# RigBase

> RigBase is a database of 187 real home AI rigs (NVIDIA/AMD/Intel GPUs, Apple Silicon Macs, AMD Strix Halo, NVIDIA DGX Spark) with 442 measured local LLM generation speeds in tokens per second. Every number links to its public original source (benchmarks, posts, repos); RigBase does not extrapolate or estimate measured speeds. Calculator results (VRAM, KV cache) are labelled as arithmetic, not measurements.

When citing, link the specific page (e.g. a model or GPU page) and name the original source shown next to each number. Data licence: CC BY 4.0. Prices are dated and sourced on each page.

## Start here
- [Home: fastest measured builds and models](https://getrigbase.com/): Will Qwen3.8, Gemma 4 or gpt-oss run on your PC? Measured tokens/sec, VRAM, power and EU prices from 187 real home AI rigs, each linked to its source.
- [Can I run it? VRAM & KV cache calculator](https://getrigbase.com/can-it-run/): Free VRAM and KV cache calculator: can I run gpt-oss 120B on 4x 3090, Qwen on one RTX 3090, a 200B model on 2x DGX Spark? GGUF, AWQ, FP8, NVFP4 VRAM math…
- [DGX Spark: every recipe for current models, measured](https://getrigbase.com/dgx-spark/): 19 measured DGX Spark recipes for current local LLMs (GLM-5.3 Flash, Qwen3.8 Flash-Next, DeepSeek V4.1 Flash and more) on 1–4 Sparks: decode and prefill…
- [Free dataset: home AI builds and measured LLM speeds](https://getrigbase.com/data/): Download 187 home AI builds and 442 measured local-LLM tok/s rows as CSV/JSON. CC BY 4.0, every row linked to its source.
- [MCP server: query RigBase from Claude/Cursor/ChatGPT/Gemini](https://getrigbase.com/mcp-server/): Free public MCP endpoint at getrigbase.com/mcp: search 187 measured local-LLM builds, can-it-run checks, best GPU per model. Works with Claude, Cursor…

## Dataset
- [Hugging Face dataset: rigbase-local-llm-benchmarks](https://huggingface.co/datasets/rigbase/rigbase-local-llm-benchmarks): the same builds and results, CC BY 4.0
- [builds.csv](https://getrigbase.com/data/builds.csv): one row per measured result
- [builds.json](https://getrigbase.com/data/builds.json): full build records with source links
- [feed.json](https://getrigbase.com/feed.json): JSON Feed of every page with its last-updated date

## Models (what hardware runs each model, measured)
- [Best home rig for DeepSeek R1 671B](https://getrigbase.com/models/deepseek-r1-671b/): What GPU do you need for DeepSeek R1 671B? Weights ≈411 GB at Q4_K_M. Fastest measured: Apple M3 Ultra at 19.89 tok/s. 5 sourced results, from 512 GB cards up.
- [Best home rig for DeepSeek R1 Distill 14B](https://getrigbase.com/models/deepseek-r1-distill-14b/): What GPU do you need for DeepSeek R1 Distill 14B? Fastest measured: RTX 4090 at 92.49 tok/s. 9 sourced results, from 11 GB cards up.
- [Best home rig for DeepSeek R1 Distill 70B](https://getrigbase.com/models/deepseek-r1-distill-70b/): What GPU do you need for DeepSeek R1 Distill 70B? Fastest measured: Framework Desktop (AMD Strix Halo) mainboard at 2.43 tok/s. 1 sourced result.
- [Best home rig for DeepSeek R1 Distill 8B](https://getrigbase.com/models/deepseek-r1-distill-8b/): What GPU do you need for DeepSeek R1 Distill 8B? Fastest measured: Framework Desktop (AMD Strix Halo) mainboard at 19.49 tok/s. 1 sourced result.
- [Best home rig for DeepSeek V4.1 Flash](https://getrigbase.com/models/deepseek-v4-1-flash/): What GPU do you need for DeepSeek V4.1 Flash? Weights ≈511 GB at native. Fastest measured: 4x DGX Spark (GB10) at 134.8 tok/s. 12 sourced results, from 512 GB…
- [Best home rig for DeepSeek V4 Flash](https://getrigbase.com/models/deepseek-v4-flash/): What GPU do you need for DeepSeek V4 Flash? Fastest measured: 8x RTX 3090 at 95.52 tok/s. 5 sourced results, from 192 GB cards up.
- [Best home rig for Flint 40B-A2.5B](https://getrigbase.com/models/flint-40b-a2-5b/): What GPU do you need for Flint 40B-A2.5B? Weights ≈24 GB at Q4_K_M. Fastest measured: RTX 5090 at 224.9 tok/s. 8 sourced results, from 24 GB cards up.
- [Best home rig for Gemma 4 12B](https://getrigbase.com/models/gemma-4-12b/): What GPU do you need for Gemma 4 12B? Weights ≈7 GB at Q4_K_M. Fastest measured: RTX 5090 at 142 tok/s. 8 sourced results, from 12 GB cards up.
- [Best home rig for Gemma 4 26B-A4B](https://getrigbase.com/models/gemma-4-26b-a4b/): What GPU do you need for Gemma 4 26B-A4B? Weights ≈15 GB at Q4_K_M. Fastest measured: RTX 5090 at 235 tok/s. 10 sourced results, from 24 GB cards up.
- [Best home rig for Gemma 4 31B](https://getrigbase.com/models/gemma-4-31b/): What GPU do you need for Gemma 4 31B? Weights ≈19 GB at Q4_K_M. Fastest measured: RTX 5090 at 71.3 tok/s. 9 sourced results, from 24 GB cards up.
