Hardware

For local models, memory bandwidth sets your generation speed and unified memory sets your ceiling. Everything else is secondary. Counts below are models that fit at their best quantization, at your current context length.

M5 generation

Chip	Bandwidth	Memory options	Models that fit	8B q4 speed est.
M5	153 GB/s	16 GB11.2 GB usable	20 of 39	27.9 tok/s
		24 GB17.3 GB usable	25 of 39
		32 GB23.0 GB usable	32 of 39
M5 Prounverified	307 GB/s	24 GB17.3 GB usable	25 of 39	59.1 tok/s
		48 GB38.4 GB usable	33 of 39
		64 GB51.2 GB usable	36 of 39
M5 Maxunverified	614 GB/s	36 GB25.9 GB usable	32 of 39	124 tok/s
		48 GB38.4 GB usable	33 of 39
		64 GB51.2 GB usable	36 of 39
		128 GB108.8 GB usable	37 of 39

M4 generation

Chip	Bandwidth	Memory options	Models that fit	8B q4 speed est.
M4	120 GB/s	16 GB11.2 GB usable	20 of 39	21.6 tok/s
		24 GB17.3 GB usable	25 of 39
		32 GB23.0 GB usable	32 of 39
M4 Pro	273 GB/s	24 GB17.3 GB usable	25 of 39	51.9 tok/s
		48 GB38.4 GB usable	33 of 39
		64 GB51.2 GB usable	36 of 39
M4 Max (14-core CPU)	410 GB/s	36 GB25.9 GB usable	32 of 39	80.9 tok/s
		48 GB38.4 GB usable	33 of 39
M4 Max (16-core CPU)	546 GB/s	48 GB38.4 GB usable	33 of 39	109 tok/s
		64 GB51.2 GB usable	36 of 39
		128 GB108.8 GB usable	37 of 39

M3 generation

Chip	Bandwidth	Memory options	Models that fit	8B q4 speed est.
M3	100 GB/s	8 GB5.6 GB usable	16 of 39	17.8 tok/s
		16 GB11.2 GB usable	20 of 39
		24 GB17.3 GB usable	25 of 39
M3 Pro	150 GB/s	18 GB13.0 GB usable	21 of 39	27.4 tok/s
		36 GB25.9 GB usable	32 of 39
M3 Max (14-core CPU)	300 GB/s	36 GB25.9 GB usable	32 of 39	57.0 tok/s
M3 Max (16-core CPU)	400 GB/s	48 GB38.4 GB usable	33 of 39	77.9 tok/s
		64 GB51.2 GB usable	36 of 39
		96 GB76.8 GB usable	37 of 39
		128 GB108.8 GB usable	37 of 39
M3 Ultra	819 GB/s	96 GB76.8 GB usable	37 of 39	166 tok/s
		256 GB217.6 GB usable	38 of 39
		512 GB435.2 GB usable	39 of 39

M2 generation

Chip	Bandwidth	Memory options	Models that fit	8B q4 speed est.
M2	100 GB/s	8 GB5.6 GB usable	16 of 39	17.8 tok/s
		16 GB11.2 GB usable	20 of 39
		24 GB17.3 GB usable	25 of 39
M2 Pro	200 GB/s	16 GB11.2 GB usable	20 of 39	37.5 tok/s
		32 GB23.0 GB usable	32 of 39
M2 Max	400 GB/s	32 GB23.0 GB usable	32 of 39	77.9 tok/s
		64 GB51.2 GB usable	36 of 39
		96 GB76.8 GB usable	37 of 39
M2 Ultra	800 GB/s	64 GB51.2 GB usable	36 of 39	160 tok/s
		128 GB108.8 GB usable	37 of 39
		192 GB163.2 GB usable	38 of 39

M1 generation

Chip	Bandwidth	Memory options	Models that fit	8B q4 speed est.
M1	68 GB/s	8 GB5.6 GB usable	16 of 39	11.9 tok/s
		16 GB11.2 GB usable	20 of 39
M1 Pro	200 GB/s	16 GB11.2 GB usable	20 of 39	37.0 tok/s
		32 GB23.0 GB usable	32 of 39
M1 Max	400 GB/s	32 GB23.0 GB usable	32 of 39	77.0 tok/s
		64 GB51.2 GB usable	36 of 39
M1 Ultra	800 GB/s	64 GB51.2 GB usable	36 of 39	158 tok/s
		128 GB108.8 GB usable	37 of 39

Bandwidth figures come from Apple's published specifications where available. Rows marked unverified are best-available community figures and are labelled everywhere they appear. TFLOPS values only influence prompt-processing estimates.