Home Comparisons A100 SXM4 vs L40S A100 or L40S: training against inference and video A100 is stronger for training and large models, thanks to its high-bandwidth HBM2e memory. L40S is a current card with 48 GB of GDDR6 that handles inference, video and graphics well, often at a lower hourly price.
Specifications A100 SXM4 80 GB L40S 48 GB Video memory 80 GB 48 GB Memory type HBM2e GDDR6 Architecture Ampere Ada Lovelace Price from 172₽ /h 119₽ /h Servers now 33 45
Prices and availability come from the live catalog and update automatically.
Which one to pick Pick A100 SXM4 if training and fine-tuning large models that need 80 GB multi-card setups on SXM Pick L40S if LLM inference up to 30B video generation and rendering a lower hourly price when 48 GB is enough Cost example A100 SXM4, per hour: about 172 ₽ L40S, per hour: about 119 ₽ Estimate at the current per-GPU hourly price. Billed per minute, the price is fixed at launch.
Questions and answers
Is the L40S good for training? For LoRA and smaller models, yes. For full training of large models, the A100 with more memory and bandwidth is usually easier to work with.
Which card is better for video? The L40S: it is built for graphics and video and often costs less per hour. For large video models, check memory size, not only the card type.