$NVIDIA (NVDA.US)$1. Model Scale & Training Compute Needs
Aspect
DeepSeek
ChatGPT (OpenAI)
Core Model Size
~100B parameters (e.g., DeepSeek-MoE-16B)
~1.8T parameters (GPT-4)
Training GPU Demand
5,000–10,000 A100/H100 GPUs (per training cycle)
~25,000 A100 GPUs (GPT-4 training)
Training Cost
Tens of millions of USD
Over $100 million (GPT-4)
Iteration Speed
6–12 months (resource-constrained)
3–6 months (Microsoft Azure infrastructure)
Key Differences:
OpenAI’s models are 5–10x larger, requiring sig...
Aspect
DeepSeek
ChatGPT (OpenAI)
Core Model Size
~100B parameters (e.g., DeepSeek-MoE-16B)
~1.8T parameters (GPT-4)
Training GPU Demand
5,000–10,000 A100/H100 GPUs (per training cycle)
~25,000 A100 GPUs (GPT-4 training)
Training Cost
Tens of millions of USD
Over $100 million (GPT-4)
Iteration Speed
6–12 months (resource-constrained)
3–6 months (Microsoft Azure infrastructure)
Key Differences:
OpenAI’s models are 5–10x larger, requiring sig...
10
1
1
$NVIDIA (NVDA.US)$ money talks..rest of something else walks
1
$Tesla (TSLA.US)$ because it has been bought the news.
2
$Tesla (TSLA.US)$ overnight 283...sell the facts....
1
JoeChg
liked
2
1
$NVIDIA (NVDA.US)$ India has successfully landed on South of Moon thanks to Nvdia
3
2
$Tesla (TSLA.US)$ 5 b of tesla going to xai? musk is pocketing the money. obviously. xai is not doing well then. how about those chips borrowed from Tesla to Xai? selling tesla stock musk again...
1
$Psyence Biomedical (PBM.US)$ let's send it to the moon...
1
$VS MEDIA Holdings (VSME.US)$ let's take it to the moon