How much VRAM does gemma-4-26B-A4B-it-uncensored-heretic need?

Its published quantizations range from 9.86 GiB to 47.04 GiB in weights alone, before KV cache and runtime overhead.

gemma-4-26B-A4B-it-uncensored-heretic on GeForce RTX 2080 Ti

Can I run gemma-4-26B-A4B-it-uncensored-heretic on a GeForce RTX 2080 Ti?

Q: Can I run gemma-4-26B-A4B-it-uncensored-heretic on a GeForce RTX 2080 Ti?

No. No published quantization of gemma-4-26B-A4B-it-uncensored-heretic fits a GeForce RTX 2080 Ti at any context with q8_0 KV cache.

Not at these settings. No indexed quantization of gemma-4-26B-A4B-it-uncensored-heretic fits GeForce RTX 2080 Ti at any context we compute, with q8_0 KV. The smallest shipped quantization is 9.86 GiB in weights alone, against 10.23 GiB usable. CPU offload can still run it, slowly.

From the file· weights summed from filesFrom the file· KV computed per layerPredicted· speed and compute buffer

KV cache dtypef16 q8_0 q4_0— 11 GB card, 10.23 GiB usable after overhead

Every quantization at every context

total memory required; green fits, red does not

Quant	Weights●	4K◐	8K◐	16K◐	32K◐	64K◐	128K◐
BF16	47.04 GiB	48.1	48.1	48.3	48.6	49.3	50.6
Q8_0	25.02 GiB	26.0	26.1	26.3	26.6	27.3	28.6
Q6_K	21.08 GiB	22.1	22.2	22.4	22.7	23.4	24.7
Q5_K_M	17.82 GiB	18.8	18.9	19.1	19.4	20.1	21.4
Q5_K_S	16.75 GiB	17.8	17.9	18.0	18.4	19.0	20.4
Q4_K_M	15.64 GiB	16.7	16.8	16.9	17.2	17.9	19.2
Q4_K_S	14.40 GiB	15.4	15.5	15.7	16.0	16.7	18.0
IQ4_XS	13.10 GiB	14.1	14.2	14.4	14.7	15.4	16.7
Q3_K_L	12.88 GiB	13.9	14.0	14.2	14.5	15.1	16.5
Q3_K_M	12.37 GiB	13.4	13.5	13.6	14.0	14.6	16.0
Q3_K_S	11.38 GiB	12.4	12.5	12.7	13.0	13.7	15.0
Q2_K	9.86 GiB	10.9	11.0	11.1	11.5	12.1	13.5

●From the file◐Predictedwhat these mean

Figures are GiB of total memory: weights plus KV cache plus compute buffer and backend overhead. Weights and KV are near-exact; the overhead term is modeled. Hover any cell for the breakdown.

Why other calculators disagree

A parameters × bits ÷ 8 estimate ignores two things that dominate at long context. First, the weights themselves are not the nominal rate — quantizations are mixtures, so the real file is consistently larger than the label implies. Second, most of this model's layers cache only a 1,024-token window rather than the full context.