New Huawei AI tech reportedly matches Nvidia in computing power

Huawei has recently introduced a new nuclear-grade CloudMatrix 384 AI Supernode that can reportedly match Nvidia in computing power. It is said to reduce the blockage in the computing segment for artificial intelligence data centers.

Before we understand the new Huawei product, let’s look at Nvidia’s AI tech.

Nvidia’s NVL72 debuted in March 2024. It contains a 72-graphics processing unit (GPU) NVLink domain that acts as a single component. It’s a high-speed interconnect technology that lets multiple GPUs interact and share data more efficiently.

NVL72 has a high-powered GPU that allows real-time inference for trillion-parameter LLMs at 30x faster speed than previous-generation AI technologies.

Although it looks like Huawei has created an alternative to this technology. The company has made a bold leap at a time when it is already struggling with sanctions.

STAR Market Daily reported that CloudMatrix 384 SuperNode is Huawei’s latest AI infrastructure architecture. The company unveiled this tech to challenge Nvidia’s dominance in AI and address a critical issue for data centers worldwide.

Huawei launched the nuclear-level AI CloudMatrix 384 Supernode tech last week, and in terms of computing power, it has matched Nvidia NVL72 AI system.

New Huawei AI tech reportedly matches Nvidia in computing power (Image Credits: Huawei/X)

CloudMatrix 384 supernode can achieve 300 petaflops of computing power at present. (Petaflop is a unit to measure computing speed and floating-point operations per second – FLOPS). It’s a big figure compared to NVL72, which offers 180 petaflops.

Huawei is currently testing and deploying the new tech in its Wuhu data centers. The company has used supernodes for its new product that has more resources like CPUs, NPUs, network bandwidth, storage, and memory over its alternatives.

These supernodes enable the AI tech to operate effectively and boost the computing performance of clusters. Thus, it increases the training of LLMs.

On the flip side, it will help meet the growing demand for computing power and efficiency for AI products and data centers. Inputs further reveal that Huawei is teaming up with SiliconFlow to use the new supernode in the DeepSeek R1.

[source]