Huawei claims new software can 'create an analogue AI chip 1000 times faster than Nvidia’s chips' — open source Flex:ai software designed to boost AI-chip utilization

News
By published

The platform promises to pool GPUs and NPUs and raise average utilization by about 30%.

Huawei
(Image credit: Getty / Bloomberg)

Huawei has introduced Flex:ai, an open-source orchestration tool designed to raise the utilization rate of AI chips in large-scale compute clusters. Announced on Friday, November 21, the platform builds on Kubernetes and will be released through Huawei’s ModelEngine developer community. It arrives amid continued U.S.export restrictions on high-end GPU hardware and reflects a growing shift inside China toward software-side efficiency gains as a stopgap for constrained silicon supply.

Aside from being equipped to help China “...create an analogue AI chip 1000 times faster than Nvidia’s chips,” Huawei claims Flex:ai can raise average utilization by around 30%. It reportedly does this by slicing individual GPU or NPU cards into multiple virtual compute instances and orchestrating workloads across heterogeneous hardware types.

Flex:ai’s architecture builds on existing open-source Kubernetes foundations but extends them in ways that are still uncommon across open deployments. Kubernetes already supports device plugins to expose accelerators and schedulers, such as Volcano, or frameworks like Ray can perform fractional allocation and gang scheduling. Flex:ai appears to unify them at a higher layer while integrating support for Ascend NPUs alongside standard GPU hardware.

The launch resembles functionality offered by Run:ai, an orchestration platform acquired by Nvidia in 2024, which enables multi-tenant scheduling and workload pre-emption across large GPU clusters. Huawei’s version, at least on paper, makes similar claims but does so with a focus on open-source deployment and cross-accelerator compatibility. That may give it broader relevance in clusters built around Chinese silicon, particularly those using Ascend chips.

Google Preferred Source

Follow Tom's Hardware on Google News, or add us as a preferred source, to get our latest news, analysis, & reviews in your feeds.

TOPICS
Luke James
Luke James
Contributor

Luke James is a freelance writer and journalist.  Although his background is in legal, he has a personal interest in all things tech, especially hardware and microelectronics, and anything regulatory. 

8 Comments Comment from the forums
  • redgarl
    And ironically, Nvidia and AMD doesn't have such tools even if AMD and OpenAI are collaborating for creating AI chips...

    right...
    Reply
  • psyconz
    The part about 1000 times faster is clearly hyperbole, in the sense that it's probably true, but only in any given non-specific time span, agreed.

    But if they could do the relatively simple work of actually releasing an open-source solution - easy compared to developing a full-blown 'security nightmare' closed source one - then the code will be easily read by anyone who wants to use it. Then the wider community would make it work.

    The part about utilising disparate parts of a machine in tandem to improve utilisation by 30% is not so much of a bold claim as "1000x faster than Nvidia." And it would be useful for anyone who cares to use it.

    I sure know if it ends up being a closed source, commercial product, it might not live that long in certain countries... so I'm betting the part about open source software through kubernetes is truth. And I welcome it. If not, it's just another potential AI development in this area, and we get a lot of them. It won't be long before other similar solutions are free and available. In fact they probably already are, on at least Linux.
    Reply
  • Giroro
    It's 1000x faster, but unfortunately unfeaseable as the design it occupies 2 square meters of silicon, consumes 300 Megawatts of power, and can only reach it's theoretically speeds when operating at -900 degrees Kelvin and under 1200 bars of atmospheric pressure.

    Or they're just lying, which Huawei does all the time
    Reply
  • psyconz
    Haha, yeah, Huawei's got this great new AI that's 1000x faster. Problem is, you need to launch yourself directly into the sun to use it 🤪
    Reply
  • HideOut
    Giroro said:
    It's 1000x faster, but unfortunately unfeaseable as the design it occupies 2 square meters of silicon, consumes 300 Megawatts of power, and can only reach it's theoretically speeds when operating at -900 degrees Kelvin and under 1200 bars of atmospheric pressure.

    Or they're just lying, which Huawei does all the time
    Its any chinese company it seems.
    Reply
  • oicu812
    Why are Chinese AI companies accumulating NVIDIA chips if their homegrown chips are so much better?
    Reply
  • JRStern
    >analogue
    Give it a break: analog

    This sounds garbled. Analog might be 1000x faster by eliminating digital multiply and sum, this be true, though it's going to require some clever fabrication to achieve. And then it has to be digitized back.

    Slicing things into service groups is an ancient approach (10 items or less!), but it's not 1000x anything.
    Reply
  • hax0red1
    HideOut said:
    Its any chinese company it seems.
    While this is likely some asic designed for specific tasks and only performs in that 1,000 X speed in certain scenarios, there are some Chinese companies actually doing amazing work on production products. Most of the DJI product line is second to none, even their FPV goggles and video transmitters are hands down the best with second best also coming from China. The same can be said with consumer 3D printers that Bambu Labs makes, and they are a fairly new entry into the 3D printing world. China was the leader in many innovations for thousands of years, only their retreat into isolation caused their progress to stagnate.
    Reply