Industry researchers dispute DeepSeek’s unusually low training cost claims
GPU smuggling that’s how. Not sure they would include the smuggled GPU’s into the numbers.
In rare disclosure, DeepSeek claims R1 model training cost just $294K
Submitted 1 day ago by BrikoX@lemmy.zip to technology@lemmy.zip
https://www.techspot.com/news/109542-rare-disclosure-deepseek-claims-r1-model-training-cost.html
Comments
87Six@lemmy.zip 1 day ago
87Six@lemmy.zip 1 day ago
Yup. No H200 and A100 and such mentioned, only H800 which were only recently banned in 2023. That’s why the cost was low.
danglybits27@sh.itjust.works 1 day ago
More clarification in this article:
www.theregister.com/2025/…/deepseek_cost_train/
“But, that’s not actually what happened. Never mind the fact that $300,000 won’t buy you anywhere close to 512 H800s (those estimates are based on GPU lease rates not actual hardware costs), the researchers aren’t talking about end-to-end model training.
Instead, it focuses on the application of reinforcement learning used to imbue its existing V3 base model with “reasoning” or “thinking” capabilities.
In other words, they’d already already done about 95 percent of the work by the time they’d reached the RL phase detailed in this paper.”