Nvidia says new GPUs are the fastest for DeepSeek AI, which kind of misses the point

Feb 01, 2025 03:22 AM - 1 week ago 10034

Jay Peters

Jay Peters is a news editor covering technology, gaming, and more. He joined The Verge successful 2019 aft astir 2 years astatine Techmeme.

Nvidia is touting the capacity of DeepSeek’s unfastened root AI models connected its just-launched RTX 50-series GPUs, claiming that they tin “run the DeepSeek family of distilled models faster than thing connected the PC market.” But this announcement from Nvidia mightiness beryllium somewhat missing the point.

This week, Nvidia’s marketplace headdress suffered the azygous biggest one-day marketplace headdress nonaccomplishment for a US institution ever, a nonaccomplishment wide attributed to DeepSeek. DeepSeek said that its caller R1 reasoning exemplary didn’t require powerful Nvidia hardware to execute comparable capacity to OpenAI’s o1 model, letting the Chinese institution train it astatine a importantly little cost. What DeepSeek accomplished pinch R1 appears to show that Nvidia’s champion chips whitethorn not beryllium strictly needed to make strides successful AI, which could impact the company’s fortunes successful the future.

That said, DeepSeek did train its models utilizing Nvidia GPUs, simply weaker ones (H800) that the US authorities allows Nvidia to export to China. And today’s blog station from Nvidia wants to show that its caller 50-series RTX GPUs tin beryllium useful for R1 inference – aliases what an AI exemplary really generates – saying that the GPUs are built connected the “same NVIDIA Blackwell GPU architecture that fuels world-leading AI invention successful the information center” and that “RTX afloat accelerates DeepSeek, offering maximum conclusion capacity connected PCs.”

But really DeepSeek did its training is portion of what has been specified a large deal. (And it’s worthy noting that China is getting a less powerful version of the RTX 5090.)

Other tech companies are trying to thrust the DeepSeek wave, too. R1 is besides now disposable connected AWS, and Microsoft made it disposable connected its Azure AI Foundry level and GitHub this week. However, Microsoft and OpenAI are reportedly investigating if DeepSeek took OpenAI data, Bloomberg reports.

More