How much VRAM does gpt-oss-20b-uncensored need?

Its published quantizations range from 11.19 GiB to 20.73 GiB in weights alone, before KV cache and runtime overhead.

gpt-oss-20b-uncensored on GeForce RTX 2080 Ti

Can I run gpt-oss-20b-uncensored on a GeForce RTX 2080 Ti?

Q: Can I run gpt-oss-20b-uncensored on a GeForce RTX 2080 Ti?

No. No published quantization of gpt-oss-20b-uncensored fits a GeForce RTX 2080 Ti at any context with q8_0 KV cache.

Not at these settings. No indexed quantization of gpt-oss-20b-uncensored fits GeForce RTX 2080 Ti at any context we compute, with q8_0 KV. The smallest shipped quantization is 11.19 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◐
Q8_0	20.73 GiB	21.6	21.6	21.7	21.9	22.3	23.1
I1-Q6_K	20.67 GiB	21.5	21.6	21.7	21.9	22.3	23.1
Q6_K	20.67 GiB	21.5	21.6	21.7	21.9	22.3	23.1
I1-Q5_K_M	15.73 GiB	16.6	16.6	16.7	16.9	17.3	18.1
Q5_K_M	15.73 GiB	16.6	16.6	16.7	16.9	17.3	18.1
I1-Q5_K_S	14.80 GiB	15.6	15.7	15.8	16.0	16.4	17.2
Q5_K_S	14.80 GiB	15.6	15.7	15.8	16.0	16.4	17.2
I1-Q4_K_M	14.72 GiB	15.6	15.6	15.7	15.9	16.3	17.1
Q4_K_M	14.72 GiB	15.6	15.6	15.7	15.9	16.3	17.1
I1-Q4_K_S	13.65 GiB	14.5	14.5	14.6	14.8	15.2	16.0
Q4_K_S	13.65 GiB	14.5	14.5	14.6	14.8	15.2	16.0
I1-Q4_1	12.45 GiB	13.3	13.3	13.4	13.6	14.0	14.8
I1-Q3_K_L	12.42 GiB	13.3	13.3	13.4	13.6	14.0	14.8
Q3_K_L	12.42 GiB	13.3	13.3	13.4	13.6	14.0	14.8
I1-Q3_K_M	12.03 GiB	12.9	12.9	13.0	13.2	13.6	14.4
Q3_K_M	12.03 GiB	12.9	12.9	13.0	13.2	13.6	14.4
IQ4_XS	11.40 GiB	12.3	12.3	12.4	12.6	13.0	13.8
I1-IQ3_M	11.36 GiB	12.2	12.3	12.4	12.6	13.0	13.8
I1-Q4_0	11.31 GiB	12.2	12.2	12.3	12.5	12.9	13.7
I1-Q2_K_S	11.30 GiB	12.2	12.2	12.3	12.5	12.9	13.7
I1-IQ4_XS	11.27 GiB	12.1	12.2	12.3	12.5	12.9	13.7
I1-IQ2_M	11.24 GiB	12.1	12.1	12.2	12.4	12.8	13.6
I1-IQ2_S	11.24 GiB	12.1	12.1	12.2	12.4	12.8	13.6
I1-IQ3_S	11.24 GiB	12.1	12.1	12.2	12.4	12.8	13.6
I1-IQ3_XS	11.24 GiB	12.1	12.1	12.2	12.4	12.8	13.6
I1-IQ3_XXS	11.24 GiB	12.1	12.1	12.2	12.4	12.8	13.6
I1-Q2_K	11.24 GiB	12.1	12.1	12.2	12.4	12.8	13.6
Q2_K	11.24 GiB	12.1	12.1	12.2	12.4	12.8	13.6
I1-Q3_K_S	11.23 GiB	12.1	12.1	12.2	12.4	12.8	13.6
Q3_K_S	11.23 GiB	12.1	12.1	12.2	12.4	12.8	13.6
I1-IQ2_XS	11.20 GiB	12.0	12.1	12.2	12.4	12.8	13.6
I1-IQ1_M	11.19 GiB	12.0	12.1	12.2	12.4	12.8	13.6
I1-IQ1_S	11.19 GiB	12.0	12.1	12.2	12.4	12.8	13.6
I1-IQ2_XXS	11.19 GiB	12.0	12.1	12.2	12.4	12.8	13.6

●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 128-token window rather than the full context.