How much VRAM does Gemma 4 12B need?

At Q4_K_M the weights are 6.7 GB; with an 8K context KV cache and runtime overhead it needs 8.2 GB, so a 11 GB card runs it with headroom.

Can a GeForce RTX 4070 run Gemma 4 12B?

Yes — Q4_K_M fits with headroom at 8K context, about 45 tokens per second estimated.

Can a GeForce RTX 4090 run Gemma 4 12B?

Yes — Q8_0 fits with headroom at 8K context, about 51 tokens per second estimated.

Can a GeForce RTX 3060 12 GB run Gemma 4 12B?

Yes — Q4_K_M fits with headroom at 8K context, about 32 tokens per second estimated.

Can a M3 Max · 64 GB run Gemma 4 12B?

Yes — Q8_0 fits with headroom at 8K context, about 19 tokens per second estimated.

Hardware requirements

What hardware do I need to run Gemma 4 12B?

12B 256K context Apache 2.0

At Q4_K_M the weights are 6.7 GB; with an 8K KV cache and runtime overhead that is 8.2 GB, so it wants a 11 GB card to run with headroom — in practice 11 GB VRAM: GeForce RTX 2080 Ti and 1 other. Every figure below is computed from the model’s published shape against each device’s usable memory; tokens per second are estimates. The method.

Minimum practical 8 GB VRAM

GeForce RTX 3070 Ti — Q3_K_M, 6.9 GB of 7.0 GB, ~67 tok/s, up to 10K context.

Also: Arc A750, GeForce RTX 5060, GeForce RTX 3070, GeForce RTX 3060 Ti, GeForce RTX 4060 Ti +2 more

the smallest card that holds it at all, at Q3_K_M — the lowest quantisation worth running

Recommended 11 GB VRAM

GeForce RTX 2080 Ti — Q4_K_M, 8.2 GB of 9.7 GB, ~55 tok/s, up to 32K context.

Also: GeForce GTX 1080 Ti

runs the recommended quantisation with headroom at 8K context

High quality 20 GB VRAM

Radeon RX 7900 XT — Q8_0, 13.3 GB of 18.4 GB, ~41 tok/s, up to 89K context.

Also: RTX 4000 Ada

near-lossless Q8_0 with headroom

Long context 24 GB VRAM

GeForce RTX 4090 — Q4_K_M, 15.7 GB of 22.4 GB, ~90 tok/s at 128K.

Also: Radeon RX 7900 XTX, GeForce RTX 3090 Ti, GeForce RTX 3090, GeForce RTX 5090 Laptop, RTX A5000 +3 more

128K tokens (or the model’s maximum) at the recommended quantisation

Apple Silicon 16 GB unified

M1 · 16 GB — Q4_K_M, 8.2 GB of 10.7 GB, ~5.6 tok/s, up to 48K context.

smallest Mac that runs the recommended quantisation with headroom

CPU only 32 GB RAM

CPU only · DDR4 dual-channel — Q8_0, 13.3 GB of 25.6 GB, ~1.8 tok/s, up to 205K context.

no GPU — weights stream from system RAM

How much VRAM at each quantisation

Quant	Weights	+KV 8K	+KV 32K	+KV 128K	Total @ 8K	Needs	Quality
F16	22.4 GB	0.81 GB	2.31 GB	8.3 GB	23.8 GB	32 GB card	Reference
Q8_0	11.9 GB	0.81 GB	2.31 GB	8.3 GB	13.3 GB	20 GB card	−0.1% ppl
Q6_K	9.2 GB	0.81 GB	2.31 GB	8.3 GB	10.6 GB	16 GB card	−0.4% ppl
Q5_K_M	7.9 GB	0.81 GB	2.31 GB	8.3 GB	9.3 GB	16 GB card	−0.8% ppl
Q4_K_M	6.7 GB	0.81 GB	2.31 GB	8.3 GB	8.2 GB	11 GB card	−1.9% ppl
Q3_K_M	5.5 GB	0.81 GB	2.31 GB	8.3 GB	6.9 GB	10 GB card	−5.4% ppl
Q2_K	4.7 GB	0.81 GB	2.31 GB	8.3 GB	6.1 GB	10 GB card	−15% ppl

Totals add 0.6 GB runtime overhead and an f16 KV cache; "needs" is the smallest discrete card class whose usable memory (after display and driver reserve) leaves 15% headroom. A q8_0 cache roughly halves the KV columns.

Every device, checked

System RAM for offload:

16 32 64 128 256

NVIDIA · GeForce

Device	Memory	Best quant with headroom	Fits at all	Max context @ Q4_K_M	Tok/s est.	128K
GeForce RTX 5090	32 GB	Q8_0	Q8_0	256K	161	yes	Check
GeForce RTX 5080	16 GB	Q6_K	Q8_0	107K	86	no	Check
GeForce RTX 5070 Ti	16 GB	Q6_K	Q8_0	107K	80	no	Check
GeForce RTX 5070	12 GB	Q4_K_M	Q6_K	47K	60	no	Check
GeForce RTX 5060 Ti 16 GB	16 GB	Q6_K	Q8_0	107K	40	no	Check
GeForce RTX 5060	8 GB	—	Q3_K_M	—	—	no	Check
GeForce RTX 4090	24 GB	Q8_0	Q8_0	235K	90	yes	Check
GeForce RTX 4080 Super	16 GB	Q6_K	Q8_0	107K	66	no	Check
GeForce RTX 4080	16 GB	Q6_K	Q8_0	107K	64	no	Check
GeForce RTX 4070 Ti Super	16 GB	Q6_K	Q8_0	107K	60	no	Check
GeForce RTX 4070 Ti	12 GB	Q4_K_M	Q6_K	47K	45	no	Check
GeForce RTX 4070 Super	12 GB	Q4_K_M	Q6_K	47K	45	no	Check
GeForce RTX 4070	12 GB	Q4_K_M	Q6_K	47K	45	no	Check
GeForce RTX 4060 Ti 16 GB	16 GB	Q6_K	Q8_0	107K	26	no	Check
GeForce RTX 4060 Ti	8 GB	—	Q3_K_M	—	—	no	Check
GeForce RTX 4060	8 GB	—	Q3_K_M	—	—	no	Check
GeForce RTX 3090 Ti	24 GB	Q8_0	Q8_0	235K	90	yes	Check
GeForce RTX 3090	24 GB	Q8_0	Q8_0	235K	84	yes	Check
GeForce RTX 3080 Ti	12 GB	Q4_K_M	Q6_K	47K	82	no	Check
GeForce RTX 3080 12 GB	12 GB	Q4_K_M	Q6_K	47K	82	no	Check
GeForce RTX 3080	10 GB	Q3_K_M	Q4_K_M	18K	68	no	Check
GeForce RTX 3070 Ti	8 GB	—	Q3_K_M	—	—	no	Check
GeForce RTX 3070	8 GB	—	Q3_K_M	—	—	no	Check
GeForce RTX 3060 Ti	8 GB	—	Q3_K_M	—	—	no	Check
GeForce RTX 3060 12 GB	12 GB	Q4_K_M	Q6_K	47K	32	no	Check
GeForce RTX 2080 Ti	11 GB	Q4_K_M	Q5_K_M	32K	55	no	Check
GeForce RTX 2060 12 GB	12 GB	Q4_K_M	Q6_K	47K	30	no	Check
GeForce GTX 1080 Ti	11 GB	Q4_K_M	Q5_K_M	32K	43	no	Check

NVIDIA · GeForce laptop

Device	Memory	Best quant with headroom	Fits at all	Max context @ Q4_K_M	Tok/s est.	128K
GeForce RTX 5090 Laptop	24 GB	Q8_0	Q8_0	235K	80	yes	Check
GeForce RTX 4090 Laptop	16 GB	Q6_K	Q8_0	107K	52	no	Check
GeForce RTX 4080 Laptop	12 GB	Q4_K_M	Q6_K	47K	39	no	Check
GeForce RTX 4070 Laptop	8 GB	—	Q3_K_M	—	—	no	Check

NVIDIA · Workstation

Device	Memory	Best quant with headroom	Fits at all	Max context @ Q4_K_M	Tok/s est.	128K
RTX 6000 Ada	48 GB	Q8_0	Q8_0	256K	86	yes	Check
RTX 5000 Ada	32 GB	Q8_0	Q8_0	256K	52	yes	Check
RTX 4000 Ada	20 GB	Q8_0	Q8_0	171K	32	tight	Check
RTX 2000 Ada	16 GB	Q6_K	Q8_0	107K	20	no	Check
RTX A6000	48 GB	Q8_0	Q8_0	256K	69	yes	Check
RTX A5000	24 GB	Q8_0	Q8_0	235K	69	yes	Check
RTX A4000	16 GB	Q6_K	Q8_0	107K	40	no	Check

NVIDIA · Data centre

Device	Memory	Best quant with headroom	Fits at all	Max context @ Q4_K_M	Tok/s est.	128K
L40S	48 GB	Q8_0	Q8_0	256K	78	yes	Check
L4	24 GB	Q8_0	Q8_0	235K	27	yes	Check
A10	24 GB	Q8_0	Q8_0	235K	54	yes	Check
A100 40 GB	40 GB	Q8_0	Q8_0	256K	140	yes	Check
A100 80 GB	80 GB	Q8_0	Q8_0	256K	183	yes	Check
H100 PCIe	80 GB	Q8_0	Q8_0	256K	179	yes	Check
H100 SXM	80 GB	Q8_0	Q8_0	256K	301	yes	Check
H200 SXM	141 GB	Q8_0	Q8_0	256K	431	yes	Check
Tesla P40	24 GB	Q8_0	Q8_0	235K	31	yes	Check

AMD · Radeon

Device	Memory	Best quant with headroom	Fits at all	Max context @ Q4_K_M	Tok/s est.	128K
Radeon RX 9070 XT	16 GB	Q6_K	Q8_0	107K	58	no	Check
Radeon RX 9070	16 GB	Q6_K	Q8_0	107K	58	no	Check
Radeon RX 7900 XTX	24 GB	Q8_0	Q8_0	235K	86	yes	Check
Radeon RX 7900 XT	20 GB	Q8_0	Q8_0	171K	72	tight	Check
Radeon RX 7900 GRE	16 GB	Q6_K	Q8_0	107K	52	no	Check
Radeon RX 7800 XT	16 GB	Q6_K	Q8_0	107K	56	no	Check
Radeon RX 7700 XT	12 GB	Q4_K_M	Q6_K	47K	39	no	Check
Radeon RX 7600 XT	16 GB	Q6_K	Q8_0	107K	26	no	Check
Radeon RX 6900 XT	16 GB	Q6_K	Q8_0	107K	46	no	Check
Radeon RX 6800 XT	16 GB	Q6_K	Q8_0	107K	46	no	Check
Radeon RX 6700 XT	12 GB	Q4_K_M	Q6_K	47K	34	no	Check

AMD · Radeon Pro

Device	Memory	Best quant with headroom	Fits at all	Max context @ Q4_K_M	Tok/s est.	128K
Radeon Pro W7900	48 GB	Q8_0	Q8_0	256K	78	yes	Check
Radeon Pro W7800	32 GB	Q8_0	Q8_0	256K	52	yes	Check

AMD · Instinct

Device	Memory	Best quant with headroom	Fits at all	Max context @ Q4_K_M	Tok/s est.	128K
Instinct MI210	64 GB	Q8_0	Q8_0	256K	147	yes	Check
Instinct MI300X	192 GB	Q8_0	Q8_0	256K	476	yes	Check

AMD · Ryzen AI Max

Device	Memory	Best quant with headroom	Fits at all	Max context @ Q4_K_M	Tok/s est.	128K
Ryzen AI Max+ 395 · 128 GB	128 GB	Q8_0	Q8_0	256K	21	yes	Check
Ryzen AI Max+ 395 · 64 GB	64 GB	Q8_0	Q8_0	256K	21	yes	Check

Apple · Apple Silicon

Device	Memory	Best quant with headroom	Fits at all	Max context @ Q4_K_M	Tok/s est.	128K
M1 · 16 GB	16 GB	Q4_K_M	Q6_K	48K	5.6	no	Check
M1 Pro · 32 GB	32 GB	Q8_0	Q8_0	256K	17	yes	Check
M1 Max · 64 GB	64 GB	Q8_0	Q8_0	256K	33	yes	Check
M1 Ultra · 128 GB	128 GB	Q8_0	Q8_0	256K	66	yes	Check
M2 · 24 GB	24 GB	Q8_0	Q8_0	165K	8.3	tight	Check
M2 Pro · 32 GB	32 GB	Q8_0	Q8_0	256K	17	yes	Check
M2 Max · 96 GB	96 GB	Q8_0	Q8_0	256K	33	yes	Check
M2 Ultra · 192 GB	192 GB	Q8_0	Q8_0	256K	66	yes	Check
M3 · 24 GB	24 GB	Q8_0	Q8_0	165K	8.3	tight	Check
M3 Pro · 36 GB	36 GB	Q8_0	Q8_0	256K	12	yes	Check
M3 Max · 64 GB	64 GB	Q8_0	Q8_0	256K	33	yes	Check
M3 Max · 128 GB	128 GB	Q8_0	Q8_0	256K	33	yes	Check
M3 Ultra · 256 GB	256 GB	Q8_0	Q8_0	256K	66	yes	Check
M3 Ultra · 512 GB	512 GB	Q8_0	Q8_0	256K	66	yes	Check
M4 · 32 GB	32 GB	Q8_0	Q8_0	256K	9.9	yes	Check
M4 Pro · 48 GB	48 GB	Q8_0	Q8_0	256K	23	yes	Check
M4 Pro · 64 GB	64 GB	Q8_0	Q8_0	256K	23	yes	Check
M4 Max · 64 GB	64 GB	Q8_0	Q8_0	256K	34	yes	Check
M4 Max · 128 GB	128 GB	Q8_0	Q8_0	256K	45	yes	Check

Intel · Arc

Device	Memory	Best quant with headroom	Fits at all	Max context @ Q4_K_M	Tok/s est.	128K
Arc B580	12 GB	Q4_K_M	Q6_K	47K	41	no	Check
Arc B570	10 GB	Q3_K_M	Q4_K_M	18K	34	no	Check
Arc A770 16 GB	16 GB	Q6_K	Q8_0	107K	50	no	Check
Arc A750	8 GB	—	Q3_K_M	—	—	no	Check

CPU · CPU + system RAM

Device	Memory	Best quant with headroom	Fits at all	Max context @ Q4_K_M	Tok/s est.	128K
CPU only · DDR4 dual-channel	32 GB	Q8_0	Q8_0	256K	3.2	yes	Check
CPU only · DDR5 dual-channel	64 GB	Q8_0	Q8_0	256K	5.6	yes	Check
CPU only · DDR5 8-channel server	256 GB	Q8_0	Q8_0	256K	19	yes	Check