| Build | GPU(s) | Memory GB | Results | Models |
|---|
| $2K EPYC 64-core, 512 GB DDR4, CPU-only | None (CPU inference) | 0 | 1 | DeepSeek R1 671B |
| 2x Mac Studio M5 Ultra cluster (Thunderbolt 5 RDMA): Muse Spark Open 816B | 2x Apple M5 Ultra (unified, Mac Studio) | | 4 | Mistral Medium 3.5 128B, Muse Spark Open 816B-A42B, Qwen3.8 27B |
| 4x Mac Studio M5 Ultra cluster (Thunderbolt 5 RDMA): Muse Spark Open at 36.7 tok/s | 4x Apple M5 Ultra (unified, Mac Studio) | | 3 | Mistral Medium 3.5 128B, Muse Spark Open 816B-A42B, Qwen3.8 27B |
| 4x RTX 3090 + 32-core Threadripper, 192 GB DDR4 ⚠ unchecked | 4x RTX 3090 | 96 | 1 | Qwen3 235B-A22B |
| 4x RTX 3090 + EPYC 7443, 256 GB DDR4 8-channel ⚠ unchecked | 4x RTX 3090 | 96 | 1 | Qwen3 235B-A22B |
| 4x RTX 3090 homelab on Threadripper 2920X ⚠ unchecked | 4x RTX 3090 (~$625 each, used) | 96 | 0 | |
| 4x RTX 3090 in a 4U rack case (AM5) ⚠ unchecked | 4x Dell OEM RTX 3090 | 96 | 0 | |
| 4x RTX 3090 lab: Qwen3.6 on vLLM vs llama.cpp | 4x RTX 3090 | 96 | 3 | Qwen3.6 35B-A3B |
| AmpereOne A192-32X 512GB (CPU only) | None (CPU only) | | 4 | DeepSeek R1 671B, Llama 3.1 405B, Llama 3.1 70B, Llama 3.2 3B |
| Apple M1 Max 64GB | Apple M1 Max (unified) | 64 | 1 | gpt-oss 20B |
| Apple M4 Max 36GB | Apple M4 Max (unified) | 36 | 1 | gpt-oss 20B |
| Dell Pro Max with GB10 (Nvidia DGX Spark) | Nvidia GB10 (unified) | | 4 | DeepSeek R1 Distill 14B, gpt-oss 20B, Llama 3.1 70B, Llama 3.2 3B |
| Dual RTX 3090 on Ryzen 3700X (X470) ⚠ unchecked | 2x Zotac RTX 3090 | 48 | 2 | Llama 3 70B |
| Dual used RTX 3090, ~2000 EUR total ⚠ unchecked | 2x RTX 3090 (used) | 48 | 1 | Llama 3.3 70B |
| EPYC + single RTX 3090 (hybrid) ⚠ unchecked | RTX 3090 | 24 | 1 | DeepSeek R1 671B |
| EPYC 7F52 jank rig: 4x3090 + 4090 + 4080 (136 GB VRAM) ⚠ unchecked | 4x RTX 3090, RTX 4090, RTX 4080 | 136 | 1 | Llama 3.1 70B |
| Framework Desktop (AMD Strix Halo) mainboard | AMD Ryzen AI Max+ 395 (these Ollama runs used CPU only) | | 4 | DeepSeek R1 Distill 70B, DeepSeek R1 Distill 8B, Llama 3.1 70B, Llama 3.2 3B |
| HP OmniBook Ultra 16 (RTX Spark N1X, 128GB) laptop: local AI benchmarks | Nvidia RTX Spark N1X, 6,144-core Blackwell GPU (unified) | 128 | 3 | gpt-oss 20B, Llama 3.1 8B, Qwen3.6 35B-A3B |
| Intel Core Ultra 265K + Radeon AI Pro R9700 | AMD Radeon AI Pro R9700 | 32 | 3 | DeepSeek R1 Distill 14B, gpt-oss 20B, Llama 3.2 3B |
| Intel Core Ultra 265K + RTX 2080 Ti | RTX 2080 Ti | 11 | 2 | DeepSeek R1 Distill 14B, Llama 3.2 3B |
| Intel Core Ultra 265K + RTX 3060 | RTX 3060 | 12 | 3 | DeepSeek R1 Distill 14B, gpt-oss 20B, Llama 3.2 3B |
| Intel Core Ultra 265K + RTX 3080 Ti | RTX 3080 Ti | 12 | 2 | DeepSeek R1 Distill 14B, Llama 3.2 3B |
| Intel Core Ultra 265K + RTX 4070 Ti | RTX 4070 Ti | 12 | 3 | DeepSeek R1 Distill 14B, gpt-oss 20B, Llama 3.2 3B |
| Intel Core Ultra 265K + RTX 4090 | RTX 4090 | 24 | 3 | DeepSeek R1 Distill 14B, gpt-oss 20B, Llama 3.2 3B |
| Intel Core Ultra 265K + RTX A4000 | RTX A4000 | 16 | 2 | gpt-oss 20B, Llama 3.2 3B |
| Intel i9-13900K + RTX 5090 | RTX 5090 | 32 | 1 | Llama 3.2 3B |
| M4 Mac mini (10-core) 32GB | Apple M4 (unified) | 32 | 1 | Llama 3.2 3B |
| Mac Studio 64 GB ⚠ unchecked | Apple unified memory | 64 | 1 | Llama 3.3 70B |
| Mac Studio M1 Ultra | Apple M1 Ultra | | 1 | Llama 3.1 8B |
| Mac Studio M1 Ultra (64-core GPU) 128GB | Apple M1 Ultra 64-core GPU (unified) | 128 | 4 | DeepSeek R1 Distill 14B, gpt-oss 20B, Llama 3.1 70B, Llama 3.2 3B |
| Mac Studio M2 Ultra 76-core GPU, 192 GB | Apple M2 Ultra 76-core GPU | 192 | 3 | gpt-oss 120B, gpt-oss 20B, Llama 2 7B |
| Mac Studio M3 Ultra 512GB (Geerling test) | Apple M3 Ultra (unified) | 512 | 6 | DeepSeek R1 671B, DeepSeek R1 Distill 14B, gpt-oss 20B, Llama 3.1 405B, Llama 3.1 70B |
| Mac Studio M3 Ultra 512GB: Muse Spark Open 816B (Hugging Face test) | Apple M3 Ultra (unified) | 512 | 2 | Muse Spark Open 816B-A42B |
| Mac Studio M3 Ultra 512GB: Qwen3.8 27B and Mistral Medium 3.5 (Apple MLX benchmark) | Apple M3 Ultra (unified) | 512 | 2 | Mistral Medium 3.5 128B, Qwen3.8 27B |
| Mac Studio M3 Ultra, 512 GB ⚠ unchecked | Apple M3 Ultra | 512 | 1 | DeepSeek R1 671B |
| Mac Studio M5 Ultra 512GB: Qwen3.8 27B and Mistral Medium 3.5 (Apple MLX benchmark) | Apple M5 Ultra (unified) | 512 | 2 | Mistral Medium 3.5 128B, Qwen3.8 27B |
| MacBook Pro M1 Max 32-core GPU, 64 GB | Apple M1 Max 32-core GPU | 64 | 1 | Llama 3 8B |
| MacBook Pro M2 Max | Apple M2 Max | | 1 | Llama 3.1 8B |
| MacBook Pro M3 Max 40-core GPU, 64 GB | Apple M3 Max 40-core GPU | 64 | 1 | Llama 3 8B |
| Microsoft Surface Laptop Ultra (RTX Spark N1X): llama.cpp speeds on 2026 models (Micro Center) | Nvidia RTX Spark N1X, 6,144-core Blackwell GPU (unified) | | 5 | GLM 4.7 Flash, Muse Glimmer 30B, Nemotron 3 Nano Omni 30B, Qwen3.5 122B, Qwen3.8 27B |
| RTX 3060 12GB: 2026 models (llamabench community) | RTX 3060 | 12 | 3 | Gemma 4 12B, LFM2.5 24B-A2B, Qwen3.8 27B |
| RTX 3090 + RTX 4060 Ti 16GB mixed pair ⚠ unchecked | RTX 3090 + RTX 4060 Ti 16GB | 40 | 1 | Llama 3.3 70B |
| RTX 3090: 2026 models (llamabench community) | RTX 3090 | 24 | 3 | Qwen3.6 27B, Qwen3.6 35B-A3B, Qwen3.8 27B |
| RTX 4060 Ti 16GB: 2026 models (llamabench community) | RTX 4060 Ti 16GB | 16 | 2 | LFM2.5 24B-A2B, Qwen3.8 27B |
| RTX 4090 desktop (speed comparison vs Macs) | RTX 4090 24GB | 24 | 2 | Llama 3.1 8B, Mistral Small 22B |
| RTX 4090: 2026 models (llamabench community) | RTX 4090 | 24 | 8 | Gemma 4 12B, Gemma 4 26B-A4B, Gemma 4 31B, gpt-oss 20B, LFM2.5 24B-A2B, Nemotron 3.5 Bolt 74B-A9B, Qwen3.5 9B, Qwen3.6 27B |
| RTX 5070 Ti: 2026 models (llamabench community) | RTX 5070 Ti | 16 | 1 | Gemma 4 12B |
| RTX 5090: 2026 models (llamabench community) | RTX 5090 | 32 | 7 | Gemma 4 12B, Gemma 4 26B-A4B, Gemma 4 31B, gpt-oss 20B, Qwen3.5 9B, Qwen3.6 27B, Qwen3.6 35B-A3B |
| RTX Pro 6000 96GB workstation card | RTX Pro 6000 (Blackwell) | 96 | 2 | gpt-oss 120B, gpt-oss 20B |
| Ryzen 9 7900X + RTX 4090 | RTX 4090 | 24 | 2 | Llama 3.1 70B, Llama 3.2 3B |
| Ryzen 9 7950X + RTX 4070 Ti Super | RTX 4070 Ti Super | 16 | 1 | Llama 3.2 3B |
| Ryzen 9 7950X + RTX 4080 | RTX 4080 | 16 | 1 | Llama 3.2 3B |
| Ryzen 9 9950X + Radeon RX 7900 XT | AMD Radeon RX 7900 XT | 20 | 1 | Llama 3.2 3B |
| Single RTX 3090 24GB | RTX 3090 | 24 | 1 | gpt-oss 20B |
| Single RTX 3090: Qwen3.6 35B-A3B (Giles Thomas) | RTX 3090 | 24 | 1 | Qwen3.6 35B-A3B |
| Single RTX 3090: Qwen3.8 27B at 78-140 tok/s with the OpenJet megakernel vs llama.cpp | RTX 3090 | 24 | 6 | Qwen3.8 27B |
| Single RTX 4060 Ti 16GB + 64 GB RAM (partial offload) ⚠ unchecked | RTX 4060 Ti 16GB | 16 | 1 | Llama 3.3 70B |
| Single RTX 4090 (llama.cpp CUDA scoreboard) | RTX 4090 24GB | 24 | 1 | Llama 2 7B |
| Single RTX 5060 Ti 16GB | RTX 5060 Ti 16GB | 16 | 1 | gpt-oss 20B |
| Single RTX 5070 Ti 16GB | RTX 5070 Ti | 16 | 1 | gpt-oss 20B |
| Single RTX 5080 16GB | RTX 5080 | 16 | 1 | gpt-oss 20B |
| Single RTX 5090 32GB | RTX 5090 | 32 | 1 | gpt-oss 20B |
| Strix Halo (Radeon 8060S, 96 GB): 2026 models (llamabench community) | AMD Ryzen AI Max+ (Radeon 8060S iGPU) | 96 | 1 | Qwen3.6 35B-A3B |