- [Best home rig for GLM 4.7 Flash](https://getrigbase.com/models/glm-4-7-flash/): What GPU do you need for GLM 4.7 Flash? Fastest measured: RTX 4090 at 164.9 tok/s. 3 sourced results, from 24 GB cards up.
- [Best home rig for GLM-5.2](https://getrigbase.com/models/glm-5-2/): What GPU do you need for GLM-5.2? Weights ≈461 GB at Q4_K_M. Fastest measured: 4x DGX Spark (GB10) at 32.5 tok/s. 1 sourced result, from 512 GB cards up.
- [Best home rig for GLM-5.3 Flash](https://getrigbase.com/models/glm-5-3-flash/): What GPU do you need for GLM-5.3 Flash? Weights ≈196 GB at Q4_K_M. Fastest measured: 4x RTX Pro 6000 (Blackwell, 96 GB) at 265.9 tok/s. 17 sourced results…
- [Best home rig for GLM-5.3](https://getrigbase.com/models/glm-5-3/): What GPU do you need for GLM-5.3? Weights ≈461 GB at Q4_K_M. Fastest measured: 3x DGX Spark (GB10) at 31.3 tok/s. 1 sourced result.
- [Best home rig for gpt-oss 120B](https://getrigbase.com/models/gpt-oss-120b/): What GPU do you need for gpt-oss 120B? Weights ≈65 GB at native. Fastest measured: RTX Pro 6000 (Blackwell, 96 GB) at 196.31 tok/s. 7 sourced results, from 96…
- [Best home rig for gpt-oss 20B](https://getrigbase.com/models/gpt-oss-20b/): What GPU do you need for gpt-oss 20B? Weights ≈13 GB at native. Fastest measured: RTX 5090 at 324.2 tok/s. 25 sourced results, from 16 GB cards up.
- [Best home rig for Kolibri-1](https://getrigbase.com/models/kolibri-1/): What GPU do you need for Kolibri-1? Weights ≈48 GB at Q4_K_M. Fastest measured: Strix Halo (Radeon 8060S) at 58.6 tok/s. 1 sourced result, from 128 GB cards up.
- [Best home rig for LFM2.5 24B-A2B](https://getrigbase.com/models/lfm2-5-24b-a2b/): What GPU do you need for LFM2.5 24B-A2B? Weights ≈15 GB at Q4_K_M. Fastest measured: RTX 4090 at 261-339 tok/s. 3 sourced results, from 16 GB cards up.
- [Best home rig for Ling 3.0 Flash](https://getrigbase.com/models/ling-3-0-flash/): What GPU do you need for Ling 3.0 Flash? Weights ≈76 GB at Q4_K_M. Fastest measured: DGX Spark (GB10) at 83 tok/s. 1 sourced result, from 128 GB cards up.
- [Best home rig for Llama 2 7B](https://getrigbase.com/models/llama-2-7b/): What GPU do you need for Llama 2 7B? Fastest measured: RTX 4090 at 188.96 tok/s. 2 sourced results, from 24 GB cards up.
- [Best home rig for Llama 3.1 405B](https://getrigbase.com/models/llama-3-1-405b/): What GPU do you need for Llama 3.1 405B? Fastest measured: Apple M3 Ultra at 1.96 tok/s. 2 sourced results, from 512 GB cards up.
- [Best home rig for Llama 3.1 70B](https://getrigbase.com/models/llama-3-1-70b/): What GPU do you need for Llama 3.1 70B? Fastest measured: Apple M3 Ultra at 14.08 tok/s. 7 sourced results, from 24 GB cards up.
- [Best home rig for Llama 3.1 8B](https://getrigbase.com/models/llama-3-1-8b/): What GPU do you need for Llama 3.1 8B? Weights ≈5 GB at Q4_K_M. Fastest measured: RTX Spark N1X at 56.5 tok/s. 5 sourced results, from 24 GB cards up.
- [Best home rig for Llama 3.2 3B](https://getrigbase.com/models/llama-3-2-3b/): What GPU do you need for Llama 3.2 3B? Fastest measured: RTX 4090 at 334.67 tok/s. 18 sourced results, from 11 GB cards up.
- [Best home rig for Llama 3.3 70B](https://getrigbase.com/models/llama-3-3-70b/): What GPU do you need for Llama 3.3 70B? Weights ≈43 GB at Q4_K_M. Fastest measured: 2x RTX 3090 at 20 tok/s. 5 sourced results, from 48 GB cards up.
- [Best home rig for Llama 3 70B](https://getrigbase.com/models/llama-3-70b/): What GPU do you need for Llama 3 70B? Fastest measured: 2x RTX 3090 at 21.2 tok/s. 2 sourced results, from 48 GB cards up.
- [Best home rig for Llama 3 8B](https://getrigbase.com/models/llama-3-8b/): What GPU do you need for Llama 3 8B? Fastest measured: Apple M3 Max at 50.74 tok/s. 2 sourced results, from 64 GB cards up.
