Computed

Llama 3.3 70B Instruct on an Apple M4 Max (128GB)

Yes, quantized. Q8_0 is the largest that fits, leaving room for 84k tokens of context.

Change the numbers →Llama 3.3 70B Instruct on every card Apple M4 Max (128GB) across every model

Verdict

Q8_0

96.0 GiB usable of 128 GB

Weights (Q4_K_M)

39.6 GiB

131.4 GiB at FP16 · 70.6B params · 6 of 6 quant rows are measured files

KV cache

320.0 KiB per token at FP16

80 layers × 8 KV heads × 128 dimensions

Speed ceiling

7 tok/s

546 GB/s ÷ bytes read per token

Every quant, against this card

Weight bytes are the size of the real published file wherever one exists — 6 of these6 rows are measured from bartowski/Llama-3.3-70B-Instruct-GGUF, the rest computed from the parameter count. Max context is what the KV cache can grow to in whatever memory the weights leave behind, capped at the 128k tokens this model was trained to address.

Bars are the weight bytes at each precision; the dashed line is the usable memory of an Apple M4 Max (128GB). Drawn from the computed byte counts, not sketched.

Precision	Bits/weight	Weights	Source	Fits	Max context	Ceiling
FP16 / BF16	16.00	131.4 GiB	measured file	−35.4 GiB	—	1.6 tok/s offloaded
Q8_0	8.50	69.8 GiB	measured file	yes	85,767	7 tok/s
Q6_K	6.56	53.9 GiB	measured file	yes	131,072 (model cap)	9 tok/s
Q5_K_M	5.66	46.5 GiB	measured file	yes	131,072 (model cap)	10 tok/s
Q4_K_M	4.82	39.6 GiB	measured file	yes	131,072 (model cap)	12 tok/s
Q3_K_M	3.89	31.9 GiB	measured file	yes	131,072 (model cap)	15 tok/s

What the context actually costs

The KV cache is 2 · layers · kv_heads · head_dim bytes per token per element — 2 · 80 · 8 · 128 · 2 B = 320.0 KiB. It depends on the key/value head count, not the attention-head count and not the parameter count. This model shares 8 KV heads across 64 query heads, which divides the cache by 8 against multi-head attention.

Context	KV cache (FP16)	KV cache (8-bit)	Plus Q4_K_M weights
4,096	1.3 GiB	640 MiB	40.9 GiB
8,192	2.5 GiB	1.3 GiB	42.1 GiB
32,768	10.0 GiB	5.0 GiB	49.6 GiB
131,072	40.0 GiB	20.0 GiB	79.6 GiB

The speed ceiling, and where it comes from

Generating one token reads every active weight from memory once. At Q8_0 that is 69.8 GiB, plus a pass over the KV cache. An Apple M4 Max (128GB) moves 546 GB/s, so the arithmetic ceiling is 7 tok/s. Treat it as a bound, not an estimate: attention overhead, kernel launches and imperfect memory access keep real runtimes at roughly 60–80% of it, and nothing pushes past it.

Where these numbers come from

The model

Parameters: 70,553,706,496
Layers: 80
Attention / KV heads: 64 / 8
Head dimension: 128
Trained context: 131,072
Checkpoint as published: 131.4 GiB

Read from unsloth/Llama-3.3-70B-Instruct — an ungated mirror of meta-llama/Llama-3.3-70B-Instruct, whose own config cannot be fetched without an access token. The parameter count is the Hub's own total over the tensor shapes, not a figure taken from the model's name.

The accelerator

Memory: 128 GB LPDDR5X unified
Bandwidth: 546 GB/s
Assumed usable: 75% → 96.0 GiB

Capacity and bandwidth from the vendor's specification. The usable fraction is an assumption, not a spec: unified memory is shared with the OS and the display, and the GPU working-set cap is raisable on Apple silicon with `sudo sysctl iogpu.wired_limit_mb`.

Questions this pairing answers

How much VRAM does Llama 3.3 70B Instruct need?

131.4 GiB for the weights at FP16 — 70.6B parameters at two bytes each — and 39.6 GiB at Q4_K_M. The KV cache is on top of that and is not a fixed number: this model spends 320.0 KiB per token of context, so 8,192 tokens costs a further 2.5 GiB. An Apple M4 Max (128GB) makes 96.0 GiB of its 128 GB available on the assumption below.

Can an Apple M4 Max (128GB) run Llama 3.3 70B Instruct?

Yes, quantized. Q8_0 is the largest that fits, leaving room for 84k tokens of context. That is the weights and the KV cache together, against 96.0 GiB of usable memory.

How fast will Llama 3.3 70B Instruct run on an Apple M4 Max (128GB)?

No faster than 7 tokens/second at Q8_0, and in practice below it. Decoding is memory-bound: every token reads all 69.8 GiB of weights plus the cache, and this card moves 546 GB/s. That division is the ceiling — no kernel, runtime or driver beats it, and a real runtime typically reaches 60–80% of it.

Why does the context length change how much memory Llama 3.3 70B Instruct needs?

Because the KV cache holds one key and one value vector per token, per layer, for the whole conversation, and it is allocated separately from the weights. This model has 80 layers and 8 key/value heads of 128 dimensions, shared across 64 query heads — grouped-query attention, which divides the cache by 8. That works out at 320.0 KiB per token. Parameter count tells you nothing about this number.

The same model on a different card

A different model on the same card

All 62 models on Apple M4 Max (128GB) →

Method and limits. Weight bytes are the byte size of the real published file wherever one exists, and the model's exact parameter count times the published llama.cpp bits-per-weight where it does not. That distinction is on every row above and it matters at both ends: a sub-1B model's Q4_K_M file runs a third larger than the nominal figure because k-quants keep its embedding tables at higher precision, and an already-4-bit release cannot be quantized upward at all. The KV cache is 2 · layers · kv_heads · head_dim · bytes per token, summed over layers with each sliding-window layer capped at its window. The speed figure is a roofline bound, not a benchmark: bandwidth divided by bytes read per token, which no runtime exceeds and every runtime falls short of. The usable fraction of card memory is an assumption: 75% here, which is what this lane assumes for unified memory, where the OS and window server share the same pool — the other class assumes 92%, so it is not one number applied to every device. That is the only assumed input on this page; every other figure is computed from the model config and the card's published specification. Nothing on this page is written by a language model. Architecture from the model's published config, fetched 2026-08-07.