QDNAAdvisory and architecture for LLM inference and training platforms, on-premises or hybrid

How much memory does Kimi K3 need in FP8?

Kimi K3 has 2800 billion parameters. The FP8 format decides how much memory is needed to load them.

Short answer. In FP8, the weights of Kimi K3 take about 3,220 GB including the runtime margin. 1 of the 8 platforms in the catalogue have enough memory.

How is this footprint calculated?

Kimi K3 totals 2800 billion parameters. The FP8 format takes 1 byte per parameter (8 bits per parameter (E4M3), hence 1 byte, block scales not counted). The product gives the weights, to which a 15% runtime margin is added for activations and buffers: 2,800 billion parameters × 1 byte (FP8) = 2,800.0 GB of weights; × 1.15 runtime margin = 3,220.0 GB. The attention cache is not calculated on this page. These models’ architectures (compressed latent, linear or sparse attention, Mamba layers) share no per-token formula; it is measured on the machine, with the target context and concurrency [TO BE MEASURED].

What does the format change?

FP8 brings negligible loss on most tasks.

FormatWeights in memory Compatible platforms
FP166,440 GB1
FP83,220 GB1
NVFP41,610 GB2
Q41,932 GB2

Which platforms qualify?

See the Kimi K3 fact sheet.

How much memory for Kimi K3 in FP8?

About 3,220 GB for the weights, excluding the attention cache.

Which format should I choose for Kimi K3?

FP8 brings negligible loss on most tasks. The most precise format that fits the target machine remains the best choice.

Method and sources

Memory footprints are calculated, not measured: a reading on real hardware may differ depending on the engine and the exact weight format. Every figure below carries its source and the date of its last check.