Is unified memory on Mac Studio M2 Ultra 128GB as fast as VRAM for Devstral 2 123B Instruct?

Not always. Mac Studio M2 Ultra 128GB 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 Devstral 2 123B Instruct run on Mac Studio M2 Ultra 128GB?

Q: What should I upgrade first if Devstral 2 123B Instruct feels slow on Mac Studio M2 Ultra 128GB?

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

S89Excellent

Estimated from fit model

Devstral 2 123B Instruct needs ~95.1 GB VRAM. Mac Studio M2 Ultra 128GB has 92.2 GB. With Q4_K_M quantization, expect ~6 tok/s.

Runtime: llama.cppCapacity: OffloadBandwidth: HighStack: 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) — 95.1 GB, 6.3 tok/s, Runs with offload (needs ~2.3 GB host RAM)

95.1 GB required92.2 GB available

103% VRAM needed

2.9 GB over capacity — needs offload or smaller quantization

Fit status

Runs with offload (needs ~2.3 GB host RAM)

Decode

6.3 tok/s

TTFT

30702 ms

Safe context

Memory

95.1 GB / 92.2 GB

Memory breakdown

Weights75.0 GB

KV Cache5.4 GB

Runtime0.9 GB

Headroom13.8 GB

See how fast it feels

See how fast it feelsDevstral 2 123B Instruct on Mac Studio M2 Ultra 128GB
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: 6.3 tok/s decode · 30.7s TTFT (warm) · 16 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	S	Runs with offload (needs ~0.2 GB host RAM)	6.7 tok/s	15855 ms	7K
Coding	S	Runs with offload (needs ~2.3 GB host RAM)	6.3 tok/s	30702 ms	7K
Agentic Coding	A	Very compromised (needs ~6.2 GB host RAM)	5.8 tok/s	48489 ms	7K
Reasoning	S	Runs with offload (needs ~2.3 GB host RAM)	6.3 tok/s	36284 ms	7K
RAG	A	Very compromised (needs ~6.2 GB host RAM)	5.8 tok/s

Quantization options

How Devstral 2 123B Instruct (123B params) fits at each quantization level on Mac Studio M2 Ultra 128GB (92.2 GB usable).

Quant	Bits	VRAM	Quality	Fit
Q2_K	2	48.0 GB	Low	S91
Q3_K_S	3	60.3 GB	Low	S91
NVFP4Best for your GPU

Get started

Copy-paste commands to run Devstral 2 123B Instruct on your machine.

Run

lms load Devstral-2-123B-Instruct-2512 && lms server start

Frequently asked questions

Can Mac Studio M2 Ultra 128GB run Devstral 2 123B Instruct?

Yes, Mac Studio M2 Ultra 128GB can run Devstral 2 123B Instruct with a S grade (Runs with offload (needs ~2.3 GB host RAM)). Expected decode speed: 6.3 tok/s.

How much VRAM does Devstral 2 123B Instruct need?

Devstral 2 123B Instruct (123B parameters) requires approximately 95.1 GB of memory with Q4_K_M quantization.

What is the best quantization for Devstral 2 123B Instruct?

The recommended quantization for Devstral 2 123B Instruct is Q4_K_M, which balances quality and memory efficiency.

What speed will Devstral 2 123B Instruct run at on Mac Studio M2 Ultra 128GB?

On Mac Studio M2 Ultra 128GB, Devstral 2 123B Instruct achieves approximately 6.3 tokens per second decode speed with a time-to-first-token of 30702ms using Q4_K_M quantization.

Can Mac Studio M2 Ultra 128GB run Devstral 2 123B Instruct for coding?

For coding workloads, Devstral 2 123B Instruct on Mac Studio M2 Ultra 128GB receives a S grade with 6.3 tok/s and 7K context.

What context window can Devstral 2 123B Instruct use on Mac Studio M2 Ultra 128GB?

On Mac Studio M2 Ultra 128GB, Devstral 2 123B Instruct can safely use up to 7K tokens of context. The model's official context limit is 256K, but available memory constrains the safe maximum.

See all results for Mac Studio M2 Ultra 128GB See all hardware for Devstral 2 123B Instruct

Embed this result▼

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

<iframe src="https://willitrunai.com/embed/devstral-2-123b-on-m2-ultra-128gb" width="400" height="180" frameborder="0" style="border:none;border-radius:12px;overflow:hidden;" title="Will It Run AI — fit result"></iframe>

Preview: