Chinese AI team wins global award for replacing Nvidia GPU with industrial chip
researchers have trained a cutting-edge video-generation model on an off-the shelf industrial chip – outperforming high-end GPUs in both speed and efficiency.
Advertisement
Their system, FlightVGM, recorded a 30 per cent performance boost and had an energy efficiency that was 4½ times greater than
Nvidia's flagship RTX 3090 GPU – all while running on the widely available V80 FPGA chip from
Advanced Micro Devices (AMD), another leading US semiconductor firm.
The innovation earned top honours at the prestigious FPGA 2025 conference which concluded on March 1. The win marked the first time a mainland Chinese team had claimed the event's Best Paper Award, signalling a seismic shift in the global race to optimise
AI hardware
Developed by scientists from Shanghai Jiao Tong University, Tsinghua University and Beijing-based start-up Infinigence-AI, the model could redefine how industries deploy cost-effective, energy-efficient AI systems, from robotic controls to
autonomous vehicles
FPGAs or field-programmable gate arrays, are programmable semiconductor devices that allow post-manufacturing modifications to their circuitry and functionality. In contrast, conventional
chips like CPUs or central processing units, GPUs or graphics processing units, and ASICs or application-specific integrated circuits have fixed functionalities once fabricated.
Advertisement
In video-generation and general computing, FPGAs and Nvidia GPUs each have distinct advantages. FPGAs offer customisable architecture tailored to specific applications, resulting in higher energy efficiency and lower latency. Meanwhile, Nvidia GPUs, known for their massive parallel computing power, excel in processing large-scale data and handling complex computational tasks.
Building on previous research, the Chinese team developed FlightVGM, the first FPGA-trained video-generation AI model. Through innovations in data architecture and scheduling methods, FlightVGM achieved computational performance that could outpace GPUs.
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
South China Morning Post
6 hours ago
- South China Morning Post
New committee, bailout talk: is China launching a fresh property rescue effort?
China's central bank confirmed on Friday that it had established a new financial stability committee to defuse the nation's mounting debt risks, as a senior government adviser suggested that Beijing needed to inject 1 trillion yuan (US$139 billion) into the property sector to stabilise developers' balance sheets. The suggestion by Yin Zhongli, a counsellor for China's State Council, comes as Beijing continues to wrestle with a four-year property crisis sparked by developer Evergrande Group's default in mid-2021. Yin, also a senior real estate finance expert at the Chinese Academy of Social Sciences, warned that developers were still weighed down by high debts despite a recent uptick in home sales in the property market. China's central bank announced in its midyear report on Friday that it had set up a new macroprudential and financial stability committee as part of an ongoing derisking effort. More robust action was needed to help developers liquidate their assets, Yin suggested. 'Given the large capital shortfall facing these companies, I suggest … setting up a dedicated state-backed agency to provide capital injections to property developers that meet certain criteria,' he wrote in the July issue of the Tsinghua Financial Review. The Ministry of Finance should fund the injections by issuing special treasury bonds, with a proposed 1 trillion yuan 'real estate stabilisation trust' set up to channel the funds to eligible developers, Yin said. The funds should be channelled to some or all of the country's 30 largest private property firms, with the government taking convertible preferred shares from the developers with a conversion period of five years or more, he added.