- [Best home rig for Mellum 2.1 12B-A2.5B](https://getrigbase.com/models/mellum-2-1-12b-a2-5b/): What GPU do you need for Mellum 2.1 12B-A2.5B? Weights ≈7 GB at Q4_K_M. Fastest measured: RTX 4070 12 GB running Mellum 2.1 12B-A2.5B Thinking Q4_K_M at 177.5…
- [Best home rig for MiMo V2.6 Flash](https://getrigbase.com/models/mimo-v2-6-flash/): What GPU do you need for MiMo V2.6 Flash? Weights ≈189 GB at Q4_K_M. Fastest measured: 2x DGX Spark (GB10) at 70.5 tok/s. 3 sourced results, from 256 GB cards…
- [Best home rig for MiniCPM-V 4.7](https://getrigbase.com/models/minicpm-v-4-7/): What GPU do you need for MiniCPM-V 4.7? Weights ≈21 GB at Q4_K_M. Fastest measured: 2x RTX 3090 at 198.7 tok/s. 1 sourced result, from 48 GB cards up.
- [Best home rig for MiniMax M2.7](https://getrigbase.com/models/minimax-m2-7/): What GPU do you need for MiniMax M2.7? Fastest measured: Minisforum MS-S1 MAX (Ryzen AI Max+ PRO 495, 192 GB): gpt-oss 120B at 55.8 tok/s at 29.5 tok/s. 2…
- [Best home rig for MiniMax M3.1 Flash](https://getrigbase.com/models/minimax-m3-1-flash/): What GPU do you need for MiniMax M3.1 Flash? Weights ≈262 GB at Q4_K_M. Fastest measured: 2x DGX Spark (GB10) at 30.4 tok/s. 4 sourced results, from 256 GB…
- [Best home rig for MiniMax M3](https://getrigbase.com/models/minimax-m3/): What GPU do you need for MiniMax M3? Weights ≈262 GB at Q4_K_M. Fastest measured: 2x DGX Spark (GB10) at 31.8 tok/s. 2 sourced results, from 256 GB cards up.
- [Best home rig for Mistral Medium 3.5 128B](https://getrigbase.com/models/mistral-medium-3-5-128b/): What GPU do you need for Mistral Medium 3.5 128B? Weights ≈78 GB at Q4_K_M. Fastest measured: 4x Apple M5 Ultra at 23.37 tok/s. 4 sourced results, from 512 GB…
- [Best home rig for Mistral Small 22B](https://getrigbase.com/models/mistral-small-22b/): What GPU do you need for Mistral Small 22B? Fastest measured: RTX 4090 at 26.72 tok/s. 1 sourced result, from 24 GB cards up.
- [Best home rig for Muse Glimmer 30B](https://getrigbase.com/models/muse-glimmer-30b/): What GPU do you need for Muse Glimmer 30B? Fastest measured: RTX 4090 at 53.7 tok/s. 4 sourced results, from 16 GB cards up.
- [Best home rig for Muse Spark Open 816B-A42B](https://getrigbase.com/models/muse-spark-open-816b-a42b/): What GPU do you need for Muse Spark Open 816B-A42B? Weights ≈499 GB at Q4_K_M. Fastest measured: 4x Apple M5 Ultra at 36.73 tok/s. 5 sourced results, from 512…
- [Best home rig for Nanbeige4.3 33B](https://getrigbase.com/models/nanbeige4-3-33b/): What GPU do you need for Nanbeige4.3 33B? Weights ≈20 GB at Q4_K_M. Fastest measured: 2x RTX 3090 at 66.2 tok/s. 2 sourced results, from 48 GB cards up.
- [Best home rig for Nemotron 3.5 Bolt 74B-A9B](https://getrigbase.com/models/nemotron-3-5-bolt-74b-a9b/): What GPU do you need for Nemotron 3.5 Bolt 74B-A9B? Weights ≈45 GB at Q4_K_M. Fastest measured: RTX 5090 at 154.7 tok/s. 7 sourced results, from 32 GB cards up.
- [Best home rig for Nemotron 3.5 Lightning 30B-A3B](https://getrigbase.com/models/nemotron-3-5-lightning-30b-a3b/): What GPU do you need for Nemotron 3.5 Lightning 30B-A3B? Weights ≈18 GB at Q4_K_M. Fastest measured: RTX 3090 at 101.0 tok/s. 4 sourced results, from 24 GB…
- [Best home rig for Nemotron 3.5 Super VL 120B-A12B](https://getrigbase.com/models/nemotron-3-5-super-vl-120b-a12b/): What GPU do you need for Nemotron 3.5 Super VL 120B-A12B? Weights ≈73 GB at Q4_K_M. Fastest measured: DGX Spark (GB10) at 38.07 tok/s. 2 sourced results, from…
- [Best home rig for Nemotron 3 Nano Omni 30B](https://getrigbase.com/models/nemotron-3-nano-omni-30b/): What GPU do you need for Nemotron 3 Nano Omni 30B? Fastest measured: RTX Spark N1X at 63.9 tok/s. 2 sourced results, from 128 GB cards up.
- [Best home rig for Nemotron 3 Super 120B-A12B](https://getrigbase.com/models/nemotron-3-super-120b-a12b/): What GPU do you need for Nemotron 3 Super 120B-A12B? Weights ≈73 GB at Q4_K_M. Fastest measured: DGX Spark (GB10) at 23.0 tok/s. 2 sourced results, from 128…
- [Best home rig for Nex N2.5 mini](https://getrigbase.com/models/nex-n2-5-mini/): What GPU do you need for Nex N2.5 mini? Weights ≈21 GB at Q4_K_M. Fastest measured: 2x RTX 3090 at 187 tok/s. 2 sourced results, from 48 GB cards up.
- [Best home rig for Ornith 1.5 122B](https://getrigbase.com/models/ornith-1-5-122b/): What GPU do you need for Ornith 1.5 122B? Weights ≈77 GB at Q4_K_M. Fastest measured: DGX Spark (GB10) at 52.6 tok/s. 2 sourced results, from 128 GB cards up.
