Can Qwen 3.5 35B A3B run on NVIDIA L4 24GB?

Q: What should I upgrade first if Qwen 3.5 35B A3B feels slow on NVIDIA L4 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

A80Great

Estimated — low-sample bucket· few comparable runs

Qwen 3.5 35B A3B needs ~26.1 GB VRAM. NVIDIA L4 24GB has 24.0 GB. With Q4_K_M quantization, expect ~16 tok/s.

Runtime: llama.cppCapacity: OffloadBandwidth: 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) — 26.1 GB, 18.1 tok/s, Very compromised (needs ~1.7 GB host RAM)

26.1 GB required24.0 GB available

109% VRAM needed

2.1 GB over capacity — needs offload or smaller quantization

Fit status

Very compromised (needs ~1.7 GB host RAM)

Decode

18.1 tok/s

TTFT

10693 ms

Safe context

Memory

26.1 GB / 24.0 GB

Offload

10%

Memory breakdown

Weights21.3 GB

KV Cache1.5 GB

Runtime0.9 GB

Headroom2.4 GB

See how fast it feels

See how fast it feelsQwen 3.5 35B A3B on NVIDIA L4 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: 18.1 tok/s decode · 10.7s TTFT (warm) · 45 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 {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	16.5 tok/s	6393 ms	4K
Coding	A	Very compromised	15.6 tok/s	12443 ms	4K
Agentic Coding	A	Very compromised	13.9 tok/s	20303 ms	4K
Reasoning	A	Very compromised	15.6 tok/s	14706 ms	4K
RAG	A	Very compromised	13.9 tok/s	25378 ms	4K

Quantization options

How Qwen 3.5 35B A3B (35B params) fits at each quantization level on NVIDIA L4 24GB (24.0 GB usable).

Quant	Bits	VRAM	Quality	Fit
Q2_K	2	13.7 GB	Low	S92
Q3_K_SBest for your GPU	3	17.2 GB	Low	S91

Get started

Copy-paste commands to run Qwen 3.5 35B A3B on your machine.

Run

ollama run qwen3.5:35b-a3b

Frequently asked questions

Can NVIDIA L4 24GB run Qwen 3.5 35B A3B?

Yes, NVIDIA L4 24GB can run Qwen 3.5 35B A3B with a A grade (Very compromised). Expected decode speed: 15.6 tok/s.

How much VRAM does Qwen 3.5 35B A3B need?

Qwen 3.5 35B A3B (35B parameters) requires approximately 26.1 GB of memory with Q4_K_M quantization.

What is the best quantization for Qwen 3.5 35B A3B?

The recommended quantization for Qwen 3.5 35B A3B is Q4_K_M, which balances quality and memory efficiency.

What speed will Qwen 3.5 35B A3B run at on NVIDIA L4 24GB?

On NVIDIA L4 24GB, Qwen 3.5 35B A3B achieves approximately 15.6 tokens per second decode speed with a time-to-first-token of 12443ms using Q4_K_M quantization.

Can NVIDIA L4 24GB run Qwen 3.5 35B A3B for coding?

For coding workloads, Qwen 3.5 35B A3B on NVIDIA L4 24GB receives a A grade with 15.6 tok/s and 4K context.

What context window can Qwen 3.5 35B A3B use on NVIDIA L4 24GB?

On NVIDIA L4 24GB, Qwen 3.5 35B A3B can safely use up to 4K tokens of context. The model's official context limit is 131K, but available memory constrains the safe maximum.

What should I upgrade first if Qwen 3.5 35B A3B feels slow on NVIDIA L4 24GB?

See all results for NVIDIA L4 24GB See all hardware for Qwen 3.5 35B A3B

Embed this result▼

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

<iframe src="https://willitrunai.com/embed/qwen-3.5-35b-a3b-on-l4-24gb" width="400" height="180" frameborder="0" style="border:none;border-radius:12px;overflow:hidden;" title="Will It Run AI — fit result"></iframe>

Preview: