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

Llama-3.3-70B-Instruct

Meta
GermanLlama 3.1 Community

Open language model for chat and text. 2 sizes from 8B to 70.6B parameters.

Settings

Quality
4 bit (Q4_K_M) is the most common choice: good quality with low memory requirements.
Context length: 8k tokens
4k128k
How much text the model keeps in view at once. More context needs more memory.
Recommendation
1× H100 80 GB

The model needs about 46.5 GB of memory. Hopper, FP8 support.

Weights
42.4 GB
KV cache
2.7 GB
Speed
~23.4 Tok./s

More suitable configurations

CardQuantityTok./sStatus
Tesla T4 16 GB4×3.32-3 weeks delivery time
Tesla P100 16 GB4×4.12-3 weeks delivery time
Tesla P40 24 GB4×3.92-3 weeks delivery time
Tesla V100 16 GB4×9.1only 1 in stock
Tesla V100 32 GB2×10.4On order, delivery time approx. 1 week
AMD Instinct MI50 16 GB4×7.0only 2 in stock
AMD Instinct MI50 32 GB2×8.0in stock
AMD Instinct MI60 32 GB2×7.42-3 weeks delivery time
AMD Instinct MI100 32 GB2×8.92-3 weeks delivery time
A30 24 GB4×8.82-3 weeks delivery time
RTX A5000 24 GB4×9.22-3 weeks delivery time
A40 48 GB2×10.02-3 weeks delivery time