Can Qwen 3.5 35B A3B run on NVIDIA T4 16GB?

YES — With Q2_K

A81Great

Estimated from fit model

Qwen 3.5 35B A3B needs ~17.9 GB VRAM. NVIDIA T4 16GB has 16.0 GB. With Q2_K quantization, expect ~22 tok/s.

Runtime: OllamaCapacity: OffloadBandwidth: LowStack: BasicBottleneck: 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 35B A3B at Q4_K_M needs 25.6 GB — too much for NVIDIA T4 16GB (16.0 GB). Runs at Q2_K (17.9 GB) with low quality.

Capabilities:

Select quantization to explore

Q4_K_M (Medium quality) — 25.6 GB, exceeds 16.0 GB available

25.6 GB required16.0 GB available

160% VRAM needed

9.6 GB over capacity — needs offload or smaller quantization

Fit status

Too heavy

Decode

7.5 tok/s

TTFT

25951 ms

Safe context

Memory

25.6 GB / 16.0 GB

Offload

40%

Memory breakdown

Weights21.3 GB

KV Cache1.5 GB

Runtime1.2 GB

Headroom1.6 GB

See how fast it feels

With memory offload — actual speed may be lower

See how fast it feelsQwen 3.5 35B A3B on NVIDIA T4 16GB
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: 7.5 tok/s decode · 26.0s TTFT (warm) · 19 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.

Older PCIe generation

PCIe 3.0 is workable, but it compounds the penalty when you offload heavily or try to scale across multiple cards.

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 1.5 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	8.0 tok/s	13282 ms	4K
Coding	F	Too heavy	7.5 tok/s	25951 ms	4K
Agentic Coding	F	Too heavy	6.6 tok/s	42651 ms	4K
Reasoning	F	Too heavy	7.5 tok/s	30670 ms	4K
RAG	F	Too heavy	6.6 tok/s	53313 ms	4K

Quantization options

How Qwen 3.5 35B A3B (35B params) fits at each quantization level on NVIDIA T4 16GB (16.0 GB usable).

Quant	Bits	VRAM	Quality	Fit
Q2_K	2	13.7 GB	Low	F0
Q3_K_S	3	17.2 GB	Low	F0
NVFP4	4	19.6 GB	Medium	F0
Q4_K_M	4	21.3 GB	Medium	F0
Q5_K_M	5	25.2 GB	High	F0
Q6_K	6	28.7 GB	High	F0
Q8_0	8	37.5 GB	Very High	F0
F16	16	71.8 GB	Maximum	F0

Get started

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

Run

ollama run qwen3.5:35b-a3b

Upgrade options

Hardware that runs Qwen 3.5 35B A3B well

RTX 3090 24GBBest value

24 GB VRAM (+8)936 GB/s (+616)

Makes the model fit on the accelerator instead of staying completely out of reach.55.5 tok/s decode

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

Raises estimated decode speed by about 640%.

~$1,499 MSRP

RTX PRO 4000 Blackwell 24GBNVIDIA upgrade

24 GB VRAM (+8)672 GB/s (+352)

Makes the model fit on the accelerator instead of staying completely out of reach.49.1 tok/s decode

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

Raises estimated decode speed by about 555%.

~$1,599 MSRP

RTX 5090 32GBBudget pick

32 GB VRAM (+16)1792 GB/s (+1472)

Makes the model fit on the accelerator instead of staying completely out of reach.139.4 tok/s decode

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.

~$1,999 MSRP

Frequently asked questions

Can NVIDIA T4 16GB run Qwen 3.5 35B A3B?

Yes, NVIDIA T4 16GB can run Qwen 3.5 35B A3B at Q2_K quantization (Very compromised (needs ~1.5 GB host RAM)). The recommended Q4_K_M requires 25.6 GB which exceeds available memory, but at Q2_K it needs only 17.9 GB. Expected decode speed: 21.7 tok/s.

How much VRAM does Qwen 3.5 35B A3B need?

Qwen 3.5 35B A3B (35B parameters) requires approximately 25.6 GB at Q4_K_M quantization. On NVIDIA T4 16GB, it fits at Q2_K using 17.9 GB.

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

The recommended quantization is Q4_K_M, but on NVIDIA T4 16GB the best fitting quantization is Q2_K, which uses 17.9 GB.

What speed will Qwen 3.5 35B A3B run at on NVIDIA T4 16GB?

On NVIDIA T4 16GB, Qwen 3.5 35B A3B achieves approximately 21.7 tokens per second decode speed with a time-to-first-token of 8902ms using Q2_K quantization.

Can NVIDIA T4 16GB run Qwen 3.5 35B A3B for coding?

For coding workloads, Qwen 3.5 35B A3B on NVIDIA T4 16GB receives a F grade with 7.5 tok/s and 4K context.

What context window can Qwen 3.5 35B A3B use on NVIDIA T4 16GB?

On NVIDIA T4 16GB, Qwen 3.5 35B A3B 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 35B A3B feels slow on NVIDIA T4 16GB?

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 T4 16GB 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-t4-16gb" width="400" height="180" frameborder="0" style="border:none;border-radius:12px;overflow:hidden;" title="Will It Run AI — fit result"></iframe>

Preview:

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.

Older PCIe generation

PCIe 3.0 is workable, but it compounds the penalty when you offload heavily or try to scale across multiple cards.

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 1.5 GB of extra host RAM just for the offloaded portion, before OS and other tools.

Workload

Grade

Fit

Decode

TTFT

Context

Chat

Too heavy

8.0 tok/s

13282 ms

Coding

Too heavy

7.5 tok/s

25951 ms

Agentic Coding

Too heavy

6.6 tok/s

42651 ms

Reasoning

Too heavy

7.5 tok/s

30670 ms

RAG

Too heavy

6.6 tok/s

53313 ms

Quant

Bits

VRAM

Quality

Fit

Q2_K

13.7 GB

Low

Q3_K_S

17.2 GB

Low

NVFP4

19.6 GB

Medium

Q4_K_M

21.3 GB

Medium

Q5_K_M

25.2 GB

High

Q6_K

28.7 GB

High

Q8_0

37.5 GB

Very High

F16

71.8 GB

Maximum

Frequently asked questions

Can NVIDIA T4 16GB run Qwen 3.5 35B A3B?

How much VRAM does Qwen 3.5 35B A3B need?

Qwen 3.5 35B A3B (35B parameters) requires approximately 25.6 GB at Q4_K_M quantization. On NVIDIA T4 16GB, it fits at Q2_K using 17.9 GB.

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

The recommended quantization is Q4_K_M, but on NVIDIA T4 16GB the best fitting quantization is Q2_K, which uses 17.9 GB.

What speed will Qwen 3.5 35B A3B run at on NVIDIA T4 16GB?

On NVIDIA T4 16GB, Qwen 3.5 35B A3B achieves approximately 21.7 tokens per second decode speed with a time-to-first-token of 8902ms using Q2_K quantization.

Can NVIDIA T4 16GB run Qwen 3.5 35B A3B for coding?

For coding workloads, Qwen 3.5 35B A3B on NVIDIA T4 16GB receives a F grade with 7.5 tok/s and 4K context.

What context window can Qwen 3.5 35B A3B use on NVIDIA T4 16GB?

On NVIDIA T4 16GB, Qwen 3.5 35B A3B 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 35B A3B feels slow on NVIDIA T4 16GB?

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.