Can Qwen 3.6 35B A3B run on AMD Instinct MI250 128GB?

YES — Runs Great

S91Excellent

Estimated from fit model

Qwen 3.6 35B A3B needs ~40.9 GB VRAM. AMD Instinct MI250 128GB has 128.0 GB. With Q4_K_M quantization, expect ~277 tok/s.

Runtime: SGLangCapacity: RoomyBandwidth: HighStack: OptimizedBottleneck: Balanced

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) — 40.9 GB, 276.5 tok/s, Runs well

40.9 GB required128.0 GB available

32% VRAM used

Fit status

Runs well

Decode

276.5 tok/s

TTFT

700 ms

Safe context

262K

Memory

40.9 GB / 128.0 GB

Memory breakdown

Weights21.3 GB

KV Cache4.1 GB

Runtime2.6 GB

Headroom12.8 GB

See how fast it feels

See how fast it feelsQwen 3.6 35B A3B on AMD Instinct MI250 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: 276.5 tok/s decode · 700ms TTFT (warm) · 691 tok/s prefill

What limits this setup

This setup is broadly balanced for this model.

No major red flags

This recommendation has enough memory headroom and acceptable estimated speed for the selected workload.

Best improvement path

Performance by workload

Workload	Grade	Fit	Decode	TTFT	Context
Chat	S	Runs well	276.5 tok/s	382 ms	262K
Coding	S	Runs well	276.5 tok/s	700 ms	262K
Agentic Coding	S	Runs well	276.5 tok/s	1018 ms	262K
Reasoning	S	Runs well	276.5 tok/s	827 ms	262K
RAG	S	Runs well	276.5 tok/s	1273 ms	262K

Inference speed

Qwen 3.6 35B A3B inference speed — tokens per second by GPU & Mac

Estimated decode speed (tokens/sec) for Qwen 3.6 35B A3B at Q4_K_M across popular GPUs and Apple Silicon, using the fastest local runtime per device. Fastest is RTX 5090 32GB at ~153 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?
RTX 5090 32GB	32 GB	Q4_K_M	152.6	Offloads
Mac Studio M3 Ultra 256GB	256 GB	Q4_K_M	70.8	Fits
Mac Studio M2 Ultra 128GB	128 GB	Q4_K_M	59.0	Fits
Mac Studio M1 Ultra 128GB	128 GB	Q4_K_M	55.9	Fits
MacBook Pro M4 Max 128GB	128 GB	Q4_K_M	43.7	Fits
MacBook Pro M4 Max 64GB	64 GB	Q4_K_M	43.7	Fits
RTX 4090 24GB	24 GB	Q4_K_M	34.1	Too big
RX 7900 XTX 24GB	24 GB	Q4_K_M	30.8	Too big
MacBook Pro M3 Max 64GB	64 GB	Q4_K_M	30.5	Fits
RTX 3090 24GB	24 GB	Q4_K_M	29.2	Too big
MacBook Pro M1 Max 64GB	64 GB	Q4_K_M	28.0	Fits
MacBook Pro M4 Pro 48GB	48 GB	Q4_K_M	26.7	Tight
RTX 4080 Super 16GB	16 GB	Q4_K_M	12.2	Too big
RTX 4070 12GB	12 GB	Q4_K_M	5.5	Too big
RTX 3060 12GB	12 GB	Q4_K_M	3.4	Too big
RTX 4060 8GB	8 GB	Q4_K_M	2.9	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 3.6 35B A3B (35B params) fits at each quantization level on AMD Instinct MI250 128GB (128.0 GB usable).

Quant	Bits	VRAM	Quality	Fit
Q2_K	2	13.7 GB	Low	A81
Q3_K_S	3	17.2 GB	Low	A81
NVFP4	4	19.6 GB	Medium	A81
Q4_K_M	4	21.3 GB	Medium	A81
Q5_K_M	5	25.2 GB	High	A82
Q6_K	6	28.7 GB	High	A83
Q8_0	8	37.5 GB	Very High	A84
F16Best for your GPU	16	71.8 GB	Maximum	S90

Get started

Copy-paste commands to run Qwen 3.6 35B A3B on your machine.

Run

docker run --rm -it ghcr.io/ggerganov/llama.cpp:full \
  --hf-repo "Qwen/Qwen3.6-35B-A3B" \
  --hf-file "Qwen3.6-35B-A3B-Q4_K_M.gguf" \
  -c 4096 -ngl 99

Your hardware

More models your AMD Instinct MI250 128GB can run

Model	Params	Grade	Decode	Capabilities
Devstral 2 123B Instruct	123B	S	31.5 tok/s
Qwen 3.5 122B A10B	122B	S	87.5 tok/s

Frequently asked questions

Can AMD Instinct MI250 128GB run Qwen 3.6 35B A3B?

Yes, AMD Instinct MI250 128GB can run Qwen 3.6 35B A3B with a S grade (Runs well). Expected decode speed: 276.5 tok/s.

How much VRAM does Qwen 3.6 35B A3B need?

Qwen 3.6 35B A3B (35B parameters) requires approximately 40.9 GB of memory with Q4_K_M quantization.

What is the best quantization for Qwen 3.6 35B A3B?

The recommended quantization for Qwen 3.6 35B A3B is Q4_K_M, which balances quality and memory efficiency.

What speed will Qwen 3.6 35B A3B run at on AMD Instinct MI250 128GB?

On AMD Instinct MI250 128GB, Qwen 3.6 35B A3B achieves approximately 276.5 tokens per second decode speed with a time-to-first-token of 700ms using Q4_K_M quantization.

Can AMD Instinct MI250 128GB run Qwen 3.6 35B A3B for coding?

For coding workloads, Qwen 3.6 35B A3B on AMD Instinct MI250 128GB receives a S grade with 276.5 tok/s and 262K context.

What context window can Qwen 3.6 35B A3B use on AMD Instinct MI250 128GB?

On AMD Instinct MI250 128GB, Qwen 3.6 35B A3B can safely use up to 262K tokens of context. The model's official context limit is 262K, but available memory constrains the safe maximum.

See all results for AMD Instinct MI250 128GB See all hardware for Qwen 3.6 35B A3B

Embed this result▼

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

<iframe src="https://willitrunai.com/embed/qwen-3.6-35b-a3b-on-instinct-mi250-128gb" width="400" height="180" frameborder="0" style="border:none;border-radius:12px;overflow:hidden;" title="Will It Run AI — fit result"></iframe>

Preview:

Workload

Grade

Fit

Decode

TTFT

Context

Chat

Runs well

276.5 tok/s

382 ms

262K

Coding

Runs well

276.5 tok/s

700 ms

262K

Agentic Coding

Runs well

276.5 tok/s

1018 ms

262K

Reasoning

Runs well

276.5 tok/s

827 ms

262K

RAG

Runs well

276.5 tok/s

1273 ms

262K

Qwen 3.6 35B A3B inference speed — tokens per second by GPU & Mac

GPU / Mac

Memory

Quant

Speed (tok/s)

Fits?

RTX 5090 32GB

32 GB

Q4_K_M

152.6

Offloads

Mac Studio M3 Ultra 256GB

256 GB

Q4_K_M

70.8

Fits

Mac Studio M2 Ultra 128GB

128 GB

Q4_K_M

59.0

Fits

Mac Studio M1 Ultra 128GB

128 GB

Q4_K_M

55.9

Fits

MacBook Pro M4 Max 128GB

128 GB

Q4_K_M

43.7

Fits

MacBook Pro M4 Max 64GB

64 GB

Q4_K_M

43.7

Fits

RTX 4090 24GB

24 GB

Q4_K_M

34.1

Too big

RX 7900 XTX 24GB

24 GB

Q4_K_M

30.8

Too big

MacBook Pro M3 Max 64GB

64 GB

Q4_K_M

30.5

Fits

RTX 3090 24GB

24 GB

Q4_K_M

29.2

Too big

MacBook Pro M1 Max 64GB

64 GB

Q4_K_M

28.0

Fits

MacBook Pro M4 Pro 48GB

48 GB

Q4_K_M

26.7

Tight

RTX 4080 Super 16GB

16 GB

Q4_K_M

12.2

Too big

RTX 4070 12GB

12 GB

Q4_K_M

5.5

Too big

RTX 3060 12GB

12 GB

Q4_K_M

3.4

Too big

RTX 4060 8GB

8 GB

Q4_K_M

2.9

Too big

Quantization options

How Qwen 3.6 35B A3B (35B params) fits at each quantization level on AMD Instinct MI250 128GB (128.0 GB usable).

Quant	Bits	VRAM	Quality	Fit
Q2_K	2	13.7 GB	Low	A81
Q3_K_S	3	17.2 GB	Low	A81
NVFP4	4	19.6 GB	Medium	A81
Q4_K_M	4	21.3 GB	Medium	A81
Q5_K_M	5	25.2 GB	High	A82
Q6_K	6	28.7 GB	High	A83
Q8_0	8	37.5 GB	Very High	A84
F16Best for your GPU	16	71.8 GB	Maximum	S90

Model

Params

Grade

Decode

Capabilities

Devstral 2 123B Instruct

123B

31.5 tok/s

Qwen 3.5 122B A10B

122B

87.5 tok/s

Frequently asked questions

Can AMD Instinct MI250 128GB run Qwen 3.6 35B A3B?

Yes, AMD Instinct MI250 128GB can run Qwen 3.6 35B A3B with a S grade (Runs well). Expected decode speed: 276.5 tok/s.

How much VRAM does Qwen 3.6 35B A3B need?

Qwen 3.6 35B A3B (35B parameters) requires approximately 40.9 GB of memory with Q4_K_M quantization.

What is the best quantization for Qwen 3.6 35B A3B?

The recommended quantization for Qwen 3.6 35B A3B is Q4_K_M, which balances quality and memory efficiency.

What speed will Qwen 3.6 35B A3B run at on AMD Instinct MI250 128GB?

On AMD Instinct MI250 128GB, Qwen 3.6 35B A3B achieves approximately 276.5 tokens per second decode speed with a time-to-first-token of 700ms using Q4_K_M quantization.

Can AMD Instinct MI250 128GB run Qwen 3.6 35B A3B for coding?

For coding workloads, Qwen 3.6 35B A3B on AMD Instinct MI250 128GB receives a S grade with 276.5 tok/s and 262K context.

What context window can Qwen 3.6 35B A3B use on AMD Instinct MI250 128GB?

On AMD Instinct MI250 128GB, Qwen 3.6 35B A3B can safely use up to 262K tokens of context. The model's official context limit is 262K, but available memory constrains the safe maximum.