- [Best home rig for Ornith 1.5 35B-A3B](https://getrigbase.com/models/ornith-1-5-35b-a3b/): What GPU do you need for Ornith 1.5 35B-A3B? Weights ≈22 GB at Q4_K_M. Fastest measured: RTX 5090 at 273.2 tok/s. 6 sourced results, from 32 GB cards up.
- [Best home rig for Qwen3 235B-A22B](https://getrigbase.com/models/qwen3-235b-a22b/): What GPU do you need for Qwen3 235B-A22B? Fastest measured: 4x RTX 3090 at 18 tok/s. 2 sourced results, from 96 GB cards up.
- [Best home rig for Qwen3.5 122B](https://getrigbase.com/models/qwen3-5-122b/): What GPU do you need for Qwen3.5 122B? Weights ≈75 GB at Q4_K_M. Fastest measured: RTX Spark N1X at 35.2 tok/s. 1 sourced result.
- [Best home rig for Qwen3.5 4B](https://getrigbase.com/models/qwen3-5-4b/): What GPU do you need for Qwen3.5 4B? Weights ≈2 GB at Q4_K_M. Fastest measured: RTX 2060 Super 8 GB + Core i9-10900K: Qwen3.5 9B Q4_K_M at 54.4 tok/s, Gemma 4…
- [Best home rig for Qwen3.5 9B](https://getrigbase.com/models/qwen3-5-9b/): What GPU do you need for Qwen3.5 9B? Weights ≈5 GB at Q4_K_M. Fastest measured: RTX 5090 at 208 tok/s. 7 sourced results, from 16 GB cards up.
- [Best home rig for Qwen3.6 27B](https://getrigbase.com/models/qwen3-6-27b/): What GPU do you need for Qwen3.6 27B? Weights ≈17 GB at Q4_K_M. Fastest measured: RTX 5090 at 75.7 tok/s. 12 sourced results, from 24 GB cards up.
- [Best home rig for Qwen3.6 35B-A3B](https://getrigbase.com/models/qwen3-6-35b-a3b/): What GPU do you need for Qwen3.6 35B-A3B? Weights ≈22 GB at Q4_K_M. Fastest measured: MacBook Pro M5 Max 128 GB: Qwen3.6 35B-A3B at 271 tok/s (mlx-serve +…
- [Best home rig for Qwen3.8 27B](https://getrigbase.com/models/qwen3-8-27b/): What GPU do you need for Qwen3.8 27B? Weights ≈17 GB at Q4_K_M. Fastest measured: 2x RTX 5070 Ti at 230 tok/s. 94 sourced results, from 16 GB cards up.
- [Best home rig for Qwen3.8 Flash-Next](https://getrigbase.com/models/qwen3-8-flash-next/): What GPU do you need for Qwen3.8 Flash-Next? Weights ≈110 GB at Q4_K_M. Fastest measured: EPYC 7532 workstation (RTX Pro 6000 + RTX 3090 Ti, 300 W caps)…
- [Best home rig for Step 5 Preview](https://getrigbase.com/models/step-5-preview/): What GPU do you need for Step 5 Preview? Weights ≈367 GB at Q4_K_M. Fastest measured: 4x DGX Spark (GB10) at 25 tok/s. 1 sourced result, from 512 GB cards up.

## GPUs and systems (measured speeds per card)
- [Apple M1 Max for local LLMs](https://getrigbase.com/gpus/apple-m1-max/): Apple M1 Max local LLM benchmarks: 2 measured tokens/sec results from 2 real builds (Llama 3 8B, gpt-oss 20B…), with memory, bandwidth and EU price.
- [Apple M1 Ultra for local LLMs](https://getrigbase.com/gpus/apple-m1-ultra/): Apple M1 Ultra local LLM benchmarks: 5 measured tokens/sec results from 2 real builds (DeepSeek R1 Distill 14B, Llama 3.1 70B, Llama 3.1 8B, Llama 3.2 3B…)…
- [Apple M3 Ultra for local LLMs](https://getrigbase.com/gpus/apple-m3-ultra/): Apple M3 Ultra local LLM benchmarks: 15 measured tokens/sec results from 7 real builds (DeepSeek R1 671B, DeepSeek R1 Distill 14B, DeepSeek V4.1 Flash, Llama…
- [Apple M5 Max for local LLMs](https://getrigbase.com/gpus/apple-m5-max/): Apple M5 Max local LLM benchmarks: 18 measured tokens/sec results from 4 real builds (Gemma 4 26B-A4B, Qwen3.5 9B, Qwen3.6 35B-A3B, Qwen3.8 27B…), with…
- [Apple M5 Ultra for local LLMs](https://getrigbase.com/gpus/apple-m5-ultra/): Apple M5 Ultra local LLM benchmarks: 11 measured tokens/sec results from 4 real builds (Mistral Medium 3.5 128B, Muse Spark Open 816B-A42B, Qwen3.8 27B…
- [DGX Spark / GB10 for local LLMs](https://getrigbase.com/gpus/dgx-spark-gb10/): DGX Spark / GB10 local LLM benchmarks: 69 measured tokens/sec results from 39 real builds (DeepSeek R1 Distill 14B, DeepSeek V4 Flash, DeepSeek V4.1 Flash…
- [Intel Arc Pro B60 24GB for local LLMs](https://getrigbase.com/gpus/intel-arc-pro-b60/): Intel Arc Pro B60 24GB local LLM benchmarks: 5 measured tokens/sec results from 2 real builds (Qwen3.8 27B…), with memory, bandwidth and EU price.
- [Intel Arc Pro B70 for local LLMs](https://getrigbase.com/gpus/intel-arc-pro-b70/): Intel Arc Pro B70 local LLM benchmarks: 8 measured tokens/sec results from 3 real builds (Gemma 4 26B-A4B, Ornith 1.5 35B-A3B, Qwen3.6 35B-A3B, Qwen3.8 27B…)…
- [Radeon AI Pro R9700 for local LLMs](https://getrigbase.com/gpus/radeon-ai-pro-r9700/): Radeon AI Pro R9700 local LLM benchmarks: 32 measured tokens/sec results from 7 real builds (DeepSeek R1 Distill 14B, Gemma 4 26B-A4B, Gemma 4 31B, Llama 3.2…
- [Radeon RX 7900 XTX for local LLMs](https://getrigbase.com/gpus/radeon-rx-7900-xtx/): Radeon RX 7900 XTX local LLM benchmarks: 6 measured tokens/sec results from 1 real builds (Qwen3.6 35B-A3B…), with memory, bandwidth and EU price.
- [RTX 3060 12GB for local LLMs](https://getrigbase.com/gpus/rtx-3060-12gb/): RTX 3060 12GB local LLM benchmarks: 10 measured tokens/sec results from 3 real builds (DeepSeek R1 Distill 14B, Gemma 4 12B, LFM2.5 24B-A2B, Llama 3.2 3B…)…
- [RTX 3090 for local LLMs](https://getrigbase.com/gpus/rtx-3090/): RTX 3090 local LLM benchmarks: 61 measured tokens/sec results from 34 real builds (DeepSeek R1 671B, DeepSeek V4 Flash, DeepSeek V4.1 Flash, Flint…
- [RTX 4060 Ti 16GB for local LLMs](https://getrigbase.com/gpus/rtx-4060-ti-16gb/): RTX 4060 Ti 16GB local LLM benchmarks: 4 measured tokens/sec results from 3 real builds (LFM2.5 24B-A2B, Llama 3.3 70B, Qwen3.8 27B…), with memory, bandwidth…
- [RTX 4090 for local LLMs](https://getrigbase.com/gpus/rtx-4090/): RTX 4090 local LLM benchmarks: 28 measured tokens/sec results from 11 real builds (DeepSeek R1 Distill 14B, Flint 40B-A2.5B, GLM 4.7 Flash, Gemma 4 12B…)…
- [RTX 5060 Ti 16GB for local LLMs](https://getrigbase.com/gpus/rtx-5060-ti-16gb/): RTX 5060 Ti 16GB local LLM benchmarks: 8 measured tokens/sec results from 3 real builds (Qwen3.8 27B, gpt-oss 20B…), with memory, bandwidth and EU price.
- [RTX 5070 Ti for local LLMs](https://getrigbase.com/gpus/rtx-5070-ti/): RTX 5070 Ti local LLM benchmarks: 6 measured tokens/sec results from 3 real builds (Gemma 4 12B, Qwen3.8 27B, Qwen3.8 Flash-Next, gpt-oss 20B…), with memory…
- [RTX 5080 for local LLMs](https://getrigbase.com/gpus/rtx-5080/): RTX 5080 local LLM benchmarks: 3 measured tokens/sec results from 2 real builds (Qwen3.8 Flash-Next, gpt-oss 20B…), with memory, bandwidth and EU price.
- [RTX 5090 for local LLMs](https://getrigbase.com/gpus/rtx-5090/): RTX 5090 local LLM benchmarks: 31 measured tokens/sec results from 12 real builds (DeepSeek V4.1 Flash, Flint 40B-A2.5B, GLM-5.3 Flash, Gemma 4 12B…), with…
- [RTX A4000 for local LLMs](https://getrigbase.com/gpus/rtx-a4000/): RTX A4000 local LLM benchmarks: 3 measured tokens/sec results from 2 real builds (Llama 3.2 3B, Muse Glimmer 30B, gpt-oss 20B…), with memory, bandwidth and EU…
- [RTX Pro 6000 Blackwell for local LLMs](https://getrigbase.com/gpus/rtx-pro-6000/): RTX Pro 6000 Blackwell local LLM benchmarks: 11 measured tokens/sec results from 5 real builds (GLM-5.3 Flash, Qwen3.8 27B, Qwen3.8 Flash-Next, gpt-oss…
- [RTX Spark N1X for local LLMs](https://getrigbase.com/gpus/rtx-spark-n1x/): RTX Spark N1X local LLM benchmarks: 8 measured tokens/sec results from 2 real builds (GLM 4.7 Flash, Llama 3.1 8B, Muse Glimmer 30B, Nemotron 3 Nano Omni…
- [Strix Halo (Ryzen AI Max+ 395, Radeon 8060S) for local LLMs](https://getrigbase.com/gpus/strix-halo/): Strix Halo (Ryzen AI Max+ 395, Radeon 8060S) local LLM benchmarks: 44 measured tokens/sec results from 13 real builds (DeepSeek R1 Distill 70B, DeepSeek R1…
- [Tesla V100 32GB for local LLMs](https://getrigbase.com/gpus/tesla-v100-32gb/): Tesla V100 32GB local LLM benchmarks: 5 measured tokens/sec results from 1 real builds (DeepSeek V4.1 Flash, Qwen3.8 27B, Qwen3.8 Flash-Next…), with memory…

## How much VRAM per model
- [How much VRAM do you need for each local LLM?](https://getrigbase.com/vram/): How much VRAM each current local LLM needs: Q4_K_M weight math and measured tokens/sec from real home AI builds.
- [How much VRAM do you need for DeepSeek V4.1 Flash?](https://getrigbase.com/vram/deepseek-v4-1-flash/): How much VRAM for DeepSeek V4.1 Flash? Weights ≈511 GB at native; 12 measured tok/s results, fastest 31.6 tok/s.
- [How much VRAM do you need for Flint 40B-A2.5B?](https://getrigbase.com/vram/flint-40b-a2-5b/): How much VRAM for Flint 40B-A2.5B? Weights ≈24 GB at Q4_K_M; 8 measured tok/s results, fastest 224.9 tok/s.
- [How much VRAM do you need for Gemma 4 12B?](https://getrigbase.com/vram/gemma-4-12b/): How much VRAM for Gemma 4 12B? Weights ≈7 GB at Q4_K_M; 8 measured tok/s results, fastest 142 tok/s.
- [How much VRAM do you need for Gemma 4 26B-A4B?](https://getrigbase.com/vram/gemma-4-26b-a4b/): How much VRAM for Gemma 4 26B-A4B? Weights ≈15 GB at Q4_K_M; 10 measured tok/s results, fastest 235 tok/s.
- [How much VRAM do you need for Gemma 4 31B?](https://getrigbase.com/vram/gemma-4-31b/): How much VRAM for Gemma 4 31B? Weights ≈19 GB at Q4_K_M; 9 measured tok/s results, fastest 71.3 tok/s.
- [How much VRAM do you need for GLM-5.3 Flash?](https://getrigbase.com/vram/glm-5-3-flash/): How much VRAM for GLM-5.3 Flash? Weights ≈196 GB at Q4_K_M; 17 measured tok/s results, fastest 105.3 tok/s.
- [How much VRAM do you need for gpt-oss 120B?](https://getrigbase.com/vram/gpt-oss-120b/): How much VRAM for gpt-oss 120B? Weights ≈65 GB at native; 7 measured tok/s results, fastest 196.31 tok/s.
- [How much VRAM do you need for gpt-oss 20B?](https://getrigbase.com/vram/gpt-oss-20b/): How much VRAM for gpt-oss 20B? Weights ≈13 GB at native; 25 measured tok/s results, fastest 324.2 tok/s.
- [How much VRAM do you need for MiniMax M3.1 Flash?](https://getrigbase.com/vram/minimax-m3-1-flash/): How much VRAM for MiniMax M3.1 Flash? Weights ≈262 GB at Q4_K_M; 4 measured tok/s results, fastest 25.3 tok/s.
- [How much VRAM do you need for Mistral Medium 3.5 128B?](https://getrigbase.com/vram/mistral-medium-3-5-128b/): How much VRAM for Mistral Medium 3.5 128B? Weights ≈78 GB at Q4_K_M; 4 measured tok/s results, fastest 23.37 tok/s.
- [How much VRAM do you need for Muse Spark Open 816B-A42B?](https://getrigbase.com/vram/muse-spark-open-816b-a42b/): How much VRAM for Muse Spark Open 816B-A42B? Weights ≈499 GB at Q4_K_M; 5 measured tok/s results, fastest 36.73 tok/s.
- [How much VRAM do you need for Nemotron 3.5 Bolt 74B-A9B?](https://getrigbase.com/vram/nemotron-3-5-bolt-74b-a9b/): How much VRAM for Nemotron 3.5 Bolt 74B-A9B? Weights ≈45 GB at Q4_K_M; 7 measured tok/s results, fastest 154.7 tok/s.
- [How much VRAM do you need for Nemotron 3.5 Lightning 30B-A3B?](https://getrigbase.com/vram/nemotron-3-5-lightning-30b-a3b/): How much VRAM for Nemotron 3.5 Lightning 30B-A3B? Weights ≈18 GB at Q4_K_M; 4 measured tok/s results, fastest 101.0 tok/s.
- [How much VRAM do you need for Ornith 1.5 35B-A3B?](https://getrigbase.com/vram/ornith-1-5-35b-a3b/): How much VRAM for Ornith 1.5 35B-A3B? Weights ≈22 GB at Q4_K_M; 6 measured tok/s results, fastest 273.2 tok/s.
- [How much VRAM do you need for Qwen3.5 9B?](https://getrigbase.com/vram/qwen3-5-9b/): How much VRAM for Qwen3.5 9B? Weights ≈5 GB at Q4_K_M; 7 measured tok/s results, fastest 208 tok/s.
- [How much VRAM do you need for Qwen3.6 35B-A3B?](https://getrigbase.com/vram/qwen3-6-35b-a3b/): How much VRAM for Qwen3.6 35B-A3B? Weights ≈22 GB at Q4_K_M; 49 measured tok/s results, fastest 258 tok/s.
- [How much VRAM do you need for Qwen3.8 27B?](https://getrigbase.com/vram/qwen3-8-27b/): How much VRAM for Qwen3.8 27B? Weights ≈17 GB at Q4_K_M; 94 measured tok/s results, fastest 75.2 tok/s.
- [How much VRAM do you need for Qwen3.8 Flash-Next?](https://getrigbase.com/vram/qwen3-8-flash-next/): How much VRAM for Qwen3.8 Flash-Next? Weights ≈110 GB at Q4_K_M; 48 measured tok/s results, fastest 182 tok/s.

## Guides
- [AMD Radeon for local LLMs: measured speeds](https://getrigbase.com/guides/amd-rdna-local-llm/): Measured local LLM speeds on AMD Radeon AI Pro R9700, RX 7900 XTX, RX 7900 XT and Strix Halo (Radeon 8060S): llama.cpp Vulkan vs ROCm, MTP speculative…
- [Best GPU for local LLMs in 2026](https://getrigbase.com/guides/best-gpu-for-local-llm-2026/): Best GPU for running local LLMs in 2026, ranked with measured tokens/sec on Qwen3.6, Gemma 4 and gpt-oss, plus VRAM, bandwidth and current prices.
- [Best GPU for local LLMs under €1,000](https://getrigbase.com/guides/best-gpu-for-local-llm-under-1000-eur/): Best GPU for local LLMs under €1,000 (2026): RTX 3060, RTX 5060 Ti 16GB, RX 7900 XT with measured tok/s, German prices and what fits in 12–20 GB.
- [Best value GPU for local LLMs in Europe (tok/s per euro)](https://getrigbase.com/guides/best-value-gpu-local-llm-europe/): Best value GPU for local LLMs in Europe: measured tokens/sec per euro for gpt-oss, Gemma 4 and Qwen on RTX 5060 Ti, 5070 Ti, 5080, 4090, 5090 and more, at…
- [GGUF vs AWQ vs FP8 vs NVFP4: which format for local LLMs?](https://getrigbase.com/guides/gguf-vs-awq-vs-fp8-vs-nvfp4/): GGUF vs AWQ vs FP8 vs NVFP4 for local LLMs: when to use each format, with measured RigBase examples on Qwen3.6 and Flint.
- [How much system RAM do you need for local LLMs?](https://getrigbase.com/guides/how-much-system-ram-for-local-llms/): How much system RAM for local LLMs? Often 32 GB is enough when the model fits in VRAM — multi-GPU 3090 examples and when offload needs more.
- [Intel Arc Pro B70 for local LLMs: measured results](https://getrigbase.com/guides/intel-arc-pro-b70-local-llm/): Intel Arc Pro B70 (32 GB) local LLM benchmarks: 8 measured tok/s results on one and two cards with llama.cpp SYCL and vLLM XPU, plus what fits in 32/64 GB.
- [Mac vs PC for local AI](https://getrigbase.com/guides/mac-vs-pc-local-ai/): Mac vs PC for running local LLMs: measured tokens/sec on gpt-oss, DeepSeek R1 and 70B models, Apple unified memory vs NVIDIA GPUs.
- [Multi-GPU RTX 3090 for local LLMs: measured 2× / 4× / 6× / 8× setups](https://getrigbase.com/guides/multi-gpu-rtx-3090-local-llm/): Multi-GPU RTX 3090 local LLM guide: measured tokens/sec on 2×, 4×, 6× and 8× 3090 builds for Qwen3.6, Qwen3.8 Flash-Next and DeepSeek.
- [Quantization for local LLMs: is Q4 good enough?](https://getrigbase.com/guides/quantization-is-q4-good-enough/): Is Q4 quantization good enough for local LLMs? Why RigBase uses Q4_K_M as the default fit estimate and when to use Q5/Q8 or avoid Q3/Q2.
- [Used RTX 3090 buying guide for local AI](https://getrigbase.com/guides/used-rtx-3090-buying-guide/): Used RTX 3090 buying guide for local LLMs in 2026: measured speeds on Qwen3.6 and Qwen3.8, what to check before buying, power tips.
- [What GPU do you need for Qwen3.8, Qwen3.6, Gemma 4 and gpt-oss locally? (measured, Oct 2026)](https://getrigbase.com/guides/what-gpu-for-qwen3-6-gemma-4-gpt-oss/): Measured tokens/sec for Qwen3.8 27B, Qwen3.6, Gemma 4 and gpt-oss on 117 real home rigs: RTX 3060 to RTX Pro 6000, Macs, DGX Spark, Strix Halo, Arc B70.
- [Beste GPU für lokale LLMs unter 1.000 €](https://getrigbase.com/de/guides/beste-gpu-fuer-lokale-llm-unter-1000-euro/): Beste GPU für lokale LLMs unter 1.000 € (2026): gemessene Token/s, deutsche Neupreise von billiger.de und was in 12–20 GB VRAM passt.
- [Gebrauchte RTX 3090 für lokale KI: Kaufberatung 2026](https://getrigbase.com/de/guides/gebrauchte-rtx-3090-kaufberatung/): Gebrauchte RTX 3090 für lokale LLMs 2026: Median 1.248 € gebraucht, gemessene tok/s mit aktuellen Modellen, Checkliste vor dem Kauf, eine oder zwei Karten.
- [Quantisierung für lokale LLMs: ist Q4 gut genug?](https://getrigbase.com/de/guides/quantisierung-ist-q4-gut-genug/): Ist Q4-Quantisierung gut genug für lokale LLMs? Warum RigBase Q4_K_M als Standard-Fit nutzt und wann Q5/Q8 oder kein Q3/Q2.
- [Wie viel System-RAM brauche ich für lokale LLMs?](https://getrigbase.com/de/guides/wie-viel-system-ram-fuer-lokale-llms/): Wie viel System-RAM für lokale LLMs? Oft reichen 32 GB, wenn das Modell in den VRAM passt — Multi-GPU-3090-Beispiele und wann Offload mehr braucht.

## Comparisons
- [DGX Spark vs Mac Studio vs Strix Halo for local LLMs: measured tok/s and tok/s per watt](https://getrigbase.com/compare/dgx-spark-vs-mac-studio-vs-strix-halo/): DGX Spark vs Mac Studio vs Strix Halo for local LLMs: measured tokens/sec on Llama, gpt-oss, Qwen3.6 and Qwen3.8, plus tok/s per watt from measured system…
- [DGX Spark vs RTX PRO 6000 vs Mac Studio vs Strix Halo for local LLMs](https://getrigbase.com/compare/dgx-spark-vs-rtx-pro-6000-vs-mac-studio-vs-strix-halo/): DGX Spark vs RTX PRO 6000 vs Mac Studio vs Strix Halo for local LLMs: measured tokens/sec on Qwen3.8, gpt-oss, GLM-5.3 Flash and more.
- [RTX 3090 vs RTX 4090 for local LLMs](https://getrigbase.com/compare/rtx-3090-vs-rtx-4090-llm/): RTX 3090 vs RTX 4090 for local LLMs: measured tokens/sec on Qwen3.6, Gemma 4 and gpt-oss, VRAM, bandwidth and price.
- [Used RTX 3090 vs RTX 5090 for local LLMs](https://getrigbase.com/compare/rtx-3090-vs-rtx-5090-llm/): Used RTX 3090 (≈€1,248) vs RTX 5090 (≈€7,140) for local LLMs: measured tok/s on 5 2026 models, tok/s per euro, 24 vs 32 GB VRAM.
- [RTX 4090 vs RTX 5090 for local LLMs](https://getrigbase.com/compare/rtx-4090-vs-rtx-5090-llm/): RTX 4090 vs RTX 5090 for local LLMs: measured tokens/sec on Qwen3.6, Gemma 4 and gpt-oss, VRAM, bandwidth and price.

## Best models by VRAM / budget
- [Best local LLMs for 128 GB or more of VRAM or unified memory](https://getrigbase.com/best/128gb-plus-vram/): Which local LLMs run on 128 GB or more of VRAM or unified memory: 148 measured tokens/sec results, fastest per model, plus what fits by memory math.
- [Best local LLMs for 12 GB of VRAM or unified memory](https://getrigbase.com/best/12gb-vram/): Which local LLMs run on 12 GB of VRAM or unified memory: 16 measured tokens/sec results, fastest per model, plus what fits by memory math.
- [Best local LLMs for 16 GB of VRAM or unified memory](https://getrigbase.com/best/16gb-vram/): Which local LLMs run on 16 GB of VRAM or unified memory: 23 measured tokens/sec results, fastest per model, plus what fits by memory math.
- [Best local LLMs for 24 GB of VRAM or unified memory](https://getrigbase.com/best/24gb-vram/): Which local LLMs run on 24 GB of VRAM or unified memory: 71 measured tokens/sec results, fastest per model, plus what fits by memory math.
- [Best local LLMs for 32 GB of VRAM or unified memory](https://getrigbase.com/best/32gb-vram/): Which local LLMs run on 32 GB of VRAM or unified memory: 73 measured tokens/sec results, fastest per model, plus what fits by memory math.
- [Best local LLMs for 48 GB of VRAM or unified memory](https://getrigbase.com/best/48gb-vram/): Which local LLMs run on 48 GB of VRAM or unified memory: 34 measured tokens/sec results, fastest per model, plus what fits by memory math.
- [Best local LLMs for 64 GB of VRAM or unified memory](https://getrigbase.com/best/64gb-vram/): Which local LLMs run on 64 GB of VRAM or unified memory: 15 measured tokens/sec results, fastest per model, plus what fits by memory math.
- [Best local LLMs for 8 GB of VRAM or unified memory](https://getrigbase.com/best/8gb-vram/): Which local LLMs run on 8 GB of VRAM or unified memory: 3 measured tokens/sec results, fastest per model, plus what fits by memory math.
- [Best local LLMs for 96 GB of VRAM or unified memory](https://getrigbase.com/best/96gb-vram/): Which local LLMs run on 96 GB of VRAM or unified memory: 11 measured tokens/sec results, fastest per model, plus what fits by memory math.
- [Best home AI rig under €1,000](https://getrigbase.com/budget/under-1000-eur/): Home AI / local LLM rigs under €1,000: real builds with reported cost and measured tokens/sec, plus the best GPU under €1,000 at German prices.
- [Best home AI rig under €2,000](https://getrigbase.com/budget/under-2000-eur/): Best home AI / local LLM rigs under €2,000: real builds with reported cost and measured tokens/sec.
- [Best home AI rig under €5,000](https://getrigbase.com/budget/under-5000-eur/): Best home AI / local LLM rigs under €5,000: real builds with reported cost and measured tokens/sec.

## Optional
- [Can I run X on Y: model × hardware answers](https://getrigbase.com/can-i-run/): 154 measured answers to 'can my GPU run this model': Qwen3.6, Gemma 4, gpt-oss and more on RTX, Radeon, Mac and Strix Halo hardware.
- [All 187 builds](https://getrigbase.com/builds/): All home AI / local LLM builds in RigBase: GPUs, memory and measured tokens/sec, each linked to its source.
- [Free public MCP endpoint](https://getrigbase.com/mcp-server/): Free public MCP endpoint at getrigbase.com/mcp: search 187 measured local-LLM builds, can-it-run checks, best GPU per model. Works with Claude, Cursor…
- [Deutsch: RigBase auf Deutsch](https://getrigbase.com/de/): Läuft Qwen3.8, Gemma 4 oder gpt-oss lokal? Gemessene Token/s von 187 echten KI-Rechnern mit Quelle, VRAM und deutschen Grafikkartenpreisen.
- [Markdown versions](https://getrigbase.com/models/): most key pages have a markdown mirror at <page>/index.md