Is unified memory on MacBook Air M3 24GB as fast as VRAM for internlm2 limarp chat 20b?

Not always. MacBook Air M3 24GB can often fit larger models thanks to unified memory, but a discrete GPU with dedicated high-bandwidth VRAM may still decode faster once the model fits. For this combination, the important distinction is capacity versus sustained throughput.

Can internlm2 limarp chat 20b run on MacBook Air M3 24GB?

Q: What should I upgrade first if internlm2 limarp chat 20b feels slow on MacBook Air M3 24GB?

Prioritize bandwidth, not only capacity. If this workload feels slow, the next useful step is often a GPU tier with materially faster memory bandwidth rather than only a small bump in capacity.

YES — With Offload

C44Usable

Estimated from fit model

internlm2 limarp chat 20b needs ~18.0 GB VRAM. MacBook Air M3 24GB has 17.3 GB. With Q4_K_M quantization, expect ~5 tok/s.

Runtime: llama.cppCapacity: OffloadBandwidth: Very 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) — 18.0 GB, 5.1 tok/s, Runs with offload (needs ~0.5 GB host RAM)

18.0 GB required17.3 GB available

104% VRAM needed

0.7 GB over capacity — needs offload or smaller quantization

Fit status

Runs with offload (needs ~0.5 GB host RAM)

Decode

5.1 tok/s

TTFT

37715 ms

Safe context

11K

Memory

18.0 GB / 17.3 GB

Memory breakdown

Weights12.2 GB

KV Cache2.3 GB

Runtime0.9 GB

Headroom2.6 GB

See how fast it feels

See how fast it feelsinternlm2 limarp chat 20b on MacBook Air M3 24GB
1st promptCold start — includes initialization
>What is local AI inference?
Local AI inference means running an AI language model directly on your own hardware — your laptop, desktop, or server — instead of sending requests to a remote cloud API.

When you run inference locally the model weights are loaded into your GPU or unified memory. Each token you generate requires reading those weights from memory, so memory bandwidth is the main bottleneck for decode speed.

Key benefits of running locally:
  - Full privacy: your prompts never leave your machine
  - No per-token cost or rate limits
  - Works offline once the model is downloaded
  - Latency depends only on your hardware
2nd promptWarm — model ready, no init overhead
>How much VRAM do I need?
It depends on the model size and quantization level. A rough rule of thumb:

  Model size     Q4 (4-bit)    Q8 (8-bit)    FP16
  7B params      ~4.3 GB       ~7.5 GB       ~14 GB
  13B params     ~7.9 GB       ~13.9 GB      ~26 GB
  70B params     ~42.7 GB      ~74.9 GB      ~140 GB

Most people use 4-bit quantization (Q4_K_M) which gives 90-95% of full quality at a fraction of the memory. A 24 GB GPU can comfortably run most 7B-13B models.
Estimated: 5.1 tok/s decode · 37.7s TTFT (warm) · 13 tok/s prefill

What limits this setup

The model fits in shared memory, but shared-memory bandwidth is now the real limiter.

Fit does not mean dedicated-VRAM speed

Unified or shared memory can make a model technically fit, but sustained tokens per second may still trail a discrete high-bandwidth GPU with less total memory.

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.

Shared-memory contention still exists

The OS, browser, and inference runtime all compete for the same physical memory pool, so real-world headroom is less forgiving than raw capacity suggests.

Best improvement path

Prioritize bandwidth, not only capacity

If this workload feels slow, the next useful step is often a GPU tier with materially faster memory bandwidth rather than only a small bump in capacity.

Buy headroom, not only minimum fit

A slightly larger memory tier gives you safer context growth and makes the recommendation more future-proof.

Performance by workload

Workload	Grade	Fit	Decode	TTFT	Context
Chat	C	Runs with offload	5.6 tok/s	18946 ms	11K
Coding	C	Runs with offload	5.1 tok/s	37715 ms	11K
Agentic Coding	D	Very compromised (needs ~1.9 GB host RAM)	4.3 tok/s	64910 ms	11K
Reasoning	C	Runs with offload (needs ~0.5 GB host RAM)	5.1 tok/s	44572 ms	11K
RAG	D	Very compromised (needs ~1.9 GB host RAM)	4.3 tok/s	81138 ms

Quantization options

How internlm2 limarp chat 20b (20B params) fits at each quantization level on MacBook Air M3 24GB (17.3 GB usable).

Quant	Bits	VRAM	Quality	Fit
Q2_K	2	7.8 GB	Low	C51
Q3_K_S	3	9.8 GB	Low	C51
NVFP4	4

Get started

Copy-paste commands to run internlm2 limarp chat 20b on your machine.

Run

lms load hf-intervitens-archive--internlm2-limarp-chat-20b-gguf && lms server start

Upgrade options

Hardware that runs internlm2 limarp chat 20b well

MacBook Pro M4 32GBBudget pick

32 GB Unified (+8)120 GB/s (+20)

Raises estimated decode speed by about 80%.9.2 tok/s decode

Raises estimated decode speed by about 80%.

Adds memory headroom for longer context windows and future model growth.

~$799 MSRP

Mac mini M4 32GBBest value

32 GB Unified (+8)120 GB/s (+20)

Raises estimated decode speed by about 80%.9.2 tok/s decode

Raises estimated decode speed by about 80%.

Adds memory headroom for longer context windows and future model growth.

~$1,099 MSRP

Mac mini M4 64GBApple upgrade

64 GB Unified (+40)120 GB/s (+20)

Raises estimated decode speed by about 80%.9.2 tok/s decode

Raises estimated decode speed by about 80%.

Adds memory headroom for longer context windows and future model growth.

~$1,099 MSRP

RTX 5090 Laptop 24GBBiggest leap

896 GB/s (+796)

Raises estimated decode speed by about 1110%.61.7 tok/s decode

Raises estimated decode speed by about 1110%.

Adds memory headroom for longer context windows and future model growth.

Frequently asked questions

Can MacBook Air M3 24GB run internlm2 limarp chat 20b?

Yes, MacBook Air M3 24GB can run internlm2 limarp chat 20b with a C grade (Runs with offload). Expected decode speed: 5.1 tok/s.

How much VRAM does internlm2 limarp chat 20b need?

internlm2 limarp chat 20b (20B parameters) requires approximately 18.0 GB of memory with Q4_K_M quantization.

What is the best quantization for internlm2 limarp chat 20b?

The recommended quantization for internlm2 limarp chat 20b is Q4_K_M, which balances quality and memory efficiency.

What speed will internlm2 limarp chat 20b run at on MacBook Air M3 24GB?

On MacBook Air M3 24GB, internlm2 limarp chat 20b achieves approximately 5.1 tokens per second decode speed with a time-to-first-token of 37715ms using Q4_K_M quantization.

Can MacBook Air M3 24GB run internlm2 limarp chat 20b for coding?

For coding workloads, internlm2 limarp chat 20b on MacBook Air M3 24GB receives a C grade with 5.1 tok/s and 11K context.

What context window can internlm2 limarp chat 20b use on MacBook Air M3 24GB?

On MacBook Air M3 24GB, internlm2 limarp chat 20b can safely use up to 11K tokens of context. The model's official context limit is —, but available memory constrains the safe maximum.

See all results for MacBook Air M3 24GB See all hardware for internlm2 limarp chat 20b

Embed this result▼

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

<iframe src="https://willitrunai.com/embed/hf-intervitens-archive--internlm2-limarp-chat-20b-gguf-on-m3-air-24gb" width="400" height="180" frameborder="0" style="border:none;border-radius:12px;overflow:hidden;" title="Will It Run AI — fit result"></iframe>

Preview: