Hardware · Clusters · Example cluster

Two older laptops on Wi-Fi: what it runs together

A 16 GB MacBook Air (M1) and a gaming laptop with a 6 GB RTX 3060, joined over home Wi-Fi.

The cluster many people can build today, and the one that shows the limits: together the two laptops run 30B-class models neither holds alone, but slowly, and Wi-Fi adds a delay to every word and seconds to every long prompt. Use a cable if you can.

Change the devices How clustering works

16.7 GBusable memory together (the largest device has 10.7 GB)

31models only the cluster can run

143run better or faster than on any one device

57B (14.0B active)biggest model it adds: Qwen2 57B-A14B Instruct

Connected by Wi-Fi 6 or 6E. The wait for the first word is worked out for a 1,000-token prompt (a short chat).

Wi-Fi slows every word. Each token crosses the network once per device, and Wi-Fi adds a few milliseconds each time, plus seconds for long prompts. A network cable, or a Thunderbolt cable between two Macs, is much faster.

The devices

Tool to use

llama.cpp RPC best estimate for 173 of the 174 models the cluster adds

Mixed devices you already own, a Mac, a gaming PC, a mini PC or a Jetson, when one GGUF model won't fit on any of them alone.

ggml-rpc-server --host 0.0.0.0 --port 50052 -c   # on each worker
llama-server -hf <model-GGUF> -ngl 99 --rpc 192.168.1.10:50052,192.168.1.11:50052   # on the main machine

Also works. The number is how many of the added models each one runs:

459 LLMs this cluster can run

Models the cluster adds come first, best match first, then the ones a single device already runs as well. Filter vs one device to pick. Speeds are estimates for one person chatting.

Quality Model Quant Memory Speed Runs Devices Tool vs one device
75Great quality (A) · 75 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 71/100 (quality, fit, speed and memory use) sets the ranking. Flint (40B-A2.5B)Recursive · 40B (2.5B active) Q3_K_M 22.1 GB 26.0 tok/s Runs welloffloads to RAM NVIDIA PC (card + RAM) + MacPipeline split llama.cpp RPC Needs the clusterWon’t run on any one of them
51Fair quality (C) · 51 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. Kimi Linear 48B-A3BMoonshot AI · 49B (3.0B active) Q2_K 22.1 GB 27.2 tok/s Runs welloffloads to RAM NVIDIA PC (card + RAM) + MacPipeline split llama.cpp RPC Needs the clusterWon’t run on any one of them
67Good quality (B) · 67 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking. Seed-OSS 36B InstructByteDance · 36B Q3_K_M 22.2 GB 3.2 tok/s Slowoffloads to RAM NVIDIA PC (card + RAM) + MacPipeline split llama.cpp RPC Needs the clusterWon’t run on any one of them
66Good quality (B) · 66 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking. Hermes 4.3 36BNous Research · 36B Q3_K_M 22.2 GB 3.2 tok/s Slowoffloads to RAM NVIDIA PC (card + RAM) + MacPipeline split llama.cpp RPC Needs the clusterWon’t run on any one of them
65Good quality (B) · 65 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 58/100 (quality, fit, speed and memory use) sets the ranking. QwQ 32BAlibaba · 33B Q3_K_M 20.6 GB 3.6 tok/s Slowoffloads to RAM NVIDIA PC (card + RAM) + MacPipeline split llama.cpp RPC Needs the clusterWon’t run on any one of them
65Good quality (B) · 65 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 58/100 (quality, fit, speed and memory use) sets the ranking. Qwen3-VL 32B InstructAlibaba · 33B Q3_K_M 20.9 GB 3.5 tok/s Slowoffloads to RAM NVIDIA PC (card + RAM) + MacPipeline split llama.cpp RPC Needs the clusterWon’t run on any one of them
63Good quality (B) · 63 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 58/100 (quality, fit, speed and memory use) sets the ranking. OpenReasoning Nemotron 32BNVIDIA · 33B Q3_K_M 20.6 GB 3.6 tok/s Slowoffloads to RAM NVIDIA PC (card + RAM) + MacPipeline split llama.cpp RPC Needs the clusterWon’t run on any one of them
63Good quality (B) · 63 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 58/100 (quality, fit, speed and memory use) sets the ranking. Skywork OR1 32BSkywork · 33B Q3_K_M 20.6 GB 3.6 tok/s Slowoffloads to RAM NVIDIA PC (card + RAM) + MacPipeline split llama.cpp RPC Needs the clusterWon’t run on any one of them
63Good quality (B) · 63 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 58/100 (quality, fit, speed and memory use) sets the ranking. Qwen3 32BAlibaba · 33B Q3_K_M 20.6 GB 3.6 tok/s Slowoffloads to RAM NVIDIA PC (card + RAM) + MacPipeline split llama.cpp RPC Needs the clusterWon’t run on any one of them
62Good quality (B) · 62 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 57/100 (quality, fit, speed and memory use) sets the ranking. EXAONE Deep 32BLG AI Research · 32B Q3_K_M 20.2 GB 3.7 tok/s Slowoffloads to RAM NVIDIA PC (card + RAM) + MacPipeline split llama.cpp RPC Needs the clusterWon’t run on any one of them
23Basic quality (D) · 23 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 57/100 (quality, fit, speed and memory use) sets the ranking. Qwen1.5 14B ChatAlibaba · 14B Q2_K 14.7 GB 10.7 tok/s Runs welltight fit NVIDIA PC + MacPipeline split llama.cpp RPC Needs the clusterWon’t run on any one of them
60Good quality (B) · 60 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 56/100 (quality, fit, speed and memory use) sets the ranking. DeepSeek R1 Distill Qwen 32BDeepSeek · 33B Q3_K_M 20.6 GB 3.6 tok/s Slowoffloads to RAM NVIDIA PC (card + RAM) + MacPipeline split llama.cpp RPC Needs the clusterWon’t run on any one of them
58Good quality (B) · 58 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 55/100 (quality, fit, speed and memory use) sets the ranking. Baichuan M2 32BBaichuan · 33B Q3_K_M 20.6 GB 3.6 tok/s Slowoffloads to RAM NVIDIA PC (card + RAM) + MacPipeline split llama.cpp RPC Needs the clusterWon’t run on any one of them
58Good quality (B) · 58 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 55/100 (quality, fit, speed and memory use) sets the ranking. OpenHands LM 32BAll Hands AI · 33B Q3_K_M 20.6 GB 3.6 tok/s Slowoffloads to RAM NVIDIA PC (card + RAM) + MacPipeline split llama.cpp RPC Needs the clusterWon’t run on any one of them
58Good quality (B) · 58 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 55/100 (quality, fit, speed and memory use) sets the ranking. Qwen2.5 32B InstructAlibaba · 33B Q3_K_M 20.6 GB 3.6 tok/s Slowoffloads to RAM NVIDIA PC (card + RAM) + MacPipeline split llama.cpp RPC Needs the clusterWon’t run on any one of them
57Good quality (B) · 57 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 55/100 (quality, fit, speed and memory use) sets the ranking. Qwen2.5-VL 32B InstructAlibaba · 33B Q3_K_M 20.9 GB 3.5 tok/s Slowoffloads to RAM NVIDIA PC (card + RAM) + MacPipeline split llama.cpp RPC Needs the clusterWon’t run on any one of them
56Good quality (B) · 56 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 54/100 (quality, fit, speed and memory use) sets the ranking. Qwen2.5 Coder 32B InstructAlibaba · 33B Q3_K_M 20.6 GB 3.6 tok/s Slowoffloads to RAM NVIDIA PC (card + RAM) + MacPipeline split llama.cpp RPC Needs the clusterWon’t run on any one of them
55Good quality (B) · 55 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 54/100 (quality, fit, speed and memory use) sets the ranking. QwQ 32B PreviewAlibaba · 33B Q3_K_M 20.6 GB 3.6 tok/s Slowoffloads to RAM NVIDIA PC (card + RAM) + MacPipeline split llama.cpp RPC Needs the clusterWon’t run on any one of them
53Fair quality (C) · 53 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 53/100 (quality, fit, speed and memory use) sets the ranking. InternVL3.5 38BShanghai AI Lab · 38B Q2_K 19.9 GB 3.9 tok/s Slowoffloads to RAM NVIDIA PC (card + RAM) + MacPipeline split llama.cpp RPC Needs the clusterWon’t run on any one of them
39Basic quality (D) · 39 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 52/100 (quality, fit, speed and memory use) sets the ranking. Phi-3.5 MoE (42B-A6.6B)Microsoft · 42B (6.6B active) Q2_K 20.2 GB 17.0 tok/s Runs welloffloads to RAM NVIDIA PC (card + RAM) + MacPipeline split llama.cpp RPC Needs the clusterWon’t run on any one of them
49Fair quality (C) · 49 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 51/100 (quality, fit, speed and memory use) sets the ranking. InternVL3 38BShanghai AI Lab · 38B Q2_K 19.9 GB 3.9 tok/s Slowoffloads to RAM NVIDIA PC (card + RAM) + MacPipeline split llama.cpp RPC Needs the clusterWon’t run on any one of them
36Basic quality (D) · 36 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 48/100 (quality, fit, speed and memory use) sets the ranking. Nous Hermes 2 Mixtral 8x7B DPONous Research · 47B (12.9B active) Q2_K 22.1 GB 10.0 tok/s Runs welloffloads to RAM NVIDIA PC (card + RAM) + MacPipeline split llama.cpp RPC Needs the clusterWon’t run on any one of them
56Good quality (B) · 56 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 48/100 (quality, fit, speed and memory use) sets the ranking. Llama 3.3 Nemotron Super 49B v1.5NVIDIA · 50B Q2_K 25.0 GB 2.9 tok/s Very slowheavy RAM offload NVIDIA PC (card + RAM) + MacPipeline split llama.cpp RPC Needs the clusterWon’t run on any one of them
35Basic quality (D) · 35 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 48/100 (quality, fit, speed and memory use) sets the ranking. Mixtral 8x7B InstructMistral AI · 47B (12.9B active) Q2_K 22.1 GB 10.0 tok/s Runs welloffloads to RAM NVIDIA PC (card + RAM) + MacPipeline split llama.cpp RPC Needs the clusterWon’t run on any one of them
36Basic quality (D) · 36 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 46/100 (quality, fit, speed and memory use) sets the ranking. Qwen1.5 32B ChatAlibaba · 33B Q3_K_M 20.4 GB 3.6 tok/s Slowoffloads to RAM NVIDIA PC (card + RAM) + MacPipeline split llama.cpp RPC Needs the clusterWon’t run on any one of them
36Basic quality (D) · 36 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 46/100 (quality, fit, speed and memory use) sets the ranking. DeepSeek Coder 33B InstructDeepSeek · 33B Q3_K_M 20.8 GB 3.6 tok/s Slowoffloads to RAM NVIDIA PC (card + RAM) + MacPipeline split llama.cpp RPC Needs the clusterWon’t run on any one of them
35Basic quality (D) · 35 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 42/100 (quality, fit, speed and memory use) sets the ranking. Jamba 1.5 MiniAI21 Labs · 52B (12.0B active) Q2_K 23.1 GB 10.5 tok/s Slowheavy RAM offload NVIDIA PC (card + RAM) + MacPipeline split llama.cpp RPC Needs the clusterWon’t run on any one of them
35Basic quality (D) · 35 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 41/100 (quality, fit, speed and memory use) sets the ranking. Qwen2 57B-A14B InstructAlibaba · 57B (14.0B active) Q2_K 25.8 GB 9.2 tok/s Slowheavy RAM offload NVIDIA PC (card + RAM) + MacPipeline split llama.cpp RPC Needs the clusterWon’t run on any one of them
37Basic quality (D) · 37 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 40/100 (quality, fit, speed and memory use) sets the ranking. Command R 35B (v01)Cohere · 35B Q2_K 27.1 GB 4.1 tok/s Slowheavy RAM offload NVIDIA PC (card + RAM) + MacPipeline split llama.cpp RPC Needs the clusterWon’t run on any one of them
35Basic quality (D) · 35 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 39/100 (quality, fit, speed and memory use) sets the ranking. Aya 23 35BCohere · 35B Q2_K 27.1 GB 4.1 tok/s Slowheavy RAM offload NVIDIA PC (card + RAM) + MacPipeline split llama.cpp RPC Needs the clusterWon’t run on any one of them
11Basic quality (D) · 11 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 34/100 (quality, fit, speed and memory use) sets the ranking. Falcon 40B InstructTII · 42B Q2_K 19.3 GB 3.7 tok/s Slowoffloads to RAM NVIDIA PC (card + RAM) + MacPipeline split llama.cpp RPC Needs the clusterWon’t run on any one of them
77Great quality (A) · 77 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 85/100 (quality, fit, speed and memory use) sets the ranking. Gemma 4 26B-A4BGoogle · 26B (3.8B active) Q3_K_M 15.4 GB 21.7 tok/s Runs welltight fit NVIDIA PC + MacPipeline split llama.cpp RPC Better on the clusterAlone: Slow on NVIDIA PC, 35.1 tok/s
78Great quality (A) · 78 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 82/100 (quality, fit, speed and memory use) sets the ranking. Gemma 4 12BGoogle · 12B Q4_K_M 10.3 GB 10.6 tok/s Runs well NVIDIA PC + MacPipeline split llama.cpp RPC Better on the clusterAlone: Runs well on NVIDIA PC, 22.4 tok/s
66Good quality (B) · 66 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 82/100 (quality, fit, speed and memory use) sets the ranking. Maple Preview (20B-A1B)DeepGrove · 20B (1.5B active) Q4_K_M 14.8 GB 30.6 tok/s Runs welltight fit NVIDIA PC + MacPipeline split llama.cpp RPC Better on the clusterAlone: Runs well on Mac, 63.7 tok/s
76Great quality (A) · 76 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 81/100 (quality, fit, speed and memory use) sets the ranking. K2 Horizon 7BInstitute of Foundation Models · 9.0B Q4_K_M 9.1 GB 16.1 tok/s Runs well NVIDIA PC + MacPipeline split llama.cpp RPC Better on the clusterAlone: Runs well on Mac, 10.4 tok/s
70Great quality (A) · 70 of 100: how capable the model is at MXFP4, whatever it runs on. Match score 81/100 (quality, fit, speed and memory use) sets the ranking. gpt-oss 20BOpenAI · 21B (3.6B active) MXFP4 14.5 GB 20.2 tok/s Runs welltight fit NVIDIA PC + MacPipeline split llama.cpp RPC Better on the clusterAlone: Slow on NVIDIA PC, 32.3 tok/s
65Good quality (B) · 65 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 80/100 (quality, fit, speed and memory use) sets the ranking. MiniCPM-V 4.7 35B-A3BOpenBMB · 35B (3.0B active) Q2_K 16.6 GB 28.2 tok/s Runs welltight fit NVIDIA PC + MacPipeline split llama.cpp RPC Better on the clusterAlone: Slow on NVIDIA PC, 52.1 tok/s
64Good quality (B) · 64 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 80/100 (quality, fit, speed and memory use) sets the ranking. Nex N2 Mini (35B-A3B)Nex-AGI · 35B (3.0B active) Q2_K 16.6 GB 28.2 tok/s Runs welltight fit NVIDIA PC + MacPipeline split llama.cpp RPC Better on the clusterAlone: Slow on NVIDIA PC, 52.2 tok/s
63Good quality (B) · 63 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 80/100 (quality, fit, speed and memory use) sets the ranking. KAT-Coder V2.5 Dev (35B-A3B)Kuaishou · 35B (3.0B active) Q2_K 16.4 GB 28.3 tok/s Runs welltight fit NVIDIA PC + MacPipeline split llama.cpp RPC Better on the clusterAlone: Slow on NVIDIA PC, 52.4 tok/s
66Good quality (B) · 66 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 79/100 (quality, fit, speed and memory use) sets the ranking. ERNIE 4.5 21B-A3B ThinkingBaidu · 22B (3.0B active) Q4_K_M 16.1 GB 20.9 tok/s Runs welltight fit NVIDIA PC + MacPipeline split llama.cpp RPC Better on the clusterAlone: Runs well on Mac, 35.2 tok/s
62Good quality (B) · 62 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 79/100 (quality, fit, speed and memory use) sets the ranking. LFM2.5 24B-A2BLiquid AI · 24B (2.5B active) Q3_K_M 14.3 GB 28.3 tok/s Runs welltight fit NVIDIA PC + MacPipeline split llama.cpp RPC Better on the clusterAlone: Runs well on Mac, 35.7 tok/s
61Good quality (B) · 61 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 79/100 (quality, fit, speed and memory use) sets the ranking. GLM-4.7 Flash (30B-A3B)Zhipu AI · 31B (3.0B active) Q2_K 15.3 GB 28.9 tok/s Runs welltight fit NVIDIA PC + MacPipeline split llama.cpp RPC Better on the clusterAlone: Slow on NVIDIA PC, 53.3 tok/s
65Good quality (B) · 65 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 79/100 (quality, fit, speed and memory use) sets the ranking. DiffusionGemma 26B-A4B ITGoogle · 26B (4.0B active) Q3_K_M 15.4 GB 21.0 tok/s Runs welltight fit NVIDIA PC + MacPipeline split llama.cpp RPC Better on the clusterAlone: Slow on NVIDIA PC, 33.6 tok/s
60Good quality (B) · 60 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 79/100 (quality, fit, speed and memory use) sets the ranking. Qwen AgentWorld 35B-A3BAlibaba · 35B (3.0B active) Q2_K 16.4 GB 28.3 tok/s Runs welltight fit NVIDIA PC + MacPipeline split llama.cpp RPC Better on the clusterAlone: Slow on NVIDIA PC, 52.4 tok/s
60Good quality (B) · 60 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 78/100 (quality, fit, speed and memory use) sets the ranking. Agents-A1 35B-A3BInternScience · 35B (3.0B active) Q2_K 16.6 GB 28.2 tok/s Runs welltight fit NVIDIA PC + MacPipeline split llama.cpp RPC Better on the clusterAlone: Slow on NVIDIA PC, 52.2 tok/s
59Good quality (B) · 59 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 78/100 (quality, fit, speed and memory use) sets the ranking. Hunyuan Hy-Memory 30B-A3BTencent · 30B (3.0B active) Q2_K 15.2 GB 28.9 tok/s Runs welltight fit NVIDIA PC + MacPipeline split llama.cpp RPC Better on the clusterAlone: Slow on NVIDIA PC, 52.5 tok/s
60Good quality (B) · 60 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 77/100 (quality, fit, speed and memory use) sets the ranking. ERNIE 4.5 21B-A3BBaidu · 22B (3.0B active) Q4_K_M 16.2 GB 20.9 tok/s Runs welltight fit NVIDIA PC + MacPipeline split llama.cpp RPC Better on the clusterAlone: Runs well on Mac, 35.2 tok/s
57Good quality (B) · 57 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 77/100 (quality, fit, speed and memory use) sets the ranking. Kimi-VL A3B Thinking 2506Moonshot AI · 16B (3.0B active) Q4_K_M 12.6 GB 23.0 tok/s Runs well NVIDIA PC + MacPipeline split llama.cpp RPC Better on the clusterAlone: Runs well on Mac, 28.2 tok/s
56Good quality (B) · 56 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 76/100 (quality, fit, speed and memory use) sets the ranking. North Mini Code 1.0 (30B-A3B)Cohere · 30B (3.0B active) Q2_K 14.9 GB 29.1 tok/s Runs welltight fit NVIDIA PC + MacPipeline split llama.cpp RPC Better on the clusterAlone: Slow on NVIDIA PC, 41.2 tok/s
56Good quality (B) · 56 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 76/100 (quality, fit, speed and memory use) sets the ranking. Nemotron Cascade 2 30B-A3BNVIDIA · 32B (3.5B active) Q2_K 15.0 GB 26.7 tok/s Runs welltight fit NVIDIA PC + MacPipeline split llama.cpp RPC Better on the clusterAlone: Slow on NVIDIA PC, 48.0 tok/s

Other example clusters

Estimates come from published hardware specs and our open formulas, not from running the model on your machine. Real speed varies with runtime, drivers and settings. Numbers marked “Measured” come from a linked benchmark.

Was this page helpful?