Moore Threads says their GPUs have “excellent” performance by running DeepSeek

Chinese company Moore Threads decided to enter the Deepseek Hype train, promoting the performance of their own GPUS in the R1 AI model of their countryman. The video card manufacturer states that its products have “excellent” performance in Deepseek AI, but prefers not to inform the numbers.

The tests would have been specifically done with the Deepseek-R1-Distill-Qwen-7B model on the MTT S80 plates focused on household use and MTT S4000 for servers. The company used the Ollama Framework, allowing you to run the LLM directly from the GPUS, without the use of the cloud, creating an ideal situation for testing. Moore Threads also states that it created an optimized engineer engine on its signs.

According to Tom’s Hardware, Ollama has no native support for Moore Threads, but the company says its GPUs can run Cudda code, allowing to use the Framework optimization for the Nvidia platform.

The company highlights its owner engine as a major differential for the performance of its plates in DeepSeek AI. If performance information is true, inference can be a good use for the plates while Moore Threads is still running to offer more performance in games.

DeepSek Success benefits Chinese companies

The Chinese AI whale has been pulling the boat of many companies in the country as nothing in the wave of hype caused by its arrival. Giants of the country such as Alibaba and Tencent are beginning to offer deepseek -based AI solutions there, while hardware specializing promote the performance of their GPUS with technology.

While Moore Threads did not want to talk about Numbers, DeepSeek representatives claim that Huawei’s Ascend 910C accelerator can deliver 60% of an H100’s inference to its model.

Related News:

Now wait to see if Moore Threads will eventually offer more concrete data from its signs – and which other Chinese companies will also promote their services with DeepSek AI.

Via: Tom’s Hardware