Rent a GPU

Can the Apple M3 Pro (36GB) run Phi-4 Reasoning 14B?

Apple M3 Pro (36GB) × Phi-4 Reasoning 14B8K context
YesRuns fully on the GPU
Memory at Q8_017.9 GB of 24.1 GB
Weights 15.6 GBKV cache 1.7 GBRuntime 0.6 GB
QuantQ8_0Near-lossless
Speed~6tokens per second
Max context32Kthe model's full window
  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 M3 Pro (36GB) can run Phi-4 Reasoning 14B fully in unified memory at Q8_0 (near-lossless), using about 17.9 GB of the 24.1 GB available, at roughly 6 tokens/second.

Try other context lengths

Run it with Ollama

With Ollama installed, one command downloads and starts Phi-4 Reasoning 14B:

ollama run phi4-reasoning:14b

Ollama's default download is Q4_K_M (about 11.2 GB here). You have room for Q8_0, which is near-lossless: pick that GGUF in LM Studio or from Hugging Face.

New to this? Step-by-step setup for the Apple M3 Pro (36GB)

Run it in the cloud

No Apple M3 Pro (36GB)? Run Phi-4 Reasoning 14B 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 24.1 GB the Apple M3 Pro (36GB) can use. macOS lets the GPU use about 24.1 GB of the 36 GB unified memory by default.

QuantQualityMemoryFits?Speed
FP16Full quality31.7 GBToo big—
Q8_0Near-lossless17.9 GBFits~6 tok/s
Q6_KExcellent14.3 GBFits~7 tok/s
Q5_K_MVery good12.7 GBFits~9 tok/s
Q4_K_MGood, most popular11.2 GBFits~10 tok/s
Q3_K_MNoticeably degraded9.5 GBFits~13 tok/s
Q2_KHeavily degraded7.8 GBFits~16 tok/s

Measured speed from real Apple M3 Pro (36GB) owners

Tokens per second that people measured on their own Apple M3 Pro (36GB), next to our estimate.

No one has measured Phi-4 Reasoning 14B on the Apple M3 Pro (36GB) yet. The speeds above are estimates. Run the benchmark and yours becomes the first real number on this page.

Have an Apple M3 Pro (36GB)? Add your real speed

One command times Phi-4 Reasoning 14B 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 -- phi4-reasoning:14b

Windows (PowerShell)

$env:GPUFIT_MODEL="phi4-reasoning:14b"; irm https://gpufit.dev/bench.ps1 | iex

How the benchmark works

GPUs that run Phi-4 Reasoning 14B comfortably

Models that run well on the Apple M3 Pro (36GB)

Frequently asked questions

Can Apple M3 Pro (36GB) run Phi-4 Reasoning 14B?

Yes. The Apple M3 Pro (36GB) can run Phi-4 Reasoning 14B fully in unified memory at Q8_0 (near-lossless), using about 17.9 GB of the 24.1 GB available, at roughly 6 tokens/second.

How much VRAM does Phi-4 Reasoning 14B need?

About 11.2 GB at Q4_K_M with 8,192 tokens of context, and about 31.7 GB at full FP16 precision.

How fast is Phi-4 Reasoning 14B on the Apple M3 Pro (36GB)?

Roughly 6 tokens per second at Q8_0, estimated from the Apple M3 Pro (36GB)'s 150 GB/s memory bandwidth.