Connect with us

News

DeepSeek R1 is using Huawei Ascend AI chip: Report

Published

on

DeepSeek R1 Huawei

DeepSeek R1 – the latest Chinese AI model which uses the Huawei AI chip. The new artificial intelligence model has now gathered the center stage for its processors, amid the ongoing US-China chip war and trade export controls.

DeepSeek is a Chinese AI startup established in 2023. It builds open-source LLMs (large language models) and is solely founded by China’s hedge fund High-flyer.

R1 is DeepSeek’s latest smart AI model, now available for users. It is a general-purpose AI system. It is free and probably unlimited, unlike OpenAI and other top LLMs. The startup launched its first R1-based free chatbot app on January 10. This application has surpassed OpenAI ChatGPT as the most downloaded app in the US.

Unlike other LLMs, the DeepSeek R1 is extremely cost-effective. On the flip side, Gemini, Claude Sonnet, and ChatGPT remain limited for users under subscriptions. OpenAI’s o1 costs $15 per million input tokens, but DeepSeek is for $0.55.

Cost efficiency, unlimited usage, and open-source properties are the major signs of DeepSeek R1’s popularity. DeepSeek R1 is eventually showing that one doesn’t have to spend thousands of dollars to access AI models for unlimited use.

DeepSeek R1 chips

Yes, something even more interesting is that DeepSeek R1 is running on the Huawei Ascend 910C. The Chinese model indeed uses Nvidia processors for training. But its inference is based on the Ascend chipset. Let’s understand the difference.

In the AI world, training refers to the process of teaching an AI model about task completion. Under this procedure, the model is seeded with a set of training data. It learns the pattern and details of the data to make decisions accordingly.

DeepSeek R1 Huawei

DeepSeek R1 is using Huawei Ascend AI chip: Report (Image Credits: X)

But inference is the process of using the trained LLM to make predictions. In this procedure, the trained model is fed with new data and asked to make decisions without examples of the desired result. Simply put, it applies the learned patterns to generate content or make decisions on the given command.

DeepSeek R1 using Huawei AI chip?

Looks like Huawei is playing a big role in the popularity of DeepSeek R1. As per dorialexander, DeepSeek R1 uses Huawei Ascend 910C chips for inference. The model has been trained on the Nvidia H800 processor but runs inference on the Huawei Ascend 910C processor.

Ascend 910C is Huawei’s new AI chipset. The company unveiled the processor silently last year. It is a direct alternative to the Nvidia H100 and claims to defeat the Nvidia B20 to some extent. Hence, the chip is more powerful than its predecessors.

On the other hand, the Ascend 910C chip has a decent price compared to many Nvidia processors. Thus, Huawei AI processors seem the right choice for the DeepSeek R1 AI model. Other details of the new R1 model are still unknown but the mist will be clear soon.

I like to listen to music, sing, dance, and play outdoor games. I have a huge interest in reading novels and cooking. I'm good enough as a speaker. Besides, I have the willingness to learn new things and increase my knowledge in different aspects with full dedication and determination.