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

Its published quantizations range from 10.73 GiB to 88.57 GiB in weights alone, before KV cache and runtime overhead.

gpt-oss-20b-heretic on Radeon RX 6500 XT

Can I run gpt-oss-20b-heretic on a Radeon RX 6500 XT?

Q: Can I run gpt-oss-20b-heretic on a Radeon RX 6500 XT?

No. No published quantization of gpt-oss-20b-heretic fits a Radeon RX 6500 XT at any context with q4_0 KV cache.

Not at these settings. No indexed quantization of gpt-oss-20b-heretic fits Radeon RX 6500 XT at any context we compute, with q4_0 KV. The smallest shipped quantization is 10.73 GiB in weights alone, against 3.72 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— 4 GB card, 3.72 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◐
Q5_1	88.57 GiB	89.5	89.5	89.6	89.7	89.9	90.3
IQ4_NL	67.58 GiB	68.5	68.5	68.6	68.7	68.9	69.3
BF16	38.99 GiB	39.9	39.9	40.0	40.1	40.3	40.7
Q8_0	20.73 GiB	21.7	21.7	21.7	21.8	22.0	22.5
Q6_K	20.67 GiB	21.6	21.6	21.7	21.8	22.0	22.4
Q6_K_L	20.67 GiB	21.6	21.6	21.7	21.8	22.0	22.4
Q5_K_L	15.91 GiB	16.8	16.9	16.9	17.0	17.2	17.6
Q5_K_M	15.74 GiB	16.7	16.7	16.7	16.8	17.1	17.5
Q4_K_L	14.97 GiB	15.9	15.9	16.0	16.1	16.3	16.7
Q5_K_S	14.81 GiB	15.7	15.8	15.8	15.9	16.1	16.5
Q4_K_M	14.76 GiB	15.7	15.7	15.8	15.9	16.1	16.5
Q4_K_S	13.83 GiB	14.8	14.8	14.8	14.9	15.1	15.6
Q3_K_L	12.42 GiB	13.3	13.4	13.4	13.5	13.7	14.2
Q3_K_M	12.03 GiB	12.9	13.0	13.0	13.1	13.3	13.8
IQ3_M	11.69 GiB	12.6	12.6	12.7	12.8	13.0	13.4
Q2_K_L	11.59 GiB	12.5	12.5	12.6	12.7	12.9	13.3
IQ4_XS	11.40 GiB	12.3	12.4	12.4	12.5	12.7	13.1
IQ3_XS	11.32 GiB	12.2	12.3	12.3	12.4	12.6	13.1
IQ3_XXS	11.32 GiB	12.2	12.3	12.3	12.4	12.6	13.1
Q2_K	11.32 GiB	12.2	12.3	12.3	12.4	12.6	13.1
Q3_K_S	11.32 GiB	12.2	12.3	12.3	12.4	12.6	13.1
IQ2_M	11.31 GiB	12.2	12.3	12.3	12.4	12.6	13.0
Q4_1	10.80 GiB	11.7	11.7	11.8	11.9	12.1	12.5
Q4_0	10.73 GiB	11.6	11.7	11.7	11.8	12.0	12.5

●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 128-token window rather than the full context.