How much memory does Qwen 3.8 27B need in FP8?

Qwen 3.8 27B has 27 billion parameters. The FP8 format decides how much memory is needed to load them.

By QDNA · Published on 27 July 2026 · Updated on 15 September 2026

Short answer. In FP8, the weights of Qwen 3.8 27B take about 31.0 GB including the runtime margin. 8 of the 8 platforms in the catalogue have enough memory.

How is this footprint calculated?

Qwen 3.8 27B totals 27 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: 27 billion parameters × 1 byte (FP8) = 27.0 GB of weights; × 1.15 runtime margin = 31.0 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.

Format	Weights in memory	Compatible platforms
FP16	62.1 GB	8
FP8	31.0 GB	8
NVFP4	15.5 GB	8
Q4	18.6 GB	8

Which platforms qualify?

See the Qwen 3.8 27B fact sheet.

How much memory for Qwen 3.8 27B in FP8?

About 31.0 GB for the weights, excluding the attention cache.

Which format should I choose for Qwen 3.8 27B?

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.

Qwen 3.8 27B: 27 billion parameters, 262,144-token context, Apache 2.0 licence. Hugging Face repository (config.json, API, model card), checked on 2 September 2026. Dense model: every parameter computes on each token. The BF16 repository holds 27.8 billion elements (56 GB).
Bytes per parameter: FP16 2 (16 bits per parameter, hence 2 bytes); FP8 1 (8 bits per parameter (E4M3), hence 1 byte, block scales not counted); NVFP4 0.5 (4 bits per parameter, hence 0.5 byte; 4-bit published weights weigh 0.54 to 0.56 byte per parameter with their scales (DeepSeek V4 Pro 865 GB for 1,599 billion, Kimi K3 1,561 GB for 2,780 billion)); Q4 0.6 (taken here as llama.cpp Q4_K_M, 0.6 byte per parameter including scales; Q4_K_S weighs 0.56, AWQ and GPTQ 0.55). Source: connaissance/faits.yaml, quantifications family, checked on 2026-08-31.
Runtime margin × 1.15: QDNA operating assumption, not measured: 15% above the weights for activations, buffers and fragmentation.
“Fits” threshold at 75% of memory: QDNA assumption, not measured, which keeps the remaining quarter for the attention cache and concurrency.