How much VRAM does functionary-medium-v3.2 need?

Its published quantizations range from 15.60 GiB to 69.83 GiB in weights alone, before KV cache and runtime overhead.

functionary-medium-v3.2 on GeForce RTX 3080

Can I run functionary-medium-v3.2 on a GeForce RTX 3080?

Q: Can I run functionary-medium-v3.2 on a GeForce RTX 3080?

No. No published quantization of functionary-medium-v3.2 fits a GeForce RTX 3080 at any context with f16 KV cache.

Not at these settings. No indexed quantization of functionary-medium-v3.2 fits GeForce RTX 3080 at any context we compute, with f16 KV. The smallest shipped quantization is 15.60 GiB in weights alone, against 9.30 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— 10 GB card, 9.30 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	69.83 GiB	72.0	73.3	75.8	80.8	90.8	110.8
Q6_K	53.91 GiB	56.1	57.3	59.8	64.8	74.8	94.8
Q5_K_M	46.52 GiB	48.7	49.9	52.4	57.4	67.4	87.4
Q4_K_L	40.33 GiB	42.5	43.8	46.3	51.3	61.3	81.3
Q4_K_M	39.60 GiB	41.8	43.0	45.5	50.5	60.5	80.5
Q4_K_S	37.58 GiB	39.8	41.0	43.5	48.5	58.5	78.5
IQ4_XS	35.30 GiB	37.5	38.7	41.2	46.2	56.2	76.2
Q3_K_L	34.59 GiB	36.8	38.0	40.5	45.5	55.5	75.5
Q3_K_M	31.91 GiB	34.1	35.3	37.8	42.8	52.8	72.8
IQ3_M	29.74 GiB	31.9	33.2	35.7	40.7	50.7	70.7
Q3_K_S	28.79 GiB	31.0	32.2	34.7	39.7	49.7	69.7
IQ3_XXS	25.58 GiB	27.8	29.0	31.5	36.5	46.5	66.5
Q2_K_L	25.52 GiB	27.7	28.9	31.4	36.4	46.4	66.4
Q2_K	24.56 GiB	26.7	28.0	30.5	35.5	45.5	65.5
IQ2_M	22.46 GiB	24.6	25.9	28.4	33.4	43.4	63.4
IQ2_XS	19.69 GiB	21.9	23.1	25.6	30.6	40.6	60.6
IQ2_XXS	17.79 GiB	20.0	21.2	23.7	28.7	38.7	58.7
IQ1_M	15.60 GiB	17.8	19.0	21.5	26.5	36.5	56.5

●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 8,192-token window rather than the full context.