How fast is Llama 3.3 70B on an Mac?

Around 7 tokens per second. Token generation on Apple Silicon is limited by memory bandwidth, and this machine has 400 GB/s.

How much memory does Llama 3.3 70B need?

40.3 GiB for the weights at Q4_K_M, plus 2.5 GiB for an 8K-token KV cache and about 2.8 GiB of runtime overhead — 45.6 GiB in total.

Home › Llama 3.3 70B › M2 Max 32GB

Can an M2 Max 32GB Mac run Llama 3.3 70B?

M2 Max · 32 GB unified memory · 400 GB/s · MacBook Pro 14" (M2 Max), Mac Studio (M2 Max)

No — not at Q4_K_M.

Llama 3.3 70B needs about 45.6 GiB but an M2 Max 32GB Mac has only ~27.2 GiB available for a model — you are 18.4 GiB short. macOS will swap to SSD and generation will crawl.

Weights

40.3 GiB

70.6B params @ Q4_K_M

KV cache

2.5 GiB

8K context, FP16

Overhead

2.8 GiB

runtime + activations

Total needed

45.6 GiB

of ~27.2 GiB usable

Est. speed

7 tok/s

400 GB/s bandwidth

Max context

n/a

at Q4_K_M

Where the memory goes

Three things occupy memory when a model is loaded: the weights themselves, the KV cache that holds the conversation, and the runtime's own working buffers. Only the first is fixed — the KV cache grows with every token in your context.

Quantisation	Weights	Total	Tok/s	Fits?
FP16	131.5 GiB	141.4 GiB	2	No
Q8_0	69.9 GiB	76.7 GiB	4	No
Q6_K	54.2 GiB	60.3 GiB	5	No
Q5_K_M	46.8 GiB	52.5 GiB	6	No
Q4_K_M	40.3 GiB	45.6 GiB	7	No
Q3_K_M	32.1 GiB	37.0 GiB	9	No
Q2_K	24.7 GiB	29.2 GiB	12	No

The default. Best quality-per-gigabyte for most people.

How long a context fits

The KV cache is often what breaks a setup that looked fine at load time. Quantising it to 8-bit roughly halves its footprint with little measurable quality cost.

Context	KV cache	Total	FP16 KV	Q8 KV
2K	0.6 GiB	43.5 GiB	No	No
4K	1.3 GiB	44.2 GiB	No	No
8K	2.5 GiB	45.6 GiB	No	No
16K	5.0 GiB	48.4 GiB	No	No
32K	10.0 GiB	54.0 GiB	No	No
128K	40.0 GiB	87.6 GiB	No	No

What to run instead on an M2 Max 32GB Mac

These are the largest models that fit comfortably on this machine at Q4_K_M with an 8K context.

Model	Size	Total	Tok/s
DeepSeek-R1-Distill-Qwen-32B	32.8B	22.4 GiB	16	Yes, comfortably
Qwen2.5 32B	32.8B	22.4 GiB	16	Yes, comfortably
Qwen2.5-Coder 32B	32.8B	22.4 GiB	16	Yes, comfortably
Qwen3 32B	32.8B	22.4 GiB	16	Yes, comfortably
Qwen3 30B-A3B	30.5B	19.8 GiB	99	Yes, comfortably
Gemma 3 27B	27.4B	21.1 GiB	19	Yes, comfortably

Macs that run Llama 3.3 70B comfortably

The smallest configuration that handles it is the M1 Max 64GB (MacBook Pro 16" (M1 Max)).

Mac	Memory	Bandwidth	Tok/s
M1 Max 64GB	64 GB	400 GB/s	7
M2 Ultra 64GB	64 GB	800 GB/s	15
M4 Pro 64GB	64 GB	273 GB/s	5
M4 Max 64GB	64 GB	546 GB/s	10
M2 Max 96GB	96 GB	400 GB/s	7
M3 Ultra 96GB	96 GB	819 GB/s	15
M3 Max 128GB	128 GB	400 GB/s	7
M4 Max 128GB	128 GB	546 GB/s	10
M2 Ultra 192GB	192 GB	800 GB/s	15
M3 Ultra 256GB	256 GB	819 GB/s	15
M3 Ultra 512GB	512 GB	819 GB/s	15

Common questions

Can an M2 Max 32GB Mac run Llama 3.3 70B?: No. At Q4_K_M it needs about 45.6 GiB, and an M2 Max 32GB Mac has roughly 27.2 GiB available for a model after macOS takes its share.
How fast is Llama 3.3 70B on an Mac?: Around 7 tokens per second. Token generation on Apple Silicon is limited by memory bandwidth, and this machine has 400 GB/s.
How much memory does Llama 3.3 70B need?: 40.3 GiB for the weights at Q4_K_M, plus 2.5 GiB for an 8K-token KV cache and about 2.8 GiB of runtime overhead — 45.6 GiB in total.