• 3 % discount on prepayment
  • Insured shipping from Switzerland
  • Sustainable: refurbished instead of new
  • Warranty on all cards
← Back to overview

Qwen3.5-397B-A17B

Alibaba (Qwen)
Apache 2.0

Open language model for chat and text. 8 sizes from 900M to 403.4B parameters, partly mixture-of-experts (only a part of the parameters is active per token).

  • Mixture-of-experts: 403.4B parameters in total, of which 17.0B active per token. Memory requirement depends on all parameters, speed on the active ones.

Settings

Quality
4 bit (Q4_K_M) is the most common choice: good quality with low memory requirements.
Context length: 8k tokens
4k256k
How much text the model keeps in view at once. More context needs more memory.
Recommendation
4× H100 80 GB

The model needs about 243.8 GB of memory. Hopper, FP8 support.

Weights
242.0 GB
KV cache
0.3 GB
Speed
~79.9 Tok./s

More suitable configurations

CardQuantityTok./sStatus
A40 48 GB8×29.42-3 weeks delivery time
RTX A6000 48 GB8×32.92-3 weeks delivery time
Quadro RTX 8000 48 GB8×24.32-3 weeks delivery time
L40 48 GB8×36.12-3 weeks delivery time
A100 40 GB8×36.42-3 weeks delivery time
A100 80 GB4×56.92-3 weeks delivery time
H100 80 GB4×79.9only 1 in stock