true, "speed_eligible": true, "eff_ok": false, "soft_reason": "HWBench Data has not verified tok/s/W rows for ranking yet", "eligible": false, "headroom": null, "unknown_badges": [ "price unknown" ], "group_size": 0, "group_rank": null }, { "platform": "Pi 500 - 8GB (as tested by Jeff Geerling)", "platform_url": "/platform/pi-500-8gb-as-tested-by-jeff-geerling.html", "perf": 4.53, "perf_unit": "tok/s", "metric": "generation_throughput", "model": "llama3.2:3b", "precision": "Ollama default tag (4-bit unless tag says otherwise)", "stack": "Ollama (geerlingguy/ollama-benchmark)", "size_class": "LLM ≤4B", "batch": "1", "like_for_like": true, "differs": [], "confidence": "C", "confidence_label": "C – community/review", "power_w": 10.4, "power_label": "reported in result source — Measured, Whole system", "power_kind": "Measured, Whole system", "model_key": "llama323b", "pb": { "tag": "Measured", "measured": true, "vendorMeasured": false, "typical": false, "scope": "Whole system", "rmax": false, "rmin": false }, "power_scope": "whole-system", "power_is_range_max": false, "power_is_range_min": false, "is_partial": "false", "partial_note": null, "rank_default": false, "fpsw_outlier": false, "fpsw_note": null, "rank_note": null, "rank_fps": false, "rank_fpsw": false, "fpsw_ratio_median": null, "power_basis": "measured peak system power (Jeff Geerling)", "power_note": "Measured", "power_range": null, "power_src_url": "https://github.com/geerlingguy/sbc-reviews/issues/60", "perf_per_w": 0.4356, "date": "2024-12-09 (page last edited)", "older_than_18_months": true, "superseded": false, "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": true, "date_basis": "page-edited", "form_factor": "independent test system", "vendor": "(independent test)", "source": { "name": "Jeff Geerling sbc-reviews (AI / LLM inference tables)", "url": "https://github.com/geerlingguy/sbc-reviews/issues/60", "date": null, "doc": null, "version": "ai-benchmarks README @ 2f56f19", "vendor_page": null }, "power_source": "https://github.com/geerlingguy/sbc-reviews/issues/60", "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "no_power": false, "group": "llama323b", "excluded_reason": null, "excluded": false, "fits": true, "speed_eligible": true, "eff_ok": false, "soft_reason": "HWBench Data has not verified tok/s/W rows for ranking yet", "eligible": false, "headroom": null, "unknown_badges": [ "price unknown" ], "group_size": 0, "group_rank": null }, { "platform": "Pi 500+ - 16GB (as tested by Jeff Geerling)", "platform_url": "/platform/pi-500-16gb-as-tested-by-jeff-geerling.html", "perf": 5.55, "perf_unit": "tok/s", "metric": "generation_throughput", "model": "Llama 3.2:3b", "precision": "as in linked issue (Ollama default tags are 4-bit)", "stack": "Ollama / llama.cpp (see linked issue)", "size_class": "LLM ≤4B", "batch": "1", "like_for_like": true, "differs": [], "confidence": "C", "confidence_label": "C – community/review", "power_w": 13, "power_label": "reported in result source — Measured, Whole system", "power_kind": "Measured, Whole system", "model_key": "llama323b", "pb": { "tag": "Measured", "measured": true, "vendorMeasured": false, "typical": false, "scope": "Whole system", "rmax": false, "rmin": false }, "power_scope": "whole-system", "power_is_range_max": false, "power_is_range_min": false, "is_partial": "false", "partial_note": null, "rank_default": false, "fpsw_outlier": false, "fpsw_note": null, "rank_note": null, "rank_fps": false, "rank_fpsw": false, "fpsw_ratio_median": null, "power_basis": "measured peak system power (Jeff Geerling, as published)", "power_note": "Measured", "power_range": null, "power_src_url": "https://github.com/geerlingguy/ai-benchmarks/issues/24", "perf_per_w": 0.4269, "date": "2026-09-22 (page last edited)", "older_than_18_months": false, "superseded": false, "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": true, "date_basis": "page-edited", "form_factor": "independent test system", "vendor": "(independent test)", "source": { "name": "Jeff Geerling ai-benchmarks README (Findings tables)", "url": "https://github.com/geerlingguy/ai-benchmarks/issues/24", "date": null, "doc": null, "version": "ai-benchmarks README @ 2f56f19", "vendor_page": null }, "power_source": "https://github.com/geerlingguy/ai-benchmarks/issues/24", "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "no_power": false, "group": "llama323b", "excluded_reason": null, "excluded": false, "fits": true, "speed_eligible": true, "eff_ok": false, "soft_reason": "HWBench Data has not verified tok/s/W rows for ranking yet", "eligible": false, "headroom": null, "unknown_badges": [ "price unknown" ], "group_size": 0, "group_rank": null }, { "platform": "Raspberry Pi 5 (CPU only)", "platform_url": "/platform/raspberry-pi-5-cpu-only.html", "perf": 4.88, "perf_unit": "tok/s", "metric": "generation_throughput", "model": "Llama 3.2:3b", "precision": "as in linked issue (Ollama default tags are 4-bit)", "stack": "Ollama / llama.cpp (see linked issue)", "size_class": "LLM ≤4B", "batch": "1", "like_for_like": true, "differs": [], "confidence": "C", "confidence_label": "C – community/review", "power_w": 11.9, "power_label": "reported in result source — Measured, Whole system", "power_kind": "Measured, Whole system", "model_key": "llama323b", "pb": { "tag": "Measured", "measured": true, "vendorMeasured": false, "typical": false, "scope": "Whole system", "rmax": false, "rmin": false }, "power_scope": "whole-system", "power_is_range_max": false, "power_is_range_min": false, "is_partial": "false", "partial_note": null, "rank_default": false, "fpsw_outlier": false, "fpsw_note": null, "rank_note": null, "rank_fps": false, "rank_fpsw": false, "fpsw_ratio_median": null, "power_basis": "measured peak system power (Jeff Geerling, as published)", "power_note": "Measured", "power_range": null, "power_src_url": "https://github.com/geerlingguy/ai-benchmarks/issues/70", "perf_per_w": 0.4101, "date": "2026-09-22 (page last edited)", "older_than_18_months": false, "superseded": false, "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": true, "date_basis": "page-edited", "form_factor": "independent test system", "vendor": "(independent test)", "source": { "name": "Jeff Geerling ai-benchmarks README (Findings tables)", "url": "https://github.com/geerlingguy/ai-benchmarks/issues/70", "date": null, "doc": null, "version": "ai-benchmarks README @ 2f56f19", "vendor_page": null }, "power_source": "https://github.com/geerlingguy/ai-benchmarks/issues/70", "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "no_power": false, "group": "llama323b", "excluded_reason": null, "excluded": false, "fits": true, "speed_eligible": true, "eff_ok": false, "soft_reason": "HWBench Data has not verified tok/s/W rows for ranking yet", "eligible": false, "headroom": null, "unknown_badges": [ "price unknown" ], "group_size": 0, "group_rank": null }, { "platform": "Pi 500+ - 16GB (Intel Arc Pro B50) (as tested by Jeff Geerling)", "platform_url": "/platform/pi-500-16gb-intel-arc-pro-b50-as-tested-by-jeff-geerling.html", "perf": 29.8, "perf_unit": "tok/s", "metric": "generation_throughput", "model": "Llama 3.2:3b", "precision": "as in linked issue (Ollama default tags are 4-bit)", "stack": "llama.cpp (Vulkan)", "size_class": "LLM ≤4B", "batch": "1", "like_for_like": true, "differs": [], "confidence": "C", "confidence_label": "C – community/review", "power_w": 78.5, "power_label": "reported in result source — Measured, Whole system", "power_kind": "Measured, Whole system", "model_key": "llama323b", "pb": { "tag": "Measured", "measured": true, "vendorMeasured": false, "typical": false, "scope": "Whole system", "rmax": false, "rmin": false }, "power_scope": "whole-system", "power_is_range_max": false, "power_is_range_min": false, "is_partial": "false", "partial_note": null, "rank_default": false, "fpsw_outlier": false, "fpsw_note": null, "rank_note": null, "rank_fps": false, "rank_fpsw": false, "fpsw_ratio_median": null, "power_basis": "measured peak system power (Jeff Geerling, as published)", "power_note": "Measured", "power_range": null, "power_src_url": "https://github.com/geerlingguy/ai-benchmarks/issues/27", "perf_per_w": 0.3796, "date": "2026-09-22 (page last edited)", "older_than_18_months": false, "superseded": false, "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": true, "date_basis": "page-edited", "form_factor": "independent test system", "vendor": "(independent test)", "source": { "name": "Jeff Geerling ai-benchmarks README (Findings tables)", "url": "https://github.com/geerlingguy/ai-benchmarks/issues/27", "date": null, "doc": null, "version": "ai-benchmarks README @ 2f56f19", "vendor_page": null }, "power_source": "https://github.com/geerlingguy/ai-benchmarks/issues/27", "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "no_power": false, "group": "llama323b", "excluded_reason": null, "excluded": false, "fits": true, "speed_eligible": true, "eff_ok": false, "soft_reason": "HWBench Data has not verified tok/s/W rows for ranking yet", "eligible": false, "headroom": null, "unknown_badges": [ "price unknown" ], "group_size": 0, "group_rank": null }, { "platform": "System76 Thelio Astra (Nvidia A4000) (as tested by Jeff Geerling)", "platform_url": "/platform/system76-thelio-astra-nvidia-a4000-as-tested-by-jeff-geerling.html", "perf": 90.92, "perf_unit": "tok/s", "metric": "generation_throughput", "model": "Llama 3.2:3b", "precision": "as in linked issue (Ollama default tags are 4-bit)", "stack": "Ollama / llama.cpp (see linked issue)", "size_class": "LLM ≤4B", "batch": "1", "like_for_like": true, "differs": [], "confidence": "C", "confidence_label": "C – community/review", "power_w": 244, "power_label": "reported in result source — Measured, Whole system", "power_kind": "Measured, Whole system", "model_key": "llama323b", "pb": { "tag": "Measured", "measured": true, "vendorMeasured": false, "typical": false, "scope": "Whole system", "rmax": false, "rmin": false }, "power_scope": "whole-system", "power_is_range_max": false, "power_is_range_min": false, "is_partial": "false", "partial_note": null, "rank_default": false, "fpsw_outlier": false, "fpsw_note": null, "rank_note": null, "rank_fps": false, "rank_fpsw": false, "fpsw_ratio_median": null, "power_basis": "measured peak system power (Jeff Geerling, as published)", "power_note": "Measured", "power_range": null, "power_src_url": "https://github.com/geerlingguy/ai-benchmarks/issues/5", "perf_per_w": 0.3726, "date": "2026-09-22 (page last edited)", "older_than_18_months": false, "superseded": false, "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": true, "date_basis": "page-edited", "form_factor": "independent test system", "vendor": "(independent test)", "source": { "name": "Jeff Geerling ai-benchmarks README (Findings tables)", "url": "https://github.com/geerlingguy/ai-benchmarks/issues/5", "date": null, "doc": null, "version": "ai-benchmarks README @ 2f56f19", "vendor_page": null }, "power_source": "https://github.com/geerlingguy/ai-benchmarks/issues/5", "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "no_power": false, "group": "llama323b", "excluded_reason": null, "excluded": false, "fits": true, "speed_eligible": true, "eff_ok": false, "soft_reason": "HWBench Data has not verified tok/s/W rows for ranking yet", "eligible": false, "headroom": null, "unknown_badges": [ "price unknown" ], "group_size": 0, "group_rank": null }, { "platform": "GMKtec G3 Plus (Intel N150) - 16GB (as tested by Jeff Geerling)", "platform_url": "/platform/gmktec-g3-plus-intel-n150-16gb-as-tested-by-jeff-geerling.html", "perf": 9.06, "perf_unit": "tok/s", "metric": "generation_throughput", "model": "Llama 3.2:3b", "precision": "as in linked issue (Ollama default tags are 4-bit)", "stack": "Ollama / llama.cpp (see linked issue)", "size_class": "LLM ≤4B", "batch": "1", "like_for_like": true, "differs": [], "confidence": "C", "confidence_label": "C – community/review", "power_w": 26.4, "power_label": "reported in result source — Measured, Whole system", "power_kind": "Measured, Whole system", "model_key": "llama323b", "pb": { "tag": "Measured", "measured": true, "vendorMeasured": false, "typical": false, "scope": "Whole system", "rmax": false, "rmin": false }, "power_scope": "whole-system", "power_is_range_max": false, "power_is_range_min": false, "is_partial": "false", "partial_note": null, "rank_default": false, "fpsw_outlier": false, "fpsw_note": null, "rank_note": null, "rank_fps": false, "rank_fpsw": false, "fpsw_ratio_median": null, "power_basis": "measured peak system power (Jeff Geerling, as published)", "power_note": "Measured", "power_range": null, "power_src_url": "https://github.com/geerlingguy/ai-benchmarks/issues/12", "perf_per_w": 0.3432, "date": "2026-09-22 (page last edited)", "older_than_18_months": false, "superseded": false, "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": true, "date_basis": "page-edited", "form_factor": "independent test system", "vendor": "(independent test)", "source": { "name": "Jeff Geerling ai-benchmarks README (Findings tables)", "url": "https://github.com/geerlingguy/ai-benchmarks/issues/12", "date": null, "doc": null, "version": "ai-benchmarks README @ 2f56f19", "vendor_page": null }, "power_source": "https://github.com/geerlingguy/ai-benchmarks/issues/12", "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "no_power": false, "group": "llama323b", "excluded_reason": null, "excluded": false, "fits": true, "speed_eligible": true, "eff_ok": false, "soft_reason": "HWBench Data has not verified tok/s/W rows for ranking yet", "eligible": false, "headroom": null, "unknown_badges": [ "price unknown" ], "group_size": 0, "group_rank": null }, { "platform": "Intel N150 Mini PC (GMKTek G3 Plus) (as tested by Jeff Geerling)", "platform_url": "/platform/intel-n150-mini-pc-gmktek-g3-plus-as-tested-by-jeff-geerling.html", "perf": 9.06, "perf_unit": "tok/s", "metric": "generation_throughput", "model": "llama3.2:3b", "precision": "Ollama default tag (4-bit unless tag says otherwise)", "stack": "Ollama (geerlingguy/ollama-benchmark)", "size_class": "LLM ≤4B", "batch": "1", "like_for_like": true, "differs": [], "confidence": "C", "confidence_label": "C – community/review", "power_w": 26.4, "power_label": "reported in result source — Measured, Whole system", "power_kind": "Measured, Whole system", "model_key": "llama323b", "pb": { "tag": "Measured", "measured": true, "vendorMeasured": false, "typical": false, "scope": "Whole system", "rmax": false, "rmin": false }, "power_scope": "whole-system", "power_is_range_max": false, "power_is_range_min": false, "is_partial": "false", "partial_note": null, "rank_default": false, "fpsw_outlier": false, "fpsw_note": null, "rank_note": null, "rank_fps": false, "rank_fpsw": false, "fpsw_ratio_median": null, "power_basis": "measured system power printed with the run (Jeff Geerling)", "power_note": "Measured", "power_range": null, "power_src_url": "https://github.com/geerlingguy/ai-benchmarks/issues/12#issuecomment-2634973115", "perf_per_w": 0.3432, "date": "2025-02-04 (page last edited)", "older_than_18_months": true, "superseded": false, "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": true, "date_basis": "page-edited", "form_factor": "independent test system", "vendor": "(independent test)", "source": { "name": "Jeff Geerling ai-benchmarks issue results (comments)", "url": "https://github.com/geerlingguy/ai-benchmarks/issues/12#issuecomment-2634973115", "date": null, "doc": null, "version": "ai-benchmarks README @ 2f56f19", "vendor_page": null }, "power_source": "https://github.com/geerlingguy/ai-benchmarks/issues/12#issuecomment-2634973115", "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "no_power": false, "group": "llama323b", "excluded_reason": null, "excluded": false, "fits": true, "speed_eligible": true, "eff_ok": false, "soft_reason": "HWBench Data has not verified tok/s/W rows for ranking yet", "eligible": false, "headroom": null, "unknown_badges": [ "price unknown" ], "group_size": 0, "group_rank": null }, { "platform": "Pi CM5 - 16GB (AMD AI Pro R9700) (as tested by Jeff Geerling)", "platform_url": "/platform/pi-cm5-16gb-amd-ai-pro-r9700-as-tested-by-jeff-geerling.html", "perf": 77.54, "perf_unit": "tok/s", "metric": "generation_throughput", "model": "Llama 3.2:3b", "precision": "as in linked issue (Ollama default tags are 4-bit)", "stack": "llama.cpp (Vulkan)", "size_class": "LLM ≤4B", "batch": "1", "like_for_like": true, "differs": [], "confidence": "C", "confidence_label": "C – community/review", "power_w": 281, "power_label": "reported in result source — Measured, Whole system", "power_kind": "Measured, Whole system", "model_key": "llama323b", "pb": { "tag": "Measured", "measured": true, "vendorMeasured": false, "typical": false, "scope": "Whole system", "rmax": false, "rmin": false }, "power_scope": "whole-system", "power_is_range_max": false, "power_is_range_min": false, "is_partial": "false", "partial_note": null, "rank_default": false, "fpsw_outlier": false, "fpsw_note": null, "rank_note": null, "rank_fps": false, "rank_fpsw": false, "fpsw_ratio_median": null, "power_basis": "measured peak system power (Jeff Geerling, as published)", "power_note": "Measured", "power_range": null, "power_src_url": "https://github.com/geerlingguy/ai-benchmarks/issues/35", "perf_per_w": 0.2759, "date": "2026-09-22 (page last edited)", "older_than_18_months": false, "superseded": false, "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": true, "date_basis": "page-edited", "form_factor": "independent test system", "vendor": "(independent test)", "source": { "name": "Jeff Geerling ai-benchmarks README (Findings tables)", "url": "https://github.com/geerlingguy/ai-benchmarks/issues/35", "date": null, "doc": null, "version": "ai-benchmarks README @ 2f56f19", "vendor_page": null }, "power_source": "https://github.com/geerlingguy/ai-benchmarks/issues/35", "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "no_power": false, "group": "llama323b", "excluded_reason": null, "excluded": false, "fits": true, "speed_eligible": true, "eff_ok": false, "soft_reason": "HWBench Data has not verified tok/s/W rows for ranking yet", "eligible": false, "headroom": null, "unknown_badges": [ "price unknown" ], "group_size": 0, "group_rank": null }, { "platform": "Pi 400 - 4GB (as tested by Jeff Geerling)", "platform_url": "/platform/pi-400-4gb-as-tested-by-jeff-geerling.html", "perf": 1.6, "perf_unit": "tok/s", "metric": "generation_throughput", "model": "Llama 3.2:3b", "precision": "as in linked issue (Ollama default tags are 4-bit)", "stack": "Ollama / llama.cpp (see linked issue)", "size_class": "LLM ≤4B", "batch": "1", "like_for_like": true, "differs": [], "confidence": "C", "confidence_label": "C – community/review", "power_w": 6, "power_label": "reported in result source — Measured, Whole system", "power_kind": "Measured, Whole system", "model_key": "llama323b", "pb": { "tag": "Measured", "measured": true, "vendorMeasured": false, "typical": false, "scope": "Whole system", "rmax": false, "rmin": false }, "power_scope": "whole-system", "power_is_range_max": false, "power_is_range_min": false, "is_partial": "false", "partial_note": null, "rank_default": false, "fpsw_outlier": false, "fpsw_note": null, "rank_note": null, "rank_fps": false, "rank_fpsw": false, "fpsw_ratio_median": null, "power_basis": "measured peak system power (Jeff Geerling, as published)", "power_note": "Measured", "power_range": null, "power_src_url": "https://github.com/geerlingguy/ai-benchmarks/commit/96bab78f2a8e6c996c6810c5e2119274e3eb401a", "perf_per_w": 0.2667, "date": "2026-09-22 (page last edited)", "older_than_18_months": false, "superseded": false, "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": true, "date_basis": "page-edited", "form_factor": "independent test system", "vendor": "(independent test)", "source": { "name": "Jeff Geerling ai-benchmarks README (Findings tables)", "url": "https://github.com/geerlingguy/ai-benchmarks/commit/96bab78f2a8e6c996c6810c5e2119274e3eb401a", "date": null, "doc": null, "version": "ai-benchmarks README @ 2f56f19", "vendor_page": null }, "power_source": "https://github.com/geerlingguy/ai-benchmarks/commit/96bab78f2a8e6c996c6810c5e2119274e3eb401a", "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "no_power": false, "group": "llama323b", "excluded_reason": null, "excluded": false, "fits": true, "speed_eligible": true, "eff_ok": false, "soft_reason": "HWBench Data has not verified tok/s/W rows for ranking yet", "eligible": false, "headroom": null, "unknown_badges": [ "price unknown" ], "group_size": 0, "group_rank": null }, { "platform": "Raspberry Pi 400 - 4GB (as tested by Jeff Geerling)", "platform_url": "/platform/raspberry-pi-400-4gb-as-tested-by-jeff-geerling.html", "perf": 1.6, "perf_unit": "tok/s", "metric": "generation_throughput", "model": "llama3.2:3b", "precision": "Ollama default tag (4-bit unless tag says otherwise)", "stack": "Ollama (geerlingguy/ollama-benchmark)", "size_class": "LLM ≤4B", "batch": "1", "like_for_like": true, "differs": [], "confidence": "C", "confidence_label": "C – community/review", "power_w": 6, "power_label": "reported in result source — Measured, Whole system", "power_kind": "Measured, Whole system", "model_key": "llama323b", "pb": { "tag": "Measured", "measured": true, "vendorMeasured": false, "typical": false, "scope": "Whole system", "rmax": false, "rmin": false }, "power_scope": "whole-system", "power_is_range_max": false, "power_is_range_min": false, "is_partial": "false", "partial_note": null, "rank_default": false, "fpsw_outlier": false, "fpsw_note": null, "rank_note": null, "rank_fps": false, "rank_fpsw": false, "fpsw_ratio_median": null, "power_basis": "measured peak system power (Jeff Geerling)", "power_note": "Measured", "power_range": null, "power_src_url": "https://github.com/geerlingguy/sbc-reviews/issues/59", "perf_per_w": 0.2667, "date": "2024-11-30 (page last edited)", "older_than_18_months": true, "superseded": false, "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": true, "date_basis": "page-edited", "form_factor": "independent test system", "vendor": "(independent test)", "source": { "name": "Jeff Geerling sbc-reviews (AI / LLM inference tables)", "url": "https://github.com/geerlingguy/sbc-reviews/issues/59", "date": null, "doc": null, "version": "ai-benchmarks README @ 2f56f19", "vendor_page": null }, "power_source": "https://github.com/geerlingguy/sbc-reviews/issues/59", "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "no_power": false, "group": "llama323b", "excluded_reason": null, "excluded": false, "fits": true, "speed_eligible": true, "eff_ok": false, "soft_reason": "HWBench Data has not verified tok/s/W rows for ranking yet", "eligible": false, "headroom": null, "unknown_badges": [ "price unknown" ], "group_size": 0, "group_rank": null }, { "platform": "System76 Thelio Astra (Nvidia A400) (as tested by Jeff Geerling)", "platform_url": "/platform/system76-thelio-astra-nvidia-a400-as-tested-by-jeff-geerling.html", "perf": 35.51, "perf_unit": "tok/s", "metric": "generation_throughput", "model": "Llama 3.2:3b", "precision": "as in linked issue (Ollama default tags are 4-bit)", "stack": "Ollama / llama.cpp (see linked issue)", "size_class": "LLM ≤4B", "batch": "1", "like_for_like": true, "differs": [], "confidence": "C", "confidence_label": "C – community/review", "power_w": 167, "power_label": "reported in result source — Measured, Whole system", "power_kind": "Measured, Whole system", "model_key": "llama323b", "pb": { "tag": "Measured", "measured": true, "vendorMeasured": false, "typical": false, "scope": "Whole system", "rmax": false, "rmin": false }, "power_scope": "whole-system", "power_is_range_max": false, "power_is_range_min": false, "is_partial": "false", "partial_note": null, "rank_default": false, "fpsw_outlier": false, "fpsw_note": null, "rank_note": null, "rank_fps": false, "rank_fpsw": false, "fpsw_ratio_median": null, "power_basis": "measured peak system power (Jeff Geerling, as published)", "power_note": "Measured", "power_range": null, "power_src_url": "https://github.com/geerlingguy/ai-benchmarks/issues/5", "perf_per_w": 0.2126, "date": "2026-09-22 (page last edited)", "older_than_18_months": false, "superseded": false, "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": true, "date_basis": "page-edited", "form_factor": "independent test system", "vendor": "(independent test)", "source": { "name": "Jeff Geerling ai-benchmarks README (Findings tables)", "url": "https://github.com/geerlingguy/ai-benchmarks/issues/5", "date": null, "doc": null, "version": "ai-benchmarks README @ 2f56f19", "vendor_page": null }, "power_source": "https://github.com/geerlingguy/ai-benchmarks/issues/5", "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "no_power": false, "group": "llama323b", "excluded_reason": null, "excluded": false, "fits": true, "speed_eligible": true, "eff_ok": false, "soft_reason": "HWBench Data has not verified tok/s/W rows for ranking yet", "eligible": false, "headroom": null, "unknown_badges": [ "price unknown" ], "group_size": 0, "group_rank": null }, { "platform": "Nvidia A400 (as tested by Jeff Geerling)", "platform_url": "/platform/nvidia-a400-as-tested-by-jeff-geerling.html", "perf": 35.51, "perf_unit": "tok/s", "metric": "generation_throughput", "model": "llama3.2:3b", "precision": "Ollama default tag (4-bit unless tag says otherwise)", "stack": "Ollama (geerlingguy/ollama-benchmark)", "size_class": "LLM ≤4B", "batch": "1", "like_for_like": true, "differs": [], "confidence": "C", "confidence_label": "C – community/review", "power_w": 167, "power_label": "reported in result source — Measured, Whole system", "power_kind": "Measured, Whole system", "model_key": "llama323b", "pb": { "tag": "Measured", "measured": true, "vendorMeasured": false, "typical": false, "scope": "Whole system", "rmax": false, "rmin": false }, "power_scope": "whole-system", "power_is_range_max": false, "power_is_range_min": false, "is_partial": "false", "partial_note": null, "rank_default": false, "fpsw_outlier": false, "fpsw_note": null, "rank_note": null, "rank_fps": false, "rank_fpsw": false, "fpsw_ratio_median": null, "power_basis": "measured peak system power (Jeff Geerling)", "power_note": "Measured", "power_range": null, "power_src_url": "https://github.com/geerlingguy/sbc-reviews/issues/53", "perf_per_w": 0.2126, "date": "2024-10-22 (page last edited)", "older_than_18_months": true, "superseded": false, "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": true, "date_basis": "page-edited", "form_factor": "independent test system", "vendor": "(independent test)", "source": { "name": "Jeff Geerling sbc-reviews (AI / LLM inference tables)", "url": "https://github.com/geerlingguy/sbc-reviews/issues/53", "date": null, "doc": null, "version": "ai-benchmarks README @ 2f56f19", "vendor_page": null }, "power_source": "https://github.com/geerlingguy/sbc-reviews/issues/53", "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "no_power": false, "group": "llama323b", "excluded_reason": null, "excluded": false, "fits": true, "speed_eligible": true, "eff_ok": false, "soft_reason": "HWBench Data has not verified tok/s/W rows for ranking yet", "eligible": false, "headroom": null, "unknown_badges": [ "price unknown" ], "group_size": 0, "group_rank": null }, { "platform": "HiFive Premier P550 (4-core RISC-V) (as tested by Jeff Geerling)", "platform_url": "/platform/hifive-premier-p550-4-core-risc-v-as-tested-by-jeff-geerling.html", "perf": 0.24, "perf_unit": "tok/s", "metric": "generation_throughput", "model": "Llama 3.2:3b", "precision": "as in linked issue (Ollama default tags are 4-bit)", "stack": "Ollama / llama.cpp (see linked issue)", "size_class": "LLM ≤4B", "batch": "1", "like_for_like": true, "differs": [], "confidence": "C", "confidence_label": "C – community/review", "power_w": 13.5, "power_label": "reported in result source — Measured, Whole system", "power_kind": "Measured, Whole system", "model_key": "llama323b", "pb": { "tag": "Measured", "measured": true, "vendorMeasured": false, "typical": false, "scope": "Whole system", "rmax": false, "rmin": false }, "power_scope": "whole-system", "power_is_range_max": false, "power_is_range_min": false, "is_partial": "false", "partial_note": null, "rank_default": false, "fpsw_outlier": false, "fpsw_note": null, "rank_note": null, "rank_fps": false, "rank_fpsw": false, "fpsw_ratio_median": null, "power_basis": "measured peak system power (Jeff Geerling, as published)", "power_note": "Measured", "power_range": null, "power_src_url": "https://github.com/geerlingguy/ai-benchmarks/issues/17", "perf_per_w": 0.01778, "date": "2026-09-22 (page last edited)", "older_than_18_months": false, "superseded": false, "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": true, "date_basis": "page-edited", "form_factor": "independent test system", "vendor": "(independent test)", "source": { "name": "Jeff Geerling ai-benchmarks README (Findings tables)", "url": "https://github.com/geerlingguy/ai-benchmarks/issues/17", "date": null, "doc": null, "version": "ai-benchmarks README @ 2f56f19", "vendor_page": null }, "power_source": "https://github.com/geerlingguy/ai-benchmarks/issues/17", "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "no_power": false, "group": "llama323b", "excluded_reason": null, "excluded": false, "fits": true, "speed_eligible": true, "eff_ok": false, "soft_reason": "HWBench Data has not verified tok/s/W rows for ranking yet", "eligible": false, "headroom": null, "unknown_badges": [ "price unknown" ], "group_size": 0, "group_rank": null }, { "platform": "HiFive Premier P550 (as tested by Jeff Geerling)", "platform_url": "/platform/hifive-premier-p550-as-tested-by-jeff-geerling.html", "perf": 0.24, "perf_unit": "tok/s", "metric": "generation_throughput", "model": "llama3.2:3b", "precision": "Ollama default tag (4-bit unless tag says otherwise)", "stack": "Ollama (geerlingguy/ollama-benchmark)", "size_class": "LLM ≤4B", "batch": "1", "like_for_like": true, "differs": [], "confidence": "C", "confidence_label": "C – community/review", "power_w": 13.5, "power_label": "reported in result source — Measured, Whole system", "power_kind": "Measured, Whole system", "model_key": "llama323b", "pb": { "tag": "Measured", "measured": true, "vendorMeasured": false, "typical": false, "scope": "Whole system", "rmax": false, "rmin": false }, "power_scope": "whole-system", "power_is_range_max": false, "power_is_range_min": false, "is_partial": "false", "partial_note": null, "rank_default": false, "fpsw_outlier": false, "fpsw_note": null, "rank_note": null, "rank_fps": false, "rank_fpsw": false, "fpsw_ratio_median": null, "power_basis": "measured system power printed with the run (Jeff Geerling)", "power_note": "Measured", "power_range": null, "power_src_url": "https://github.com/geerlingguy/ai-benchmarks/issues/17#issuecomment-2640127288", "perf_per_w": 0.01778, "date": "2025-02-06 (page last edited)", "older_than_18_months": true, "superseded": false, "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": true, "date_basis": "page-edited", "form_factor": "independent test system", "vendor": "(independent test)", "source": { "name": "Jeff Geerling ai-benchmarks issue results (comments)", "url": "https://github.com/geerlingguy/ai-benchmarks/issues/17#issuecomment-2640127288", "date": null, "doc": null, "version": "ai-benchmarks README @ 2f56f19", "vendor_page": null }, "power_source": "https://github.com/geerlingguy/ai-benchmarks/issues/17#issuecomment-2640127288", "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "no_power": false, "group": "llama323b", "excluded_reason": null, "excluded": false, "fits": true, "speed_eligible": true, "eff_ok": false, "soft_reason": "HWBench Data has not verified tok/s/W rows for ranking yet", "eligible": false, "headroom": null, "unknown_badges": [ "price unknown" ], "group_size": 0, "group_rank": null }, { "platform": "DC-ROMA Mainboard II (8-core RISC-V) (as tested by Jeff Geerling)", "platform_url": "/platform/dc-roma-mainboard-ii-8-core-risc-v-as-tested-by-jeff-geerling.html", "perf": 0.31, "perf_unit": "tok/s", "metric": "generation_throughput", "model": "Llama 3.2:3b", "precision": "as in linked issue (Ollama default tags are 4-bit)", "stack": "Ollama / llama.cpp (see linked issue)", "size_class": "LLM ≤4B", "batch": "1", "like_for_like": true, "differs": [], "confidence": "C", "confidence_label": "C – community/review", "power_w": 30.6, "power_label": "reported in result source — Measured, Whole system", "power_kind": "Measured, Whole system", "model_key": "llama323b", "pb": { "tag": "Measured", "measured": true, "vendorMeasured": false, "typical": false, "scope": "Whole system", "rmax": false, "rmin": false }, "power_scope": "whole-system", "power_is_range_max": false, "power_is_range_min": false, "is_partial": "false", "partial_note": null, "rank_default": false, "fpsw_outlier": false, "fpsw_note": null, "rank_note": null, "rank_fps": false, "rank_fpsw": false, "fpsw_ratio_median": null, "power_basis": "measured peak system power (Jeff Geerling, as published)", "power_note": "Measured", "power_range": null, "power_src_url": "https://github.com/geerlingguy/ai-benchmarks/issues/28", "perf_per_w": 0.01013, "date": "2026-09-22 (page last edited)", "older_than_18_months": false, "superseded": false, "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": true, "date_basis": "page-edited", "form_factor": "independent test system", "vendor": "(independent test)", "source": { "name": "Jeff Geerling ai-benchmarks README (Findings tables)", "url": "https://github.com/geerlingguy/ai-benchmarks/issues/28", "date": null, "doc": null, "version": "ai-benchmarks README @ 2f56f19", "vendor_page": null }, "power_source": "https://github.com/geerlingguy/ai-benchmarks/issues/28", "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "no_power": false, "group": "llama323b", "excluded_reason": null, "excluded": false, "fits": true, "speed_eligible": true, "eff_ok": false, "soft_reason": "HWBench Data has not verified tok/s/W rows for ranking yet", "eligible": false, "headroom": null, "unknown_badges": [ "price unknown" ], "group_size": 0, "group_rank": null }, { "platform": "Dell Optiplex 780 (C2Q Q8400) (as tested by Jeff Geerling)", "platform_url": "/platform/dell-optiplex-780-c2q-q8400-as-tested-by-jeff-geerling.html", "perf": 1.09, "perf_unit": "tok/s", "metric": "generation_throughput", "model": "Llama 3.2:3b", "precision": "as in linked issue (Ollama default tags are 4-bit)", "stack": "Ollama / llama.cpp (see linked issue)", "size_class": "LLM ≤4B", "batch": "1", "like_for_like": true, "differs": [], "confidence": "C", "confidence_label": "C – community/review", "power_w": 146, "power_label": "reported in result source — Measured, Whole system", "power_kind": "Measured, Whole system", "model_key": "llama323b", "pb": { "tag": "Measured", "measured": true, "vendorMeasured": false, "typical": false, "scope": "Whole system", "rmax": false, "rmin": false }, "power_scope": "whole-system", "power_is_range_max": false, "power_is_range_min": false, "is_partial": "false", "partial_note": null, "rank_default": false, "fpsw_outlier": false, "fpsw_note": null, "rank_note": null, "rank_fps": false, "rank_fpsw": false, "fpsw_ratio_median": null, "power_basis": "measured peak system power (Jeff Geerling, as published)", "power_note": "Measured", "power_range": null, "power_src_url": "https://github.com/geerlingguy/ai-benchmarks/issues/31", "perf_per_w": 0.007466, "date": "2026-09-22 (page last edited)", "older_than_18_months": false, "superseded": false, "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": true, "date_basis": "page-edited", "form_factor": "independent test system", "vendor": "(independent test)", "source": { "name": "Jeff Geerling ai-benchmarks README (Findings tables)", "url": "https://github.com/geerlingguy/ai-benchmarks/issues/31", "date": null, "doc": null, "version": "ai-benchmarks README @ 2f56f19", "vendor_page": null }, "power_source": "https://github.com/geerlingguy/ai-benchmarks/issues/31", "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "no_power": false, "group": "llama323b", "excluded_reason": null, "excluded": false, "fits": true, "speed_eligible": true, "eff_ok": false, "soft_reason": "HWBench Data has not verified tok/s/W rows for ranking yet", "eligible": false, "headroom": null, "unknown_badges": [ "price unknown" ], "group_size": 0, "group_rank": null }, { "platform": "Ryzen 9 9950X (AMD 7900 XT) (as tested by Jeff Geerling)", "platform_url": "/platform/ryzen-9-9950x-amd-7900-xt-as-tested-by-jeff-geerling.html", "perf": 131.2, "perf_unit": "tok/s", "metric": "generation_throughput", "model": "Llama 3.2:3b", "precision": "as in linked issue (Ollama default tags are 4-bit)", "stack": "Ollama / llama.cpp (see linked issue)", "size_class": "LLM ≤4B", "batch": "1", "like_for_like": true, "differs": [], "confidence": "D", "confidence_label": "D – vendor-published", "power_w": null, "power_label": "not published", "power_kind": "unknown", "model_key": "llama323b", "pb": { "tag": "", "measured": false, "vendorMeasured": false, "typical": false, "scope": "", "rmax": false, "rmin": false }, "power_scope": null, "power_is_range_max": false, "power_is_range_min": false, "is_partial": "false", "partial_note": null, "rank_default": false, "fpsw_outlier": false, "fpsw_note": null, "rank_note": null, "rank_fps": false, "rank_fpsw": false, "fpsw_ratio_median": null, "power_basis": "unknown", "power_note": "", "power_range": null, "power_src_url": null, "perf_per_w": null, "date": "2026-09-22 (page last edited)", "older_than_18_months": false, "superseded": false, "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": true, "date_basis": "page-edited", "form_factor": "independent test system", "vendor": "(independent test)", "source": { "name": "Jeff Geerling ai-benchmarks README (Findings tables)", "url": "https://github.com/geerlingguy/ai-benchmarks/pull/11", "date": null, "doc": null, "version": "ai-benchmarks README @ 2f56f19", "vendor_page": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "no_power": true, "group": "llama323b", "excluded_reason": null, "excluded": false, "fits": true, "speed_eligible": true, "eff_ok": false, "soft_reason": null, "eligible": false, "headroom": null, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "group_size": 0, "group_rank": null }, { "platform": "M1 Ultra (64 GPU Core) 128GB (as tested by Jeff Geerling)", "platform_url": "/platform/m1-ultra-64-gpu-core-128gb-as-tested-by-jeff-geerling.html", "perf": 127.6, "perf_unit": "tok/s", "metric": "generation_throughput", "model": "Llama 3.2:3b", "precision": "as in linked issue (Ollama default tags are 4-bit)", "stack": "Ollama / llama.cpp (see linked issue)", "size_class": "LLM ≤4B", "batch": "1", "like_for_like": true, "differs": [], "confidence": "C", "confidence_label": "C – community/review", "power_w": null, "power_label": "not published", "power_kind": "unknown", "model_key": "llama323b", "pb": { "tag": "", "measured": false, "vendorMeasured": false, "typical": false, "scope": "", "rmax": false, "rmin": false }, "power_scope": null, "power_is_range_max": false, "power_is_range_min": false, "is_partial": "false", "partial_note": null, "rank_default": false, "fpsw_outlier": false, "fpsw_note": null, "rank_note": null, "rank_fps": false, "rank_fpsw": false, "fpsw_ratio_median": null, "power_basis": "unknown", "power_note": "", "power_range": null, "power_src_url": null, "perf_per_w": null, "date": "2026-09-22 (page last edited)", "older_than_18_months": false, "superseded": false, "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": true, "date_basis": "page-edited", "form_factor": "independent test system", "vendor": "(independent test)", "source": { "name": "Jeff Geerling ai-benchmarks README (Findings tables)", "url": "https://github.com/geerlingguy/ai-benchmarks/issues/48", "date": null, "doc": null, "version": "ai-benchmarks README @ 2f56f19", "vendor_page": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "no_power": true, "group": "llama323b", "excluded_reason": null, "excluded": false, "fits": true, "speed_eligible": true, "eff_ok": false, "soft_reason": null, "eligible": false, "headroom": null, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "group_size": 0, "group_rank": null }, { "platform": "M1 Ultra (48 GPU Core) 64GB (as tested by Jeff Geerling)", "platform_url": "/platform/m1-ultra-48-gpu-core-64gb-as-tested-by-jeff-geerling.html", "perf": 108.67, "perf_unit": "tok/s", "metric": "generation_throughput", "model": "Llama 3.2:3b", "precision": "as in linked issue (Ollama default tags are 4-bit)", "stack": "Ollama / llama.cpp (see linked issue)", "size_class": "LLM ≤4B", "batch": "1", "like_for_like": true, "differs": [], "confidence": "D", "confidence_label": "D – vendor-published", "power_w": null, "power_label": "not published", "power_kind": "unknown", "model_key": "llama323b", "pb": { "tag": "", "measured": false, "vendorMeasured": false, "typical": false, "scope": "", "rmax": false, "rmin": false }, "power_scope": null, "power_is_range_max": false, "power_is_range_min": false, "is_partial": "false", "partial_note": null, "rank_default": false, "fpsw_outlier": false, "fpsw_note": null, "rank_note": null, "rank_fps": false, "rank_fpsw": false, "fpsw_ratio_median": null, "power_basis": "unknown", "power_note": "", "power_range": null, "power_src_url": null, "perf_per_w": null, "date": "2026-09-22 (page last edited)", "older_than_18_months": false, "superseded": false, "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": true, "date_basis": "page-edited", "form_factor": "independent test system", "vendor": "(independent test)", "source": { "name": "Jeff Geerling ai-benchmarks README (Findings tables)", "url": "https://github.com/geerlingguy/ai-benchmarks/pull/11", "date": null, "doc": null, "version": "ai-benchmarks README @ 2f56f19", "vendor_page": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "no_power": true, "group": "llama323b", "excluded_reason": null, "excluded": false, "fits": true, "speed_eligible": true, "eff_ok": false, "soft_reason": null, "eligible": false, "headroom": null, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "group_size": 0, "group_rank": null }, { "platform": "Minisforum MS-R1 (as tested by Jeff Geerling)", "platform_url": "/platform/minisforum-ms-r1-as-tested-by-jeff-geerling.html", "perf": 71.36, "perf_unit": "tok/s", "metric": "generation_throughput", "model": "llama3.2:3b", "precision": "Ollama default tag (4-bit unless tag says otherwise)", "stack": "Ollama (geerlingguy/ollama-benchmark)", "size_class": "LLM ≤4B", "batch": "1", "like_for_like": true, "differs": [], "confidence": "C", "confidence_label": "C – community/review", "power_w": null, "power_label": "not published", "power_kind": "unknown", "model_key": "llama323b", "pb": { "tag": "", "measured": false, "vendorMeasured": false, "typical": false, "scope": "", "rmax": false, "rmin": false }, "power_scope": null, "power_is_range_max": false, "power_is_range_min": false, "is_partial": "false", "partial_note": null, "rank_default": false, "fpsw_outlier": false, "fpsw_note": null, "rank_note": null, "rank_fps": false, "rank_fpsw": false, "fpsw_ratio_median": null, "power_basis": "unknown", "power_note": "", "power_range": null, "power_src_url": null, "perf_per_w": null, "date": "2025-11-10 (page last edited)", "older_than_18_months": false, "superseded": false, "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": true, "date_basis": "page-edited", "form_factor": "independent test system", "vendor": "(independent test)", "source": { "name": "Jeff Geerling ai-benchmarks issue results (comments)", "url": "https://github.com/geerlingguy/ai-benchmarks/issues/32#issuecomment-3512391773", "date": null, "doc": null, "version": "ai-benchmarks README @ 2f56f19", "vendor_page": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "no_power": true, "group": "llama323b", "excluded_reason": null, "excluded": false, "fits": true, "speed_eligible": true, "eff_ok": false, "soft_reason": null, "eligible": false, "headroom": null, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "group_size": 0, "group_rank": null }, { "platform": "M1 Max Mac Studio (10 core - 64GB) (as tested by Jeff Geerling)", "platform_url": "/platform/m1-max-mac-studio-10-core-64gb-as-tested-by-jeff-geerling.html", "perf": 59.38, "perf_unit": "tok/s", "metric": "generation_throughput", "model": "Llama 3.2:3b", "precision": "as in linked issue (Ollama default tags are 4-bit)", "stack": "Ollama / llama.cpp (see linked issue)", "size_class": "LLM ≤4B", "batch": "1", "like_for_like": true, "differs": [], "confidence": "D", "confidence_label": "D – vendor-published", "power_w": null, "power_label": "not published", "power_kind": "unknown", "model_key": "llama323b", "pb": { "tag": "", "measured": false, "vendorMeasured": false, "typical": false, "scope": "", "rmax": false, "rmin": false }, "power_scope": null, "power_is_range_max": false, "power_is_range_min": false, "is_partial": "false", "partial_note": null, "rank_default": false, "fpsw_outlier": false, "fpsw_note": null, "rank_note": null, "rank_fps": false, "rank_fpsw": false, "fpsw_ratio_median": null, "power_basis": "unknown", "power_note": "", "power_range": null, "power_src_url": null, "perf_per_w": null, "date": "2026-09-22 (page last edited)", "older_than_18_months": false, "superseded": false, "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": true, "date_basis": "page-edited", "form_factor": "independent test system", "vendor": "(independent test)", "source": { "name": "Jeff Geerling ai-benchmarks README (Findings tables)", "url": "https://github.com/geerlingguy/ai-benchmarks#findings", "date": null, "doc": null, "version": "ai-benchmarks README @ 2f56f19", "vendor_page": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "no_power": true, "group": "llama323b", "excluded_reason": null, "excluded": false, "fits": true, "speed_eligible": true, "eff_ok": false, "soft_reason": null, "eligible": false, "headroom": null, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "group_size": 0, "group_rank": null }, { "platform": "Axelera Metis Edge 130p PCIe card (16GB)", "platform_url": "/platform/axelera-metis-edge-130p-pcie-card-16gb.html", "perf": 5.5, "perf_unit": "tok/s", "metric": "generation_throughput", "model": "llama3.2-3b", "precision": "not stated", "stack": "Voyager SDK model_axscript.llm", "size_class": "LLM ≤4B", "batch": "1", "like_for_like": true, "differs": [], "confidence": "C", "confidence_label": "C – community/review", "power_w": null, "power_label": "not published", "power_kind": "unknown", "model_key": "llama323b", "pb": { "tag": "", "measured": false, "vendorMeasured": false, "typical": false, "scope": "", "rmax": false, "rmin": false }, "power_scope": null, "power_is_range_max": false, "power_is_range_min": false, "is_partial": "false", "partial_note": null, "rank_default": false, "fpsw_outlier": false, "fpsw_note": null, "rank_note": null, "rank_fps": false, "rank_fpsw": false, "fpsw_ratio_median": null, "power_basis": "unknown", "power_note": "", "power_range": null, "power_src_url": null, "perf_per_w": null, "date": "2026-10-15 (page last edited)", "older_than_18_months": false, "superseded": false, "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": true, "date_basis": "page-edited", "form_factor": "edge accelerator card", "vendor": "Axelera AI", "source": { "name": "Axelera Voyager SDK model zoo (docs/reference/models/model-zoo.md)", "url": "https://github.com/axelera-ai-hub/voyager-sdk/blob/latest/docs/reference/models/model-zoo.md", "date": null, "doc": null, "version": "Voyager SDK 1.9.0 (model-zoo.md @ f184b09)", "vendor_page": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "no_power": true, "group": "llama323b", "excluded_reason": null, "excluded": false, "fits": true, "speed_eligible": true, "eff_ok": false, "soft_reason": null, "eligible": false, "headroom": null, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "group_size": 0, "group_rank": null }, { "platform": "Raspberry Pi CM5 - 4GB (as tested by Jeff Geerling)", "platform_url": "/platform/raspberry-pi-cm5-4gb-as-tested-by-jeff-geerling.html", "perf": 4.58, "perf_unit": "tok/s", "metric": "generation_throughput", "model": "llama3.2:3b", "precision": "Ollama default tag (4-bit unless tag says otherwise)", "stack": "Ollama (geerlingguy/ollama-benchmark)", "size_class": "LLM ≤4B", "batch": "1", "like_for_like": true, "differs": [], "confidence": "C", "confidence_label": "C – community/review", "power_w": null, "power_label": "not published", "power_kind": "unknown", "model_key": "llama323b", "pb": { "tag": "", "measured": false, "vendorMeasured": false, "typical": false, "scope": "", "rmax": false, "rmin": false }, "power_scope": null, "power_is_range_max": false, "power_is_range_min": false, "is_partial": "false", "partial_note": null, "rank_default": false, "fpsw_outlier": false, "fpsw_note": null, "rank_note": null, "rank_fps": false, "rank_fpsw": false, "fpsw_ratio_median": null, "power_basis": "unknown", "power_note": "", "power_range": null, "power_src_url": null, "perf_per_w": null, "date": "2024-11-27 (page last edited)", "older_than_18_months": true, "superseded": false, "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": true, "date_basis": "page-edited", "form_factor": "independent test system", "vendor": "(independent test)", "source": { "name": "Jeff Geerling sbc-reviews (AI / LLM inference tables)", "url": "https://github.com/geerlingguy/sbc-reviews/issues/58", "date": null, "doc": null, "version": "ai-benchmarks README @ 2f56f19", "vendor_page": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "no_power": true, "group": "llama323b", "excluded_reason": null, "excluded": false, "fits": true, "speed_eligible": true, "eff_ok": false, "soft_reason": null, "eligible": false, "headroom": null, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "group_size": 0, "group_rank": null }, { "platform": "Dell Optiplex 780 (Intel Core 2 Quad Q8400) (as tested by Jeff Geerling)", "platform_url": "/platform/dell-optiplex-780-intel-core-2-quad-q8400-as-tested-by-jeff-geerling.html", "perf": 1.09, "perf_unit": "tok/s", "metric": "generation_throughput", "model": "llama3.2:3b", "precision": "Ollama default tag (4-bit unless tag says otherwise)", "stack": "Ollama (geerlingguy/ollama-benchmark)", "size_class": "LLM ≤4B", "batch": "1", "like_for_like": true, "differs": [], "confidence": "C", "confidence_label": "C – community/review", "power_w": null, "power_label": "not published", "power_kind": "unknown", "model_key": "llama323b", "pb": { "tag": "", "measured": false, "vendorMeasured": false, "typical": false, "scope": "", "rmax": false, "rmin": false }, "power_scope": null, "power_is_range_max": false, "power_is_range_min": false, "is_partial": "false", "partial_note": null, "rank_default": false, "fpsw_outlier": false, "fpsw_note": null, "rank_note": null, "rank_fps": false, "rank_fpsw": false, "fpsw_ratio_median": null, "power_basis": "unknown", "power_note": "", "power_range": null, "power_src_url": null, "perf_per_w": null, "date": "2025-11-05 (page last edited)", "older_than_18_months": false, "superseded": false, "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": true, "date_basis": "page-edited", "form_factor": "independent test system", "vendor": "(independent test)", "source": { "name": "Jeff Geerling ai-benchmarks issue results (comments)", "url": "https://github.com/geerlingguy/ai-benchmarks/issues/31#issuecomment-3494030172", "date": null, "doc": null, "version": "ai-benchmarks README @ 2f56f19", "vendor_page": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "no_power": true, "group": "llama323b", "excluded_reason": null, "excluded": false, "fits": true, "speed_eligible": true, "eff_ok": false, "soft_reason": null, "eligible": false, "headroom": null, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "group_size": 0, "group_rank": null }, { "platform": "DC-ROMA RISC-V Mainboard II (as tested by Jeff Geerling)", "platform_url": "/platform/dc-roma-risc-v-mainboard-ii-as-tested-by-jeff-geerling.html", "perf": 0.31, "perf_unit": "tok/s", "metric": "generation_throughput", "model": "llama3.2:3b", "precision": "Ollama default tag (4-bit unless tag says otherwise)", "stack": "Ollama (geerlingguy/ollama-benchmark)", "size_class": "LLM ≤4B", "batch": "1", "like_for_like": true, "differs": [], "confidence": "C", "confidence_label": "C – community/review", "power_w": null, "power_label": "not published", "power_kind": "unknown", "model_key": "llama323b", "pb": { "tag": "", "measured": false, "vendorMeasured": false, "typical": false, "scope": "", "rmax": false, "rmin": false }, "power_scope": null, "power_is_range_max": false, "power_is_range_min": false, "is_partial": "false", "partial_note": null, "rank_default": false, "fpsw_outlier": false, "fpsw_note": null, "rank_note": null, "rank_fps": false, "rank_fpsw": false, "fpsw_ratio_median": null, "power_basis": "unknown", "power_note": "", "power_range": null, "power_src_url": null, "perf_per_w": null, "date": "2025-10-17 (page last edited)", "older_than_18_months": false, "superseded": true, "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": true, "date_basis": "page-edited", "form_factor": "independent test system", "vendor": "(independent test)", "source": { "name": "Jeff Geerling ai-benchmarks issue results (comments)", "url": "https://github.com/geerlingguy/ai-benchmarks/issues/28#issuecomment-3417366710", "date": null, "doc": null, "version": "ai-benchmarks README @ 2f56f19", "vendor_page": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "no_power": true, "group": "llama323b", "excluded_reason": null, "excluded": false, "fits": true, "speed_eligible": true, "eff_ok": false, "soft_reason": null, "eligible": false, "headroom": null, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "group_size": 0, "group_rank": null } ], "no_data": null, "legend": "Confidence: A audited MLPerf closed, B documented third-party, C community, D vendor-published, E estimate. Unknown is never assumed." }