Is unified memory on MacBook Pro M3 24GB as fast as VRAM for Codestral RAG 19B Pruned i1?

Not always. MacBook Pro 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 Codestral RAG 19B Pruned i1 run on MacBook Pro M3 24GB?

Q: What should I upgrade first if Codestral RAG 19B Pruned i1 feels slow on MacBook Pro M3 24GB?

Prioritize bandwidth, not only capacity. If this workload feels slow, the next useful step is often a GPU tier with materially faster memory bandwidth rather than only a small bump in capacity.

YES — With Offload

C45Usable

Estimated from fit model

Codestral RAG 19B Pruned i1 needs ~17.3 GB VRAM. MacBook Pro M3 24GB has 17.3 GB. With Q4_K_M quantization, expect ~6 tok/s.

Runtime: llama.cppCapacity: OffloadBandwidth: Very lowStack: StandardBottleneck: Memory bandwidth

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) — 17.3 GB, 5.8 tok/s, Runs with offload (needs ~0 GB host RAM)

17.3 GB required17.3 GB available

100% VRAM used

Fit status

Runs with offload (needs ~0 GB host RAM)

Decode

5.8 tok/s

TTFT

33195 ms

Safe context

16K

Memory

17.3 GB / 17.3 GB

Memory breakdown

Weights11.6 GB

KV Cache2.2 GB

Runtime0.9 GB

Headroom2.6 GB

See how fast it feels

See how fast it feelsCodestral RAG 19B Pruned i1 on MacBook Pro 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: 5.8 tok/s decode · 33.2s TTFT (warm) · 15 tok/s prefill

What limits this setup

The model fits in shared memory, but shared-memory bandwidth is now the real limiter.

Fit does not mean dedicated-VRAM speed

Unified or shared memory can make a model technically fit, but sustained tokens per second may still trail a discrete high-bandwidth GPU with less total memory.

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

Prioritize bandwidth, not only capacity

If this workload feels slow, the next useful step is often a GPU tier with materially faster memory bandwidth rather than only a small bump in capacity.

Buy headroom, not only minimum fit

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

Performance by workload

Workload	Grade	Fit	Decode	TTFT	Context
Chat	C	Tight fit	5.9 tok/s	17999 ms	16K
Coding	C	Runs with offload (needs ~0 GB host RAM)	5.8 tok/s	33195 ms	16K
Agentic Coding	D	Very compromised (needs ~1.3 GB host RAM)	4.8 tok/s	58374 ms	16K
Reasoning	C	Runs with offload (needs ~0 GB host RAM)	5.8 tok/s	39230 ms	16K
RAG	D	Very compromised (needs ~1.3 GB host RAM)	4.8 tok/s

Quantization options

How Codestral RAG 19B Pruned i1 (19B params) fits at each quantization level on MacBook Pro M3 24GB (17.3 GB usable).

Quant	Bits	VRAM	Quality	Fit
Q2_K	2	7.4 GB	Low	C50
Q3_K_S	3	9.3 GB	Low	C51
NVFP4	4

Get started

Copy-paste commands to run Codestral RAG 19B Pruned i1 on your machine.

Run

lms load hf-mradermacher--codestral-rag-19b-pruned-i1-gguf && lms server start

Upgrade options

Hardware that runs Codestral RAG 19B Pruned i1 well

MacBook Pro M4 32GBBudget pick

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

Raises estimated decode speed by about 60%.9.3 tok/s decode

Raises estimated decode speed by about 60%.

Adds memory headroom for longer context windows and future model growth.

~$799 MSRP

Mac mini M4 32GBBest value

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

Raises estimated decode speed by about 60%.9.3 tok/s decode

Raises estimated decode speed by about 60%.

Adds memory headroom for longer context windows and future model growth.

~$1,099 MSRP

MacBook Pro M4 Pro 64GBApple upgrade

64 GB Unified (+40)273 GB/s (+173)

Raises estimated decode speed by about 291%.22.7 tok/s decode

Raises estimated decode speed by about 291%.

Adds memory headroom for longer context windows and future model growth.

~$1,599 MSRP

RTX 5090 Laptop 24GBBiggest leap

896 GB/s (+796)

Raises estimated decode speed by about 1019%.64.9 tok/s decode

Raises estimated decode speed by about 1019%.

Adds memory headroom for longer context windows and future model growth.

Frequently asked questions

Can MacBook Pro M3 24GB run Codestral RAG 19B Pruned i1?

Yes, MacBook Pro M3 24GB can run Codestral RAG 19B Pruned i1 with a C grade (Runs with offload (needs ~0 GB host RAM)). Expected decode speed: 5.8 tok/s.

How much VRAM does Codestral RAG 19B Pruned i1 need?

Codestral RAG 19B Pruned i1 (19B parameters) requires approximately 17.3 GB of memory with Q4_K_M quantization.

What is the best quantization for Codestral RAG 19B Pruned i1?

The recommended quantization for Codestral RAG 19B Pruned i1 is Q4_K_M, which balances quality and memory efficiency.

What speed will Codestral RAG 19B Pruned i1 run at on MacBook Pro M3 24GB?

On MacBook Pro M3 24GB, Codestral RAG 19B Pruned i1 achieves approximately 5.8 tokens per second decode speed with a time-to-first-token of 33195ms using Q4_K_M quantization.

Can MacBook Pro M3 24GB run Codestral RAG 19B Pruned i1 for coding?

For coding workloads, Codestral RAG 19B Pruned i1 on MacBook Pro M3 24GB receives a C grade with 5.8 tok/s and 16K context.

What context window can Codestral RAG 19B Pruned i1 use on MacBook Pro M3 24GB?

On MacBook Pro M3 24GB, Codestral RAG 19B Pruned i1 can safely use up to 16K tokens of context. The model's official context limit is —, but available memory constrains the safe maximum.

See all results for MacBook Pro M3 24GB See all hardware for Codestral RAG 19B Pruned i1

Embed this result▼

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

<iframe src="https://willitrunai.com/embed/hf-mradermacher--codestral-rag-19b-pruned-i1-gguf-on-m3-24gb" width="400" height="180" frameborder="0" style="border:none;border-radius:12px;overflow:hidden;" title="Will It Run AI — fit result"></iframe>

Preview: