How much VRAM does GLM-5 need?

Its published quantizations range from 164.05 GiB to 1404.42 GiB in weights alone, before KV cache and runtime overhead.

GLM-5 on Apple M5 Max

Can I run GLM-5 on a Apple M5 Max?

Q: Can I run GLM-5 on a Apple M5 Max?

No. No published quantization of GLM-5 fits a Apple M5 Max at any context with q4_0 KV cache.

Not at these settings. No indexed quantization of GLM-5 fits Apple M5 Max at any context we compute, with q4_0 KV. The smallest shipped quantization is 164.05 GiB in weights alone, against 25.11 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— 36 GB card, 25.11 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	1404.42 GiB	1405.1	1405.2	1405.4	1405.8	1406.6	1408.1
Q8_0	746.31 GiB	747.0	747.1	747.3	747.7	748.5	750.0
Q6_K	576.84 GiB	577.5	577.6	577.8	578.2	579.0	580.5
Q5_K_M	498.42 GiB	499.1	499.2	499.4	499.8	500.6	502.1
Q5_K_S	484.04 GiB	484.7	484.8	485.0	485.4	486.2	487.7
Q4_1	440.24 GiB	440.9	441.0	441.2	441.6	442.4	443.9
Q4_K_M	424.57 GiB	425.3	425.4	425.6	425.9	426.7	428.3
Q4_K_S	398.95 GiB	399.6	399.7	399.9	400.3	401.1	402.6
Q4_0	397.81 GiB	398.5	398.6	398.8	399.2	400.0	401.5
IQ4_NL	396.67 GiB	397.4	397.5	397.7	398.0	398.8	400.4
IQ4_XS	375.23 GiB	375.9	376.0	376.2	376.6	377.4	378.9
Q3_K_M	335.56 GiB	336.3	336.4	336.5	336.9	337.7	339.2
Q3_K_S	303.87 GiB	304.6	304.7	304.9	305.2	306.0	307.6
UD-IQ3_XXS	283.62 GiB	284.3	284.4	284.6	285.0	285.8	287.3
Q2_K_L	257.21 GiB	257.9	258.0	258.2	258.6	259.4	260.9
Q2_K	257.00 GiB	257.7	257.8	258.0	258.4	259.1	260.7
UD-IQ2_M	237.27 GiB	238.0	238.1	238.3	238.6	239.4	241.0
UD-IQ2_XXS	224.52 GiB	225.2	225.3	225.5	225.9	226.7	228.2
UD-IQ1_M	208.48 GiB	209.2	209.3	209.5	209.9	210.6	212.2
UD-IQ1_S	189.71 GiB	190.4	190.5	190.7	191.1	191.9	193.4
UD-TQ1_0	164.05 GiB	164.7	164.8	165.0	165.4	166.2	167.7

●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, this model uses latent attention and allocates no V cache at all, so any formula reading num_key_value_heads overstates its cache by more than an order of magnitude.