{"slug": "leaderboard-by-hardware", "title": "Leaderboard by hardware", "summary": "A hardware-grouped benchmark leaderboard published by an unnamed benchmarking site ranks local model runs by the exact GPU, PCIe and RAM configuration used, so readers can match a model to their own hardware. On a PCIe 5.0 x8/x8 setup of two RTX 5060 Ti 16GB cards, Ornith 1.5 35B-A3B NVFP4 posted 2340 tokens/s prompt processing and 112 tokens/s generation at 178 s wall time with a 4/5 quality score, while Nemotron-3.5-Bolt-74B-A9B IQ2 QAT reached 1280 PP and 107.9 TG at 284.8 s and also scored 4/5. The board notes that tensor versus layer split mode depends on the motherboard's PCIe topology when more than one GPU is involved.", "body_md": "Best Elixir runs per hardware.\n\nThis board groups model benchmarking runs by hardware, i.e. the exact GPUs, PCIe lanes and RAM they run on, so you can pick the model that best matches the hardware you have. Split mode (tensor or layer) depends on the PCIe topology of the motherboard, if more than one GPU is involved.\n\nPCIe 3.0 x16\n\n| Model | Quant | Wall | PP | TG | Needle | Quality | \n|---|---|---|---|---|---|---|\n| [Ornith-1.5-35B-A3B MiniPlus V2.1](https://overbring.com/ai-benchmarks/2026-11-01-isvalorum-ornith-1.5-35b-a3b-miniplus-v2.1-gguf-rx-9060-xt-16gb-single-card-llama-server-vulkan/) | APEX-I MiniPlus V2.1 | 1227.33 s | 819.65 | 57.92 | hit | 2/5 | \n\nPCIe 4.0 x16 (CPU) + 4.0 x4 (PCH)\n\n| Model | Quant | Wall | PP | TG | Needle | Quality | \n|---|---|---|---|---|---|---|\n| [Groot-19B-A3B-261023 MTP](https://overbring.com/ai-benchmarks/2026-10-31-prithivmlmods-groot-19b-a3b-261023-q5-k-m-gguf-rtx-4070-super-12gb-plus-4060-ti-16gb-layer-split-llama-server-cuda/) | Q5_K_M | 227.7 s | 2496 | 80.2 | hit | 2/5 | \n\nPCIe 5.0 x8/x8\n\n| Model | Quant | Wall | PP | TG | Needle | Quality | \n|---|---|---|---|---|---|---|\n| [Nemotron-3.5-Bolt-74B-A9B IQ2 QAT](https://overbring.com/ai-benchmarks/2026-10-31-nvidia-nemotron-35-bolt-74b-a9b-iq2-qat-gguf-rtx-5060-ti-16gb-x2-layer-split-llama-server-cuda/) | IQ2 QAT | 284.8 s | 1280 | 107.9 | hit | 4/5 | \n| [Muse-Glimmer NVFP4](https://overbring.com/ai-benchmarks/2026-10-31-freedom-aisvr-muse-glimmer-30b-nvfp4-gguf-rtx-5060-ti-16gb-x2-layer-split-llama-server-cuda/) | NVFP4 | 113.4 s | 1224 | 74.9 | hit | 4/5 | \n| [Ornith 1.5 35B-A3B NVFP4](https://overbring.com/ai-benchmarks/2026-10-30-tizzyt566-ornith-1.5-35b-a3b-nvfp4-gguf-rtx-5060-ti-16gb-x2-tensor-split-llama-server-cuda/) | NVFP4 | 178 s | 2340 | 112 | hit | 4/5 | \n| [flint NVFP4](https://overbring.com/ai-benchmarks/2026-10-31-recursive-si-flint-nvfp4-gguf-rtx-5060-ti-16gb-x2-layer-split-llama-server-cuda/) | NVFP4 | 371.5 s | 815 | 84.9 | cliff | 2/5 | \n| [LFM2.5-24B-A2B Q8_0](https://overbring.com/ai-benchmarks/2026-10-31-unsloth-lfm2.5-24b-a2b-q8_0-gguf-rtx-5060-ti-16gb-x2-tensor-split-llama-server-cuda/) | Q8_0 | 135.2 s | 1403 | 151 | hit | 2/5 | \n\nA paragraph about what you are working on is enough to start.", "url": "https://wpnews.pro/news/leaderboard-by-hardware", "canonical_source": "https://overbring.com/ai-benchmarks/leaderboard-by-hardware/", "published_at": "2026-10-29 21:00:00+00:00", "updated_at": "2026-11-09 01:58:48.642380+00:00", "lang": "en", "topics": ["large-language-models", "ai-infrastructure", "ai-tools"], "entities": ["Ornith-1.5-35B-A3B MiniPlus V2.1", "Groot-19B-A3B-261023 MTP", "Nemotron-3.5-Bolt-74B-A9B", "Muse-Glimmer", "LFM2.5-24B-A2B", "RTX 5060 Ti 16GB", "RTX 4070 Super 12GB", "RX 9060 XT 16GB"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/leaderboard-by-hardware", "markdown": "https://wpnews.pro/news/leaderboard-by-hardware.md", "text": "https://wpnews.pro/news/leaderboard-by-hardware.txt", "jsonld": "https://wpnews.pro/news/leaderboard-by-hardware.jsonld"}}