LOCAL AI COMPATIBILITY

AI Rig Match

Match your GPU or Apple silicon Mac with local language, coding, image, and video models.

Capacity and speed are modeled estimates, not a guarantee. Confirm runtime, driver, and model-version requirements before buying hardware.

YOUR RIG

Build a hardware profile

Set the memory budget and workload that matter for this run.

89 hardware presets · 92 models · verified 2026-08-23

MATCH REPORT

Compatible models

Results separate native VRAM fit from slower system-memory offload.

48Native

3Offload

8Upgrade

Sort models

Qwen3 0.6B

0.6Bllama.cpp

Qwen · Confidence: medium

Native VRAM Fit

Model memory0.9 / 29.4 GB

Estimated speed: 452.9–685.2 tok/s
Quantization: Q4
Context: 8K
Minimum RAM: 16 GB

llama-cli -m ./models/qwen3-0.6b-q4_K_M.gguf

Llama 3.2 1B

1Bllama.cpp

Llama · Confidence: medium

Native VRAM Fit

Model memory1.3 / 29.4 GB

Estimated speed: 423–640 tok/s
Quantization: Q4
Context: 8K
Minimum RAM: 16 GB

llama-cli -m ./models/llama-3-2-1b-q4_K_M.gguf

Gemma 3 1B

1Bllama.cpp

Gemma · Confidence: medium

Native VRAM Fit

Model memory1.3 / 29.4 GB

Estimated speed: 423–640 tok/s
Quantization: Q4
Context: 8K
Minimum RAM: 16 GB

llama-cli -m ./models/gemma-3-1b-q4_K_M.gguf

Qwen3 1.7B

1.7Bllama.cpp

Qwen · Confidence: medium

Native VRAM Fit

Model memory1.5 / 29.4 GB

Estimated speed: 355.1–537.2 tok/s
Quantization: Q4
Context: 8K
Minimum RAM: 16 GB

llama-cli -m ./models/qwen3-1.7b-q4_K_M.gguf

Llama 3.2 3B

3Bllama.cpp

Llama · Confidence: medium

Native VRAM Fit

Model memory2.4 / 29.4 GB

Estimated speed: 248.2–375.5 tok/s
Quantization: Q4
Context: 8K
Minimum RAM: 16 GB

llama-cli -m ./models/llama-3-2-3b-q4_K_M.gguf

SmolLM3 3B

3Bllama.cpp

Hugging Face · Confidence: medium

Native VRAM Fit

Model memory2.4 / 29.4 GB

Estimated speed: 256.4–387.8 tok/s
Quantization: Q4
Context: 8K
Minimum RAM: 16 GB

llama-cli -m ./models/smollm3-3b-q4_K_M.gguf

Ministral 3B

3Bllama.cpp

Mistral · Confidence: medium

Native VRAM Fit

Model memory2.6 / 29.4 GB

Estimated speed: 241–364.5 tok/s
Quantization: Q4
Context: 8K
Minimum RAM: 16 GB

llama-cli -m ./models/ministral-3b-q4_K_M.gguf

Phi-4 Mini

3.8Bllama.cpp

Microsoft · Confidence: medium

Native VRAM Fit

Model memory2.9 / 29.4 GB

Estimated speed: 221.9–335.8 tok/s
Quantization: Q4
Context: 8K
Minimum RAM: 16 GB

llama-cli -m ./models/phi-4-mini-q4_K_M.gguf

Phi-3.5 Mini

3.8Bllama.cpp

Microsoft · Confidence: medium

Native VRAM Fit

Model memory3 / 29.4 GB

Estimated speed: 215.6–326.2 tok/s
Quantization: Q4
Context: 8K
Minimum RAM: 16 GB

llama-cli -m ./models/phi-3.5-mini-q4_K_M.gguf

Qwen3 4B

4Bllama.cpp

Qwen · Confidence: medium

Native VRAM Fit

Model memory3.3 / 29.4 GB

Estimated speed: 200.2–302.9 tok/s
Quantization: Q4
Context: 8K
Minimum RAM: 16 GB

llama-cli -m ./models/qwen3-4b-q4_K_M.gguf

Gemma 3 4B

4Bllama.cpp

Gemma · Confidence: medium

Native VRAM Fit

Model memory3.5 / 29.4 GB

Estimated speed: 192–290.5 tok/s
Quantization: Q4
Context: 8K
Minimum RAM: 16 GB

llama-cli -m ./models/gemma-3-4b-q4_K_M.gguf

Gemma 3n E2B

5B / 2B activellama.cpp

Gemma · Confidence: medium

Native VRAM Fit

Model memory4 / 29.4 GB

Estimated speed: 173–261.7 tok/s
Quantization: Q4
Context: 8K
Minimum RAM: 16 GB

llama-cli -m ./models/gemma-3n-e2b-q4_K_M.gguf

Ranges include runtime overhead and an OS reserve. Real performance varies by model build, backend, driver, thermals, and prompt.

CALCULATION METHOD

Why Our Fit Estimates Are More Accurate

We treat weights, KV-cache context, runtime overhead, and memory topology as distinct constraints instead of using crude formulas.

Real Footprint vs. Marketing Labels

Each estimate starts from actual quantized weight sizes and adds runtime framework overhead, rather than simply equating raw parameter count with required VRAM.

Context (KV Cache) Costs Real Memory

Longer conversations significantly increase KV-cache pressure. A model that fits in VRAM for short chats may spill into RAM during long-context tasks.

VRAM Fit vs. CPU Offload

Offloading layers to system RAM lets models launch, but runs at much lower speeds. We clearly separate true native GPU execution from hybrid offloading.

COMMON QUESTIONS

Local AI hardware questions

How much VRAM does an 8B language model need?

A Q4 build commonly needs about 5–7 GB after basic runtime overhead. Longer context, a larger batch, or higher precision can push it beyond 8 GB.

Is Apple unified memory the same as GPU VRAM?

No. Apple silicon shares one memory pool between the CPU and GPU. That flexibility helps larger models fit, but macOS and other processes still need a reserve.

What does offload mean?

Part of the model stays in system RAM and moves through the CPU or PCIe path. It can make a model launch, but it is usually much slower than keeping all active data in GPU memory.

Are the speed ranges benchmark results?

They are conservative modeled ranges, not a benchmark claim. Exact speed depends on runtime, kernel support, model build, driver, cooling, and workload.