- Huawei Introduces Atlas 350 With Significant FP4 Computing Performance Claims
- New accelerator board focuses on inference workloads and multimodal AI processing
- Huawei Atlas 350 offers higher memory capacity and improved bandwidth efficiency
Huawei officially launched the Atlas 350 accelerator card, featuring its new Ascend 950PR processor, at the Huawei China Partner Conference 2026 in Shenzhen.
The company claims that this NPU delivers 1.56 PFLOPS of FP4 computing performance, which would be 2.87 times better than Nvidia’s H20.
Although exact verification is difficult because Hopper-era GPUs do not support FP4 natively, the Atlas 350 is China’s first accelerator optimized for this low-precision format, allowing larger AI models to run on the same hardware with reduced memory requirements.
Article continues below
Technical upgrades and memory performance
The Ascend 950PR chip introduces improvements over the previous Ascend 910 series, including improved microarchitecture, faster memory access and flexible programming modes.
Huawei equips the proprietary HBM 112GB Atlas 350, known as HiBL 1.0, offering up to 1.4TB/s of bandwidth according to current reports, with a memory access granularity of 128 bytes.
This configuration enables efficient multimodal generation and inference tasks and would quadruple the memory access efficiency for small operators compared to the previous generation.
Its interconnection bandwidth also reaches 2TB/s using the LingQu protocol, which is 2.5 times higher than that of the Ascend 910 series.
Huawei markets the Atlas 350 for recommendation inference, LLM processing, and multimodal AI workloads.
Seven key partners, including Kunlun, Huakun Zhenyu, Shenzhou Kuntai and Yangtze Computing, have developed complete system products leveraging the Atlas 350.
These brands have created high-performance custom inference solutions for enterprise customers.
The accelerator is designed to integrate with AI ecosystems, allowing partners to optimize performance for specific workloads while maintaining compatibility with Huawei’s AI software stack.
The Atlas 350 reflects China’s efforts to establish self-reliance in AI hardware under U.S. export restrictions.
Although Huawei cannot access TSMC’s CoWoS technology, the company has implemented advanced packaging alternatives for HBM and memory stacking.
Huawei did not announce specific availability dates – a common practice with AI accelerators – but launched the Ascend 950PR in the first quarter of 2026 as promised.
The Atlas 350 would cost around 111,000 yuan, or about $16,000, comparable to the Nvidia H20, which can range from $15,000 to $25,000.
Via Tom’s material
Follow TechRadar on Google News And add us as your favorite source to get our news, reviews and expert opinions in your feeds. Make sure to click the Follow button!
And of course you can too follow TechRadar on TikTok for news, reviews, unboxings in video form and receive regular updates from us on WhatsApp Also.
