QDNASales and integration of LLM inference and training platforms, on-premises or hybrid

How much memory does Kimi K3 need in FP16?

Kimi K3 has 2800 billion parameters. The FP16 format decides how much memory is needed to load them.

Short answer. In FP16, the weights of Kimi K3 take about 6,440 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 FP16 format takes 2.0 byte per parameter. The product gives the weights, to which a 15% runtime margin is added for activations and buffers. The attention cache is not included: it grows with context and concurrency.

What does the format change?

FP16 brings full precision, the quality reference.

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

Which platforms qualify?

See the Kimi K3 fact sheet.

How much memory for Kimi K3 in FP16?

About 6,440 GB for the weights, excluding the attention cache.

Which format should I choose for Kimi K3?

FP16 brings full precision, the quality reference. The most precise format that fits the target machine remains the best choice.

Method

Parameter counts and memory figures come from the site fact sheets. Memory footprints are calculated, not measured: a reading on real hardware may differ depending on the engine and the exact weight format. Prices are indicative and not contractual.