Can Devstral 2 123B Instruct run on NVIDIA A16 64GB?

YES — With Q3_K_S

A78Great

Estimated from fit model

Devstral 2 123B Instruct needs ~72.9 GB VRAM. NVIDIA A16 64GB has 64.0 GB. With Q3_K_S quantization, expect ~5 tok/s.

Runtime: llama.cppCapacity: OffloadBandwidth: MediumStack: 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.

Devstral 2 123B Instruct at Q4_K_M needs 87.7 GB — too much for NVIDIA A16 64GB (64.0 GB). Runs at Q3_K_S (72.9 GB) with low quality. 2 quantization levels fit.

Capabilities:

Select quantization to explore

Q4_K_M (Medium quality) — 87.7 GB, exceeds 64.0 GB available

87.7 GB required64.0 GB available

137% VRAM needed

23.7 GB over capacity — needs offload or smaller quantization

Fit status

Too heavy

Decode

2.6 tok/s

TTFT

73862 ms

Safe context

Memory

87.7 GB / 64.0 GB

Offload

30%

Memory breakdown

Weights75.0 GB

KV Cache5.4 GB

Runtime0.9 GB

Headroom6.4 GB

See how fast it feels

With memory offload — actual speed may be lower

See how fast it feelsDevstral 2 123B Instruct on NVIDIA A16 64GB
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: 2.6 tok/s decode · 73.9s TTFT (warm) · 7 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.

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 7.4 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	F	Too heavy	2.8 tok/s	37735 ms	4K
Coding	F	Too heavy	2.6 tok/s	73862 ms	4K
Agentic Coding	F	Too heavy	2.3 tok/s	121755 ms	4K
Reasoning	F	Too heavy	2.6 tok/s	87291 ms	4K
RAG	F	Too heavy	2.3 tok/s	152194 ms	4K

Inference speed

Devstral 2 123B Instruct inference speed — tokens per second by GPU & Mac

Estimated decode speed (tokens/sec) for Devstral 2 123B Instruct at Q4_K_M across popular GPUs and Apple Silicon, including multi-GPU rigs, using the fastest local runtime per device. Fastest is MacBook Pro M4 Max 128GB at ~8 tok/s. Speed is memory-bandwidth bound, so cards that fit the whole model in VRAM run far faster than ones that offload to system RAM.

GPU / Mac	Memory	Quant	Speed (tok/s)	Fits?
MacBook Pro M4 Max 128GB	128 GB	Q4_K_M	8.2	Offloads
Mac Studio M3 Ultra 256GB	256 GB	Q4_K_M	8.1	Fits
Mac Studio M2 Ultra 128GB	128 GB	Q4_K_M	6.3	Offloads
Mac Studio M1 Ultra 128GB	128 GB	Q4_K_M	6.0	Offloads
MacBook Pro M4 Max 64GB	64 GB	Q4_K_M	3.9	Too big
2× RX 7900 XTX 24GB	48 GB	Q4_K_M	3.7	Too big
2× RTX 4090 24GB	48 GB	Q4_K_M	2.6	Too big
MacBook Pro M4 Pro 48GB	48 GB	Q4_K_M	2.5	Too big
2× RTX 3090 24GB	48 GB	Q4_K_M	2.2	Too big
RTX 5090 32GB	32 GB	Q4_K_M	2.0	Too big
RTX 4090 24GB	24 GB	Q4_K_M	2.0	Too big
RTX 4080 Super 16GB	16 GB	Q4_K_M	2.0	Too big
RTX 3090 24GB	24 GB	Q4_K_M	2.0	Too big
RTX 4070 12GB	12 GB	Q4_K_M	2.0	Too big
RTX 3060 12GB	12 GB	Q4_K_M	2.0	Too big
RTX 4060 8GB	8 GB	Q4_K_M	2.0	Too big
RX 7900 XTX 24GB	24 GB	Q4_K_M	2.0	Too big
MacBook Pro M3 Max 64GB	64 GB	Q4_K_M	2.0	Too big
MacBook Pro M1 Max 64GB	64 GB	Q4_K_M	2.0	Too big
4× RTX 3060 12GB	48 GB	Q4_K_M	2.0	Too big

Estimates for single-stream decoding at Q4_K_M; real tokens/sec varies with prompt length, context, batch size, and runtime build. Prompt processing (prefill) is faster than the decode figures shown here.

Quantization options

How Devstral 2 123B Instruct (123B params) fits at each quantization level on NVIDIA A16 64GB (64.0 GB usable).

Quant	Bits	VRAM	Quality	Fit
Q2_KBest for your GPU	2	48.0 GB	Low	S91
Q3_K_S	3	60.3 GB	Low	F0
NVFP4	4	68.9 GB	Medium	F0
Q4_K_M	4	75.0 GB	Medium	F0
Q5_K_M	5	88.6 GB	High	F0
Q6_K	6	100.9 GB	High	F0
Q8_0	8	131.6 GB	Very High	F0
F16	16	252.2 GB	Maximum	F0

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

升级选项

能流畅运行 Devstral 2 123B Instruct 的硬件

RTX PRO 6000 Blackwell Workstation Edition 96GB经济之选

96 GB VRAM (+32)1792 GB/s (+1192)

Makes the model fit on the accelerator instead of staying completely out of reach.21.8 tok/s 解码

Makes the model fit on the accelerator instead of staying completely out of reach.

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

~$9,999 MSRP

RTX PRO 6000 Blackwell Server Edition 96GB最佳性价比

96 GB VRAM (+32)1597 GB/s (+997)

Makes the model fit on the accelerator instead of staying completely out of reach.19.4 tok/s 解码

Makes the model fit on the accelerator instead of staying completely out of reach.

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

~$9,999 MSRP

NVIDIA H20 96GBNVIDIA 升级

96 GB VRAM (+32)4000 GB/s (+3400)

Makes the model fit on the accelerator instead of staying completely out of reach.47 tok/s 解码

Makes the model fit on the accelerator instead of staying completely out of reach.

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

~$12,000 MSRP

Frequently asked questions

Can NVIDIA A16 64GB run Devstral 2 123B Instruct?

Yes, NVIDIA A16 64GB can run Devstral 2 123B Instruct at Q3_K_S quantization (Very compromised (needs ~7.4 GB host RAM)). The recommended Q4_K_M requires 87.7 GB which exceeds available memory, but at Q3_K_S it needs only 72.9 GB. Expected decode speed: 4.5 tok/s.

How much VRAM does Devstral 2 123B Instruct need?

Devstral 2 123B Instruct (123B parameters) requires approximately 87.7 GB at Q4_K_M quantization. On NVIDIA A16 64GB, it fits at Q3_K_S using 72.9 GB.

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

The recommended quantization is Q4_K_M, but on NVIDIA A16 64GB the best fitting quantization is Q3_K_S, which uses 72.9 GB.

What speed will Devstral 2 123B Instruct run at on NVIDIA A16 64GB?

On NVIDIA A16 64GB, Devstral 2 123B Instruct achieves approximately 4.5 tokens per second decode speed with a time-to-first-token of 43285ms using Q3_K_S quantization.

Can NVIDIA A16 64GB run Devstral 2 123B Instruct for coding?

For coding workloads, Devstral 2 123B Instruct on NVIDIA A16 64GB receives a F grade with 2.6 tok/s and 4K context.

What context window can Devstral 2 123B Instruct use on NVIDIA A16 64GB?

On NVIDIA A16 64GB, Devstral 2 123B Instruct can safely use up to 4K tokens of context at Q3_K_S quantization. The model's official context limit is 256K, but available memory constrains the safe maximum.

What should I upgrade first if Devstral 2 123B Instruct feels slow on NVIDIA A16 64GB?

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.

See all results for NVIDIA A16 64GB 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-a16-64gb" width="400" height="180" frameborder="0" style="border:none;border-radius:12px;overflow:hidden;" title="Will It Run AI — fit result"></iframe>

Preview: