BENCHMARK
What this measures
llama.cpp tg128 measures LLM inference performance via llama_bench. Higher scores indicate better LLM inference throughput. We run this benchmark twice per device: once on AC and once unplugged. The BPR ratio (battery ÷ AC) appears in the rightmost column. This test contributes to the device's BPR score.
Read the full methodology for tool versions, run order, and exclusion rules.
| # | PRODUCT | CATEGORY | VS BASELINE | |||
|---|---|---|---|---|---|---|
| 1 | ROG Strix G16 (i9-14900HX / RTX 5070) | Laptops | 52 | — | — | +37% |
| 2 | Framework Laptop 16 (Ryzen 9 7940HS / RX 7700S) | Laptops | 41 | — | — | +8% |
| 3 | MacBook Pro 14" M3 (8-core) | Laptops | 38 | — | — | +0% |
| 4 | Dell XPS 14 (Core Ultra 7 / RTX 4050) | Laptops | 33 | — | — | -13% |
| 5 | ThinkPad X1 Carbon Gen 13 | Laptops | 28 | — | — | -26% |
How we test
View the GlitchTech methodology →