AllAfrica
7 hours ago
- AllAfrica
Many silver linings in China's export slowdown
China's export machine—the core driver of its economic power for decades—is starting to falter. New data signals that overseas orders are weakening, raising alarm bells for those who have long viewed China as a consistent source of global momentum. A slowdown in outbound shipments suggests the country's extraordinary trade surpluses may be approaching their ceiling. Without new policy moves, the risk is that this key engine of global growth could begin to idle—just as the rest of the world is looking for new forward motion. Recent purchasing manager data reveals a clear drop in new export orders, with both official and private surveys falling short of expectations. This is significant not only for China but for the world. When China cools, the effects ripple outward—impacting commodity exporters, multinationals, emerging markets and global equity indices. Yet while the signals merit close attention, they do not warrant panic. In fact, for investors, it's a moment that demands precision, not pessimism. The July trade data due Thursday (August 7) is expected to show a 5% year-on-year rise in exports and a staggering $103.4 billion monthly surplus. But as with June's even higher surplus of US$114.8 billion, the surface numbers are misleading. Exports to the United States fell nearly 11% that month, underscoring growing trade friction and weakening demand in key markets. To remain competitive, Chinese firms have been slashing prices. Factory gate prices have been falling since late 2022, making the country's manufactured goods extraordinarily cheap. This has been a boon for overseas buyers—and a temporary cushion for exporters—but it's also a warning. Sustained deflation isn't a strength; it reflects weak domestic demand, overcapacity and thinning margins. Even so, it's critical to recognize that China's manufacturing and export sector remains vast, technologically advanced and globally integrated. The export slowdown we're now seeing is not a collapse—it's a deceleration from breakneck speed. And that shift opens a window for global investors to reposition for what comes next. As Chinese policymakers weigh their next move, the opportunity lies in anticipating where capital will flow in response. So far, Chinese government stimulus has been modest and targeted, but the pressure is growing. A more proactive policy response—be it through interest rate moves, consumption incentives or infrastructure support—could reinvigorate demand and lift sentiment. Investors with exposure to sectors sensitive to domestic recovery—such as materials, construction or consumer tech—should watch this space closely. Beyond China, this transition presents openings elsewhere. Countries across Southeast Asia, South Asia and Latin America are already capturing global interest as alternative manufacturing bases and consumption markets. This isn't about decoupling from China—it's about diversifying exposure. The relative winners will be those economies that combine political stability, reform momentum and access to global capital. China's export slowdown also forces a reassessment of risk and resilience. Currency dynamics, for instance, are shifting. The renminbi remains under pressure due to weaker export receipts and persistent capital outflows. While this weighs on investor returns in China in the short term, it improves competitiveness and may eventually support a rebound in export momentum—especially if global demand firms in late 2025. Meanwhile, global equity markets are rebalancing. China will no longer be the automatic offset to Western economic weakness it once was, and investors must rethink their global macro assumptions accordingly. But that also frees up capital for other stories—ones driven by services, innovation, energy transition and demographic strength. These themes are gaining traction in portfolios as the old China-dominated playbook fades. None of this implies abandoning China. The long-term fundamentals remain intact: vast human capital, global manufacturing scale, dominance in sectors like clean energy and EVs, and growing outbound investment. But exposure should be thoughtful—focused on quality, valuations and alignment with new policy directions. Yes, the warning lights are flashing for China's export momentum. The numbers are softening, sentiment is muted and global demand is uneven. But for those prepared to read beyond the headlines, this is not a time for retreat. It's a time to assess, realign and get ahead of the rotation that's already underway in global capital flows. Every slowdown is also a moment of recalibration. And from recalibration comes opportunity—for those willing to shift focus, reassess risk and embrace a broader global view. China's trade engine may no longer be running at full throttle, but it's far from out of fuel. And around it, new engines are revving up.
AllAfrica
8 hours ago
- AllAfrica
China's drone swarms just got smarter, faster and harder to kill
This month, the South China Morning Post (SCMP) reported that Chinese aerospace engineers have proposed a breakthrough upgrade for People's Liberation Army (PLA) drones, drawing lessons from the Russia-Ukraine war, where approximately 90% of Russian unmanned aerial vehicles (UAVs) were reportedly neutralized by advanced air defenses. Published in the July issue of Acta Armamentarii, the research—led by Bi Wenhao of Northwestern Polytechnical University—introduces a 'terminal evasion' system that equips small to medium-sized drones with compact, side-mounted rocket boosters. These boosters enable drones to execute abrupt, high-G maneuvers in the final seconds before missile impact, disrupting tracking algorithms and causing incoming warheads to detonate in empty space. Digital simulations cited in the study show survivability rates exceeding 87%, with evasive shifts occurring within a one-to-two-second window and lateral thrust reportedly surpassing 16Gs—far beyond conventional aerodynamic limits. While the concept demonstrates significant promise in controlled environments, real-world implementation poses notable challenges. Synchronizing booster activation with flight controls demands precise thrust vectoring, and the integration of additional components introduces trade-offs in payload, fuel capacity, and overall flight stability. The system's performance against faster or more adaptive interceptors remains untested, and operational deployment would require robust surveillance networks and onboard computing capable of autonomous, split-second decision-making. Still, analysts say the innovation could shift the logic of swarm warfare. Upgrading individual drones from cheap throwaways into survivable strike platforms could allow the PLA to turn vulnerable swarms into formations capable of saturating sophisticated air defenses. The proposal underscores China's growing emphasis on survivability and agility in unmanned systems as battlefield threats evolve. Tye Graham and Peter Singer note in a May 2025 Defense One article that PLA drone countermeasures in exercises already achieved a 40% kill rate, leaving a modest 60% survival rate for drones. The 'terminal evasion' concept aims to increase that rate by 27%, reflecting Beijing's belief that raising the survivability of individual drones enhances a swarm's ability to overwhelm enemy sensors and interceptors. The logic is backed by battlefield precedent. Jorge Rivero writes in a March 2025 article for the Modern Warfare Institute (MWI) that drone strikes account for 70% of Russian casualties and, in some areas of the front, 90% of Russian equipment losses in engagements with Ukrainian forces. Such figures confirm the tactical potency of drones even with high attrition rates, suggesting that improving their survivability could be a force multiplier. Yet China's military drone program faces structural weaknesses that may limit the impact of these innovations. John S Van Oudenaren points out in a May 2025 China Aeronautical Studies Institute (CASI) report that the PLA heavily relies on civilian drone manufacturers like DJI, which lack the military-grade standards and secure supply chains needed for wartime resilience. He further highlights fragmentation among suppliers, poor coordination between commercial and defense sectors, and limited operational testing as persistent barriers to effective integration. According to Oudenaren, China's drone industry also struggles with vulnerabilities to electronic warfare, insecure communications, and underdeveloped autonomy. Without centralized oversight or dedicated defense infrastructure, he says, these issues constrain the PLA's ability to militarize its commercial drone base fully. While Beijing's push for swarm resilience marks a doctrinal shift, these underlying gaps could blunt its operational effectiveness. By contrast, the US approach to drone warfare has historically prioritized quality over quantity. SCMP noted in December 2024 that the US integrates high-end drones like the Predator and autonomous wingmen into its network-centric strategy, emphasizing sensor fusion and interoperability. SCMP says that China, by contrast, leans on mass production and AI-enabled autonomy to achieve asymmetric saturation. Both sides now see drone survivability and scale not as technical details, but as the future of power projection. That dichotomy may be narrowing. In August 2023, the US Department of Defense (DoD) unveiled its Replicator Initiative to counter China's numerical advantage with a push to deploy thousands of attritable, autonomous, all-domain systems—including aerial drones and undersea vehicles—by 2025. Spearheaded by the Defense Innovation Unit (DIU), Replicator's first phase focuses on fielding low-cost, mass-producible platforms designed to survive contested environments and complicate enemy targeting through swarming tactics and distributed autonomy. Replicator reflects a broader shift toward disposability, scalability, and speed. Unlike legacy systems that are expensive and slow to produce, Replicator platforms are meant to operate with minimal human oversight and accept battlefield losses as a cost of saturation. Underscoring the shift, US Secretary of Defense Pete Hegseth outlined a sweeping overhaul in a July 2025 memo to restore US drone dominance. Citing Ukraine as proof of drones' transformative lethality, the memo rescinds procurement restrictions and delegates acquisition authority directly to warfighters to accelerate deployment. Backed by US Executive Order 14307, the plan emphasizes domestic manufacturing, streamlined acquisition, and combat-realistic training, including force-on-force drone simulations. The DoD will approve hundreds of US-made systems for frontline units, integrate emergent technologies, and break through bureaucratic inertia. Hegseth warns that outdated policies must not constrain lethality amid rapid technological change. Despite these initiatives, the US still lags behind China in drone production scale, speed, and affordability. The New York Times reported in July 2025 that China's military drone base, built on civilian mass production, vastly outpaces US capabilities. DJI alone produces millions of low-cost drones, such as the Mavic—sold for $300 to $5,000—which has been widely weaponized in Ukraine. The report states that while US defense contractors excel at developing sophisticated platforms like the Predator and Reaper, such systems are increasingly ill-suited for modern battlefields where mass-produced, expendable drones dominate. Even with recent policy shifts and executive orders aiming to revitalize domestic production, the US lacks the industrial scale to match China's rapid, cost-efficient output. The US drone shortfall is not just about budgets—it's about industrial architecture. Martin Feldman and Gene Keselman delve deeper into these production gaps in a July 2025 War on the Rocks article. They argue that although the US has industrial strengths in plastics, motor machining, and automotive-grade camera modules, it lacks a comprehensive wartime-scale mobilization strategy. While they say it's feasible to build domestic motor lines and repurpose car camera supply chains, critical bottlenecks remain in battery and circuit board capacity. By contrast, Feldman and Keselman note that Chinese civilian factories can retool within a year to produce one billion weaponized drones annually, using under 1% of their existing assembly capacity and without significantly straining battery or circuit board output. They warn that without major shifts in US industrial policy and long-term procurement contracts, America risks strategic vulnerability in a drone war driven not just by AI and autonomy, but by sheer output. Without bold industrial reforms and sustained procurement, the US risks losing not just the drone war but the ability to fight it at scale.
