China has been consistently looking to crack into NVIDIA's CUDA, and one of the proposed workarounds is defintely worth highlighting.
China's Semiconductor Official Advises the Domestic AI Industry To Shift Towards Software-Defined Chips
When Jensen is asked about what one of the biggest reasons is behind NVIDIA's dominance in the AI industry, he always refers to CUDA as the "strongest moat", saying that the persistent work done in uplifting the software ecosystem is what distinguishes Team Green from others. And it appears that China is worried about its industry's reliance on CUDA, which is why Wei Shaojun, an executive at the China Semiconductor Industry Association, says Beijing should develop alternatives to CUDA and other components sourced from the West.
Even if our own technology is not good enough at the start, it must still be used. Trial and error may not succeed, but without trying, we will certainly fall behind.
- Wei Shaojun
Talking about CUDA in particular, Shaojun says China should probably stop focusing on creating a direct alternative to the software and instead adopt a strategy that hasn't been discussed much. He proposes the idea of a "software-defined chip", or SDC, where the focus is on moving the compute intelligence towards the software, instead of a pre-layout hardware configuration. Right now, developers are inclined towards CUDA due to the ecosystem's maturity, which indirectly binds them to NVIDIA's hardware. However, an SDC flips the game, and we'll talk about the 'how' next.
With an SDC, developers wouldn't need the 'CUDA' layer to perform their computations; instead, the chips are configured with a reconfigurable grid that uses the configuration bitstream generated by the compiler. In simpler terms, what this means is that neither the compiler nor the source-level representation of the code depends on an ISA; instead, it is much more flexible. Compared to GPUs, which operate through a dedicated scheduler, SDCs rely on deterministic compilation, meaning every data movement down to the clock cycle is tracked.
The overhead of creating translation layers and independent ecosystems that aim to replicate CUDA's success is too high, according to Professor Wei Shaojun, who suggests the SDC route is a viable bet for China. If you have realized it by now, SDC's reliance on the compiler is what makes this effort a 'nightmare', and it includes routing and branching issues along with structural changes that defy the norms of hardware engineering. Popular examples of SDCs include SambaNova's RDUs and Groq's LPU units, but they are designed to complement specific workloads rather than replace GPUs entirely.
News Source: DigiTimes
About the author: Muhammad Zuhair is a hardware and technology reporter for Wccftech, specializing in the semiconductor industry and the complex interplay between technology, manufacturing, and geopolitics. His coverage focuses on the corporate strategies and technological roadmaps of industry giants like TSMC, NVIDIA, Samsung, and Intel. Zuhair's expertise lies in deconstructing complex topics such as fabrication nodes (e.g., 2nm process), the economic impact of policies like the CHIPS Act, and the strategic development of AI infrastructure from NVIDIA, AMD and Intel.
Follow Wccftech on Google to get more of our news coverage in your feeds.
