Can Qwen 3.5 122B A10B run on NVIDIA A40 48GB?

YES — With Q2_K

A81Great

Estimated from fit model

Qwen 3.5 122B A10B needs ~55.7 GB VRAM. NVIDIA A40 48GB has 48.0 GB. With Q2_K quantization, expect ~16 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.

Qwen 3.5 122B A10B at Q4_K_M needs 82.6 GB — too much for NVIDIA A40 48GB (48.0 GB). Runs at Q2_K (55.7 GB) with low quality.

Capabilities:

Select quantization to explore

Q4_K_M (Medium quality) — 82.6 GB, exceeds 48.0 GB available

82.6 GB required48.0 GB available

172% VRAM needed

34.6 GB over capacity — needs offload or smaller quantization

Fit status

Too heavy

Decode

5.2 tok/s

TTFT

37042 ms

Safe context

Memory

82.6 GB / 48.0 GB

Offload

40%

Memory breakdown

Weights74.4 GB

KV Cache2.4 GB

Runtime0.9 GB

Headroom4.8 GB

See how fast it feels

With memory offload — actual speed may be lower

See how fast it feelsQwen 3.5 122B A10B on NVIDIA A40 48GB
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 · 37.0s 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.

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 6.6 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	5.4 tok/s	19581 ms	4K
Coding	F	Too heavy	5.2 tok/s	37042 ms	4K
Agentic Coding	F	Too heavy	4.9 tok/s	57287 ms	4K
Reasoning	F	Too heavy	5.2 tok/s	43776 ms	4K
RAG	F	Too heavy	4.9 tok/s	71609 ms	4K

Inference speed

Qwen 3.5 122B A10B inference speed — tokens per second by GPU & Mac

Estimated decode speed (tokens/sec) for Qwen 3.5 122B A10B at Q4_K_M across popular GPUs and Apple Silicon, including multi-GPU rigs, using the fastest local runtime per device. Fastest is Mac Studio M3 Ultra 256GB at ~35 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?
Mac Studio M3 Ultra 256GB	256 GB	Q4_K_M	34.7	Fits
Mac Studio M2 Ultra 128GB	128 GB	Q4_K_M	28.9	Offloads
Mac Studio M1 Ultra 128GB	128 GB	Q4_K_M	27.4	Offloads
MacBook Pro M4 Max 128GB	128 GB	Q4_K_M	21.4	Offloads
2× RX 7900 XTX 24GB	48 GB	Q4_K_M	11.3	Too big
MacBook Pro M4 Max 64GB	64 GB	Q4_K_M	10.0	Too big
2× RTX 4090 24GB	48 GB	Q4_K_M	7.6	Too big
RTX 5090 32GB	32 GB	Q4_K_M	7.2	Too big
MacBook Pro M3 Max 64GB	64 GB	Q4_K_M	7.0	Too big
2× RTX 3090 24GB	48 GB	Q4_K_M	6.5	Too big
MacBook Pro M1 Max 64GB	64 GB	Q4_K_M	6.4	Too big
MacBook Pro M4 Pro 48GB	48 GB	Q4_K_M	5.9	Too big
4× RTX 3060 12GB	48 GB	Q4_K_M	5.7	Too big
RTX 4090 24GB	24 GB	Q4_K_M	4.6	Too big
RX 7900 XTX 24GB	24 GB	Q4_K_M	4.2	Too big
RTX 3090 24GB	24 GB	Q4_K_M	4.0	Too big
RTX 4080 Super 16GB	16 GB	Q4_K_M	3.7	Too big
RTX 4070 12GB	12 GB	Q4_K_M	2.3	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

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 Qwen 3.5 122B A10B (122B params) fits at each quantization level on NVIDIA A40 48GB (48.0 GB usable).

Quant	Bits	VRAM	Quality	Fit
Q2_K	2	47.6 GB	Low	F0
Q3_K_S	3	59.8 GB	Low	F0
NVFP4	4	68.3 GB	Medium	F0
Q4_K_M	4	74.4 GB	Medium	F0
Q5_K_M	5	87.8 GB	High	F0
Q6_K	6	100.0 GB	High	F0
Q8_0	8	130.5 GB	Very High	F0
F16	16	250.1 GB	Maximum	F0

Get started

Copy-paste commands to run Qwen 3.5 122B A10B on your machine.

Run

lms load Qwen3.5-122B-A10B-Instruct && lms server start

Opciones de mejora

Hardware que ejecuta bien Qwen 3.5 122B A10B

RTX PRO 6000 Blackwell Workstation Edition 96GBOpción económica

96 GB VRAM (+48)1792 GB/s (+1096)

Hace que el modelo quepa en el acelerador en lugar de seguir fuera de alcance.60.5 tok/s decodificación

Hace que el modelo quepa en el acelerador en lugar de seguir fuera de alcance.

Elimina el offload a memoria del sistema, que suele ser la mayor mejora individual en latencia y throughput.

~$9,999 MSRP

RTX PRO 6000 Blackwell Server Edition 96GBMejor relación calidad-precio

96 GB VRAM (+48)1597 GB/s (+901)

Hace que el modelo quepa en el acelerador en lugar de seguir fuera de alcance.53.9 tok/s decodificación

Hace que el modelo quepa en el acelerador en lugar de seguir fuera de alcance.

Elimina el offload a memoria del sistema, que suele ser la mayor mejora individual en latencia y throughput.

~$9,999 MSRP

NVIDIA H20 96GBMejora NVIDIA

96 GB VRAM (+48)4000 GB/s (+3304)

Hace que el modelo quepa en el acelerador en lugar de seguir fuera de alcance.130.3 tok/s decodificación

Hace que el modelo quepa en el acelerador en lugar de seguir fuera de alcance.

Elimina el offload a memoria del sistema, que suele ser la mayor mejora individual en latencia y throughput.

~$12,000 MSRP

Frequently asked questions

Can NVIDIA A40 48GB run Qwen 3.5 122B A10B?

Yes, NVIDIA A40 48GB can run Qwen 3.5 122B A10B at Q2_K quantization (Very compromised (needs ~6.6 GB host RAM)). The recommended Q4_K_M requires 82.6 GB which exceeds available memory, but at Q2_K it needs only 55.7 GB. Expected decode speed: 15.9 tok/s.

How much VRAM does Qwen 3.5 122B A10B need?

Qwen 3.5 122B A10B (122B parameters) requires approximately 82.6 GB at Q4_K_M quantization. On NVIDIA A40 48GB, it fits at Q2_K using 55.7 GB.

What is the best quantization for Qwen 3.5 122B A10B?

The recommended quantization is Q4_K_M, but on NVIDIA A40 48GB the best fitting quantization is Q2_K, which uses 55.7 GB.

What speed will Qwen 3.5 122B A10B run at on NVIDIA A40 48GB?

On NVIDIA A40 48GB, Qwen 3.5 122B A10B achieves approximately 15.9 tokens per second decode speed with a time-to-first-token of 12178ms using Q2_K quantization.

Can NVIDIA A40 48GB run Qwen 3.5 122B A10B for coding?

For coding workloads, Qwen 3.5 122B A10B on NVIDIA A40 48GB receives a F grade with 5.2 tok/s and 4K context.

What context window can Qwen 3.5 122B A10B use on NVIDIA A40 48GB?

On NVIDIA A40 48GB, Qwen 3.5 122B A10B can safely use up to 4K tokens of context at Q2_K quantization. The model's official context limit is 131K, but available memory constrains the safe maximum.

What should I upgrade first if Qwen 3.5 122B A10B feels slow on NVIDIA A40 48GB?

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 A40 48GB See all hardware for Qwen 3.5 122B A10B

Embed this result▼

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

<iframe src="https://willitrunai.com/embed/qwen-3.5-122b-a10b-on-a40-48gb" width="400" height="180" frameborder="0" style="border:none;border-radius:12px;overflow:hidden;" title="Will It Run AI — fit result"></iframe>

Preview: