Is unified memory on MacBook Air M3 24GB as fast as VRAM for Granite Code 20B?

Not always. MacBook Air M3 24GB can often fit larger models thanks to unified memory, but a discrete GPU with dedicated high-bandwidth VRAM may still decode faster once the model fits. For this combination, the important distinction is capacity versus sustained throughput.

Can Granite Code 20B run on MacBook Air M3 24GB?

Q: What should I upgrade first if Granite Code 20B feels slow on MacBook Air M3 24GB?

Remove offload with more accelerator memory. Prioritize a GPU or unified-memory tier that fits the whole model natively. Removing offload usually helps more than small compute gains.

BARELY — Tight on Memory

B64Good

Estimated from fit model

Granite Code 20B needs ~18.9 GB VRAM. MacBook Air M3 24GB has 17.3 GB. With Q4_K_M quantization, expect ~5 tok/s.

Runtime: llama.cppCapacity: OffloadBandwidth: Very lowStack: StandardBottleneck: Host offload

Operating mode

Choose the run profile you care about

Interactive favors responsiveness, while light API and scale-out lean harder on serving readiness. The fit stays the same, but the recommendation lens changes.

Current mode

Balanced

Balanced for general local use. Keeps the ranking neutral across personal and serving workflows.

Capabilities:

Select quantization to explore

Q4_K_M (Medium quality) — 18.9 GB, 5.2 tok/s, Very compromised (needs ~1 GB host RAM)

18.9 GB required17.3 GB available

109% VRAM needed

1.6 GB over capacity — needs offload or smaller quantization

Fit status

Very compromised (needs ~1 GB host RAM)

Decode

5.2 tok/s

TTFT

37307 ms

Safe context

Memory

18.9 GB / 17.3 GB

Offload

10%

Memory breakdown

Weights12.2 GB

KV Cache3.2 GB

Runtime0.9 GB

Headroom2.6 GB

See how fast it feels

See how fast it feelsGranite Code 20B on MacBook Air M3 24GB
1st promptCold start — includes initialization
>Write a Python function to sort a list
Here's a Python function that sorts a list using quicksort:

def quicksort(arr):
    if len(arr) <= 1:
        return arr
    pivot = arr[len(arr) // 2]
    left   = [x for x in arr if x < pivot]
    middle = [x for x in arr if x == pivot]
    right  = [x for x in arr if x > pivot]
    return quicksort(left) + middle + quicksort(right)

# Example usage
numbers = [3, 6, 8, 10, 1, 2, 1]
print(quicksort(numbers))  # [1, 1, 2, 3, 6, 8, 10]

The algorithm picks a pivot, partitions the list into smaller, equal, and larger elements, then recursively sorts each partition and concatenates the results.
2nd promptWarm — model ready, no init overhead
>Now add type hints and a docstring
def quicksort(arr: list[int]) -> list[int]:
    """Sort a list using the quicksort algorithm.

    Args:
        arr: List of integers to sort.

    Returns:
        A new sorted list.
    """
    if len(arr) <= 1:
        return arr
    pivot = arr[len(arr) // 2]
    left   = [x for x in arr if x < pivot]
    middle = [x for x in arr if x == pivot]
    right  = [x for x in arr if x > pivot]
    return quicksort(left) + middle + quicksort(right)

Added type hints for the input and return type, plus a Google-style docstring.
Estimated: 5.2 tok/s decode · 37.3s TTFT (warm) · 13 tok/s prefill

What limits this setup

It fits through host-memory offload, and offload is the main reason performance drops.

CPU or host-memory offload is active

About 10% of the working set spills out of accelerator memory, which usually hurts latency and sustained decode throughput.

Very little memory headroom

You can run the model, but there is not much room left for longer context, bigger batches, extra apps, or future model updates.

Shared-memory contention still exists

The OS, browser, and inference runtime all compete for the same physical memory pool, so real-world headroom is less forgiving than raw capacity suggests.

Best improvement path

Remove offload with more accelerator memory

Prioritize a GPU or unified-memory tier that fits the whole model natively. Removing offload usually helps more than small compute gains.

Buy headroom, not only minimum fit

A slightly larger memory tier gives you safer context growth and makes the recommendation more future-proof.

Increase host RAM if you keep offloading

This setup may need roughly {ram} GB of extra host RAM just for the offloaded portion, before OS and other tools.

Performance by workload

Workload	Grade	Fit	Decode	TTFT	Context
Chat	A	Runs with offload	5.6 tok/s	18946 ms	8K
Coding	B	Very compromised	4.8 tok/s	40292 ms	8K
Agentic Coding	F	Too heavy	3.9 tok/s	71503 ms	8K
Reasoning	B	Very compromised	4.8 tok/s	47617 ms	8K
RAG	F	Too heavy	3.9 tok/s	89379 ms	8K

Quantization options

How Granite Code 20B (20B params) fits at each quantization level on MacBook Air M3 24GB (17.3 GB usable).

Quant	Bits	VRAM	Quality	Fit
Q2_K	2	7.8 GB	Low	A81
Q3_K_S	3	9.8 GB	Low	A81
NVFP4	4

Get started

Copy-paste commands to run Granite Code 20B on your machine.

Run

ollama run granite-code:20b

Upgrade options

Hardware that runs Granite Code 20B well

MacBook Pro M4 32GBBudget pick

32 GB Unified (+8)120 GB/s (+20)

Removes host-memory offload, which is usually the single biggest latency and throughput win.9.9 tok/s decode

Removes host-memory offload, which is usually the single biggest latency and throughput win.

Raises estimated decode speed by about 90%.

~$799 MSRP

Mac mini M4 32GBBest value

32 GB Unified (+8)120 GB/s (+20)

Removes host-memory offload, which is usually the single biggest latency and throughput win.9.9 tok/s decode

Removes host-memory offload, which is usually the single biggest latency and throughput win.

Raises estimated decode speed by about 90%.

~$1,099 MSRP

Mac mini M4 64GBApple upgrade

64 GB Unified (+40)120 GB/s (+20)

Removes host-memory offload, which is usually the single biggest latency and throughput win.9.9 tok/s decode

Removes host-memory offload, which is usually the single biggest latency and throughput win.

Raises estimated decode speed by about 90%.

~$1,099 MSRP

RTX 5090 Laptop 24GBBiggest leap

896 GB/s (+796)

Removes host-memory offload, which is usually the single biggest latency and throughput win.66.6 tok/s decode

Removes host-memory offload, which is usually the single biggest latency and throughput win.

Raises estimated decode speed by about 1181%.

Frequently asked questions

Can MacBook Air M3 24GB run Granite Code 20B?

Yes, MacBook Air M3 24GB can run Granite Code 20B with a B grade (Very compromised). Expected decode speed: 4.8 tok/s.

How much VRAM does Granite Code 20B need?

Granite Code 20B (20B parameters) requires approximately 18.9 GB of memory with Q4_K_M quantization.

What is the best quantization for Granite Code 20B?

The recommended quantization for Granite Code 20B is Q4_K_M, which balances quality and memory efficiency.

What speed will Granite Code 20B run at on MacBook Air M3 24GB?

On MacBook Air M3 24GB, Granite Code 20B achieves approximately 4.8 tokens per second decode speed with a time-to-first-token of 40292ms using Q4_K_M quantization.

Can MacBook Air M3 24GB run Granite Code 20B for coding?

For coding workloads, Granite Code 20B on MacBook Air M3 24GB receives a B grade with 4.8 tok/s and 8K context.

What context window can Granite Code 20B use on MacBook Air M3 24GB?

On MacBook Air M3 24GB, Granite Code 20B can safely use up to 8K tokens of context. The model's official context limit is 8K, but available memory constrains the safe maximum.

What should I upgrade first if Granite Code 20B feels slow on MacBook Air M3 24GB?

See all results for MacBook Air M3 24GB See all hardware for Granite Code 20B

Embed this result▼

Paste this snippet into any page to show a live fit card.

<iframe src="https://willitrunai.com/embed/granite-code-20b-on-m3-air-24gb" width="400" height="180" frameborder="0" style="border:none;border-radius:12px;overflow:hidden;" title="Will It Run AI — fit result"></iframe>

Preview: