Rent a GPU

Can the Apple M4 Pro (24GB) run Llama 3.2 3B?

Apple M4 Pro (24GB) × Llama 3.2 3B8K context
YesRuns fully on the GPU
Memory at FP168 GB of 16.1 GB
Weights 6.4 GBKV cache 0.9 GBRuntime 0.6 GB
QuantFP16Full quality
Speed~26tokens per second
Max context76Ktokens that fit at FP16
  1. FP16
  2. Q8_0
  3. Q6_K
  4. Q5_K_M
  5. Q4_K_M
  6. Q3_K_M
  7. Q2_K

Yes. The Apple M4 Pro (24GB) can run Llama 3.2 3B fully in unified memory at FP16 (full quality), using about 8 GB of the 16.1 GB available, at roughly 26 tokens/second.

Try other context lengths

Run it with Ollama

With Ollama installed, one command downloads and starts Llama 3.2 3B:

ollama run llama3.2:3b

Ollama's default download is Q4_K_M (about 3.5 GB here). You have room for FP16, which is full quality: pick that GGUF in LM Studio or from Hugging Face.

New to this? Step-by-step setup for the Apple M4 Pro (24GB)

Run it in the cloud

No Apple M4 Pro (24GB)? Run Llama 3.2 3B in the cloud.

Rent a cloud GPU by the hour. No upfront cost, stop any time.

Rent a GPU on RunPodRent a GPU on Vast.ai

Memory needed by quantization

At 8,192 tokens of context. The black line marks the 16.1 GB the Apple M4 Pro (24GB) can use. macOS lets the GPU use about 16.1 GB of the 24 GB unified memory by default.

QuantQualityMemoryFits?Speed
FP16Full quality8 GBFits~26 tok/s
Q8_0Near-lossless5 GBFits~48 tok/s
Q6_KExcellent4.2 GBFits~62 tok/s
Q5_K_MVery good3.8 GBFits~72 tok/s
Q4_K_MGood, most popular3.5 GBFits~84 tok/s
Q3_K_MNoticeably degraded3.1 GBFits~104 tok/s
Q2_KHeavily degraded2.7 GBFits~136 tok/s

Measured speed from real Apple M4 Pro (24GB) owners

Tokens per second that people measured on their own Apple M4 Pro (24GB), next to our estimate.

No one has measured Llama 3.2 3B on the Apple M4 Pro (24GB) yet. The speeds above are estimates. Run the benchmark and yours becomes the first real number on this page.

Have an Apple M4 Pro (24GB)? Add your real speed

One command times Llama 3.2 3B in your own Ollama, shows you exactly what it would send, and asks before sending anything. No sign-up, no personal data.

macOS and Linux

curl -fsSL https://gpufit.dev/bench.sh | sh -s -- llama3.2:3b

Windows (PowerShell)

$env:GPUFIT_MODEL="llama3.2:3b"; irm https://gpufit.dev/bench.ps1 | iex

How the benchmark works

GPUs that run Llama 3.2 3B comfortably

Models that run well on the Apple M4 Pro (24GB)

Frequently asked questions

Can Apple M4 Pro (24GB) run Llama 3.2 3B?

Yes. The Apple M4 Pro (24GB) can run Llama 3.2 3B fully in unified memory at FP16 (full quality), using about 8 GB of the 16.1 GB available, at roughly 26 tokens/second.

How much VRAM does Llama 3.2 3B need?

About 3.5 GB at Q4_K_M with 8,192 tokens of context, and about 8 GB at full FP16 precision.

How fast is Llama 3.2 3B on the Apple M4 Pro (24GB)?

Roughly 26 tokens per second at FP16, estimated from the Apple M4 Pro (24GB)'s 273 GB/s memory bandwidth.