The US-China AI war continues to heat up. When Trump chose OpenAI as a strategic partner, DeepSeek suddenly rose to prominence worldwide, with its core competitive advantage, the 'reasoning model,' still relying on NVIDIA's GPU power for acceleration, quickly debunking all news suggesting that NVIDIA GPUs were to be replaced.
NVIDIA's official blog reveals that the DeepSeek-R1 model allows development teams to run complex mathematical and coding tasks while maintaining customer privacy, with NVIDIA GeForce RTX 50 series GPUs accelerating DeepSeek reasoning models at a power of 3.352 trillion calculations per second.
A new type of reasoning model allows AI to think and provide feedback.
Reasoning Models are a new type of large language model where AI takes time to 'think' like humans and provides 'feedback' on mistakes, hence the name 'reasoning.' Just like humans solve problems, AI solves issues through more thinking, understanding, and feedback. Reasoning models can understand users' deep needs and allow AI to provide feedback on the thinking process of the model, further enhancing user experience and unlocking AI agent workflows to solve complex multi-step tasks like market analysis, complex mathematical problems, or coding, among others.
DeepSeek reasoning models can run quickly on NVIDIA's RTX AI PCs locally.
The DeepSeek-R1 series streamlined models are based on a large expert mixture (MoE) model with 671 billion parameters. The MoE model consists of multiple smaller Expert Models to solve complex problems. The DeepSeek model further allocates tasks and assigns sub-tasks to smaller groups of experts.
DeepSeek employs a technique called Distillation, similar to the process of making whiskey, which evaporates most of the alcohol through distillation, leaving the essence. The large DeepSeek model with 671 billion parameters can distill six smaller student models, with parameters ranging from 1.5 billion to 70 billion. The larger DeepSeek-R1 model imparts its reasoning capabilities to the smaller Llama and Qwen student models, resulting in powerful yet smaller reasoning models that run quickly on NVIDIA RTX AI PCs.
RTX equipped with DeepSeek-R1 reasoning model
NVIDIA GeForce RTX 50 GPUs are built on the fifth-generation Tensor Core, based on the same Blackwell GPU architecture. They accelerate DeepSeek reasoning models, providing the highest reasoning efficiency on AI PC computers.
NVIDIA's RTX AI platform offers a wide range of AI tools, software development kits, and model options. Currently, over 100 million NVIDIA RTX AI PCs (including PCs equipped with GeForce RTX 50 series GPUs) are capable of utilizing the features of DeepSeek-R1.
Killer GPUs can be used without an internet connection.
RTX GPUs can operate in low-latency environments even without internet connectivity, protecting customer privacy as users do not need to upload sensitive personal data and search histories to AI service systems.
This article NVIDIA Latest: GeForce RTX 50 Equipped with DeepSeek Accelerated Reasoning Models Running Locally on PC first appeared in Chain News ABMedia.
