Can 1-bit Bonsai 27B run on GTX 1660 Super 6GB?

YES — With Offload

A75Great

Estimated from fit model

1-bit Bonsai 27B needs ~6.4 GB VRAM. GTX 1660 Super 6GB has 6.0 GB. With Q1_0_G128 quantization, expect ~26 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

Q1_0_G128 (Very Low quality) — 6.4 GB, 26.3 tok/s, Runs with offload (needs ~0.2 GB host RAM)

6.4 GB required6.0 GB available

107% VRAM needed

0.4 GB over capacity — needs offload or smaller quantization

Fit status

Runs with offload (needs ~0.2 GB host RAM)

Decode

26.3 tok/s

TTFT

7373 ms

Safe context

10K

Memory

6.4 GB / 6.0 GB

Offload

10%

Memory breakdown

Weights3.9 GB

KV Cache1.0 GB

Runtime0.9 GB

Headroom0.6 GB

See how fast it feels

See how fast it feels1-bit Bonsai 27B on GTX 1660 Super 6GB
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: 26.3 tok/s decode · 7.4s TTFT (warm) · 66 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 0.2 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	S	Runs with offload	41.0 tok/s	2577 ms	10K
Coding	A	Runs with offload (needs ~0.2 GB host RAM)	26.3 tok/s	7373 ms	10K
Agentic Coding	F	Too heavy	19.2 tok/s	14674 ms	10K
Reasoning	A	Runs with offload (needs ~0.2 GB host RAM)	26.3 tok/s	8714 ms	10K
RAG	F	Too heavy	19.2 tok/s	18342 ms	10K

Inference speed

1-bit Bonsai 27B inference speed — tokens per second by GPU & Mac

Estimated decode speed (tokens/sec) for 1-bit Bonsai 27B at Q1_0_G128 across popular GPUs and Apple Silicon, using the fastest local runtime per device. Fastest is RTX 5090 32GB at ~156 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	Q1_0_G128	156.1	Fits
RX 7900 XTX 24GB	24 GB	Q1_0_G128	132.5	Fits
Mac Studio M3 Ultra 256GB	256 GB	Q1_0_G128	123.5	Fits
MacBook Pro M4 Max 128GB	128 GB	Q1_0_G128	122.1	Fits
MacBook Pro M4 Max 64GB	64 GB	Q1_0_G128	122.1	Fits
Mac Studio M2 Ultra 128GB	128 GB	Q1_0_G128	102.9	Fits
Mac Studio M1 Ultra 128GB	128 GB	Q1_0_G128	97.6	Fits
RTX 4090 24GB	24 GB	Q1_0_G128	89.7	Fits
RTX 3090 24GB	24 GB	Q1_0_G128	87.4	Fits
RTX 4080 Super 16GB	16 GB	Q1_0_G128	86.1	Fits
MacBook Pro M4 Pro 48GB	48 GB	Q1_0_G128	76.9	Fits
RTX 3060 12GB	12 GB	Q1_0_G128	69.5	Fits
RTX 4070 12GB	12 GB	Q1_0_G128	53.3	Fits
MacBook Pro M3 Max 64GB	64 GB	Q1_0_G128	53.2	Fits
MacBook Pro M1 Max 64GB	64 GB	Q1_0_G128	48.8	Fits
RTX 4060 8GB	8 GB	Q1_0_G128	28.0	Tight

Estimates for single-stream decoding at Q1_0_G128; 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 1-bit Bonsai 27B (27B params) fits at each quantization level on GTX 1660 Super 6GB (6.0 GB usable).

Quant	Bits	VRAM	Quality	Fit
Q1_0_G128	1.125	3.9 GB	Very Low	F0
Q2_0_G128	1.71	7.2 GB	Low	F0
Q2_K	2	10.5 GB	Low	F0
Q3_K_S	3	13.2 GB	Low	F0
NVFP4	4	15.1 GB	Medium	F0
Q4_K_M	4	16.5 GB	Medium	F0
Q5_K_M	5	19.4 GB	High	F0
Q6_K	6	22.1 GB	High	F0
Q8_0	8	28.9 GB	Very High	F0
F16	16	55.4 GB	Maximum	F0

Get started

Copy-paste commands to run 1-bit Bonsai 27B on your machine.

Run

docker run --rm -it ghcr.io/ggerganov/llama.cpp:full \
  --hf-repo "prism-ml/Bonsai-27B-gguf" \
  --hf-file "Bonsai-27B-gguf-Q1_0_G128.gguf" \
  -c 4096 -ngl 99

Frequently asked questions

Can GTX 1660 Super 6GB run 1-bit Bonsai 27B?

Yes, GTX 1660 Super 6GB can run 1-bit Bonsai 27B with a A grade (Runs with offload (needs ~0.2 GB host RAM)). Expected decode speed: 26.3 tok/s.

How much VRAM does 1-bit Bonsai 27B need?

1-bit Bonsai 27B (27B parameters) requires approximately 6.4 GB of memory with Q1_0_G128 quantization.

What is the best quantization for 1-bit Bonsai 27B?

The recommended quantization for 1-bit Bonsai 27B is Q1_0_G128, which balances quality and memory efficiency.

What speed will 1-bit Bonsai 27B run at on GTX 1660 Super 6GB?

On GTX 1660 Super 6GB, 1-bit Bonsai 27B achieves approximately 26.3 tokens per second decode speed with a time-to-first-token of 7373ms using Q1_0_G128 quantization.

Can GTX 1660 Super 6GB run 1-bit Bonsai 27B for coding?

For coding workloads, 1-bit Bonsai 27B on GTX 1660 Super 6GB receives a A grade with 26.3 tok/s and 10K context.

What context window can 1-bit Bonsai 27B use on GTX 1660 Super 6GB?

On GTX 1660 Super 6GB, 1-bit Bonsai 27B can safely use up to 10K tokens of context. The model's official context limit is 262K, but available memory constrains the safe maximum.

What should I upgrade first if 1-bit Bonsai 27B feels slow on GTX 1660 Super 6GB?

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 GTX 1660 Super 6GB See all hardware for 1-bit Bonsai 27B

Embed this result▼

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

<iframe src="https://willitrunai.com/embed/bonsai-27b-on-gtx-1660-super-6gb" width="400" height="180" frameborder="0" style="border:none;border-radius:12px;overflow:hidden;" title="Will It Run AI — fit result"></iframe>

Preview: