How much VRAM does Hypernova-60B-2605 need?

Its published quantizations range from 24.84 GiB to 58.13 GiB in weights alone, before KV cache and runtime overhead.

Can I run Hypernova-60B-2605 on a Apple M3 Pro?

Q: Can I run Hypernova-60B-2605 on a Apple M3 Pro?

No. No published quantization of Hypernova-60B-2605 fits a Apple M3 Pro at any context with q8_0 KV cache.

Not at these settings. No indexed quantization of Hypernova-60B-2605 fits Apple M3 Pro at any context we compute, with q8_0 KV. The smallest shipped quantization is 24.84 GiB in weights alone, against 12.56 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— 18 GB card, 12.56 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	58.13 GiB	58.8	58.8	59.0	59.2	59.7	60.8
I1-Q6_K	53.79 GiB	54.4	54.5	54.6	54.9	55.4	56.5
Q6_K	53.79 GiB	54.4	54.5	54.6	54.9	55.4	56.5
I1-Q5_K_M	41.29 GiB	41.9	42.0	42.1	42.4	42.9	44.0
Q5_K_M	41.29 GiB	41.9	42.0	42.1	42.4	42.9	44.0
I1-Q5_K_S	40.12 GiB	40.7	40.8	40.9	41.2	41.7	42.8
Q5_K_S	40.12 GiB	40.7	40.8	40.9	41.2	41.7	42.8
I1-Q4_K_M	37.89 GiB	38.5	38.6	38.7	39.0	39.5	40.6
Q4_K_M	37.89 GiB	38.5	38.6	38.7	39.0	39.5	40.6
I1-Q4_K_S	35.89 GiB	36.5	36.6	36.7	37.0	37.5	38.6
Q4_K_S	35.89 GiB	36.5	36.6	36.7	37.0	37.5	38.6
I1-Q4_1	34.48 GiB	35.1	35.2	35.3	35.6	36.1	37.2
I1-Q3_K_L	33.35 GiB	34.0	34.0	34.2	34.4	35.0	36.0
Q3_K_L	33.35 GiB	34.0	34.0	34.2	34.4	35.0	36.0
I1-Q3_K_M	31.25 GiB	31.9	31.9	32.1	32.3	32.9	33.9
Q3_K_M	31.25 GiB	31.9	31.9	32.1	32.3	32.9	33.9
I1-Q4_0	31.24 GiB	31.9	31.9	32.1	32.3	32.9	33.9
IQ4_XS	30.89 GiB	31.5	31.6	31.7	32.0	32.5	33.6
I1-IQ4_XS	30.55 GiB	31.2	31.2	31.4	31.6	32.2	33.2
I1-IQ3_M	29.06 GiB	29.7	29.7	29.9	30.1	30.7	31.7
I1-IQ3_S	28.73 GiB	29.3	29.4	29.5	29.8	30.3	31.4
I1-IQ3_XS	28.73 GiB	29.3	29.4	29.5	29.8	30.3	31.4
I1-Q2_K	28.73 GiB	29.3	29.4	29.5	29.8	30.3	31.4
Q2_K	28.73 GiB	29.3	29.4	29.5	29.8	30.3	31.4
I1-Q3_K_S	28.72 GiB	29.3	29.4	29.5	29.8	30.3	31.4
Q3_K_S	28.72 GiB	29.3	29.4	29.5	29.8	30.3	31.4
I1-IQ3_XXS	27.90 GiB	28.5	28.6	28.7	29.0	29.5	30.6
I1-Q2_K_S	27.42 GiB	28.0	28.1	28.2	28.5	29.0	30.1
I1-IQ2_M	27.04 GiB	27.7	27.7	27.9	28.1	28.7	29.7
I1-IQ2_S	26.56 GiB	27.2	27.2	27.4	27.6	28.2	29.2
I1-IQ2_XS	26.29 GiB	26.9	27.0	27.1	27.4	27.9	29.0
I1-IQ2_XXS	25.80 GiB	26.4	26.5	26.6	26.9	27.4	28.5
I1-IQ1_M	25.20 GiB	25.8	25.9	26.0	26.3	26.8	27.9
I1-IQ1_S	24.84 GiB	25.5	25.5	25.7	25.9	26.5	27.5

●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.