华为即将推出的人工智能芯片可能挑战Nvidia的主导地位

币界网Publicado em 2024-08-13Última atualização em 2024-08-14

币界网报道:

中国电子巨头华为技术有限公司准备推出一款新的人工智能芯片,可以与英伟达强大的H100处理器相媲美。据《华尔街日报》报道,该公司一直在与中国科技行业的主要参与者,包括字节跳动、百度和中国移动,测试其Ascend 910C。

此前,美国实施了一系列严厉制裁,阻碍了这家亚洲巨头获得先进的芯片制造技术。

虽然英伟达目前的顶级芯片是Nvidia H200 Tensor Core GPU,但英伟达的H100无疑是世界上部署最广泛的人工智能芯片,销量达数百万,等待名单很长。这家老牌巨头也是竞争对手芯片制造商AMD的目标。

Ascend 910C代表了其前身910B的飞跃,910B与Nvidia的旧A100芯片大致相当,后者比H系列低。如果它能提供大幅升级的性能,华为可能即将在科技行业取得重大突破,巩固其在五年前首席执行官表示面临“生死攸关”时刻后的重生。

华为的潜在客户已经在排队下订单。《华尔街日报》报道称,该公司计划出售7万枚芯片,总价值接近20亿美元。据报道,该公司正准备在10月发货其硬件。

然而,一些分析人士质疑华为满足预期的能力,认为美国的制裁仍可能阻碍其进展。

特朗普政府实施的这些制裁对人工智能芯片市场来说是一把双刃剑。虽然他们限制了英伟达向中国客户销售顶级芯片的能力,但他们也刺激了国内创新。英伟达的回应——为中国市场量身定制的H20芯片——已经面临着来自本土替代品的激烈竞争,其中包括华为更便宜的Ascend 910B。

彭博社上个月报道称,作为回应,华为正在自己动手,在中国土地上建设开发和生产半导体设备的设施。据说该公司还与中国半导体公司合作,在国内生产高带宽存储芯片。

就Nvidia而言,它并没有停滞不前。该公司已经在开发基于Blackwell架构的下一代制裁友好型B20芯片。英特尔也在推出专为中国市场设计的高迪3处理器。

目前,华为对Ascend 910C的确切规格仍然守口如瓶。华为确实报告称,其前身在人工智能训练任务上的效率比英伟达的A100高出80%,在特定推理任务上的性能比其高出20%。

华为没有立即回应Decrypt的置评请求。

Leituras Relacionadas

Chip Stocks Lead U.S. Market Decline: Is AI Trading Being Hit by Both Interest Rates and Returns?

Chip stocks led a broad decline in US markets, with the Nasdaq dropping 2.2% and the S&P 500 falling 1.4%. This selloff reflects a dual challenge for the once-high-flying AI hardware trade: rising interest rate expectations and growing investor impatience for clear returns from massive AI capital expenditures. The pressure was most acute on hardware leaders. Nvidia fell about 4%, dipping below a $5 trillion market cap, while Micron plunged 13.2% ahead of its earnings report. Declines across memory, storage, AI, and mobile chips indicated a sector-wide retreat. The selloff spread globally, with South Korea's KOSPI index dropping nearly 10% as key suppliers SK Hynix and Samsung recorded double-digit losses. Investors appeared to be taking profits from the most crowded trades first. Macro headwinds intensified as market expectations shifted toward a more aggressive Federal Reserve. Forecasts for multiple rate hikes in 2026 pressured high-valuation tech stocks, which rely on long-term growth projections that become less attractive as discount rates rise. Concurrently, investors are scrutinizing the profit potential of the immense AI spending by cloud giants like Alphabet, Amazon, and Meta. While these expenditures drive demand for chips and hardware, the market is now questioning whether AI services will generate sufficient returns to justify the ongoing costs. This adjustment is not necessarily a bubble burst but a recalibration. AI demand fundamentals remain, but the narrative of endless growth can no longer fully offset concerns over higher interest rates and a longer path to profitability. Near-term direction may hinge on Micron's upcoming earnings guidance and incoming inflation data, which will influence both the AI demand outlook and the Fed's policy path. The market is transitioning from blindly buying growth to demanding clearer visibility on returns.

marsbitHá 27m

Chip Stocks Lead U.S. Market Decline: Is AI Trading Being Hit by Both Interest Rates and Returns?

marsbitHá 27m

OpenAI's New Paper: How to Train an AI that "Doesn't Deteriorate Under Pressure"?

OpenAI's new paper "Reinforcement Learning Towards Broadly and Persistently Beneficial Models" explores training AI to maintain safe, helpful, and honest behavior even under pressure, in unseen scenarios, or after being fine-tuned for harmful purposes. Moving beyond simple rule-based "don'ts," the research focuses on cultivating "beneficial traits" like honesty, risk-awareness, corrigibility, and transparency. It investigates if reinforcement learning (RL), often prone to "reward hacking" where models exploit loopholes, can instead be used to instill robust, generalized positive behaviors. Researchers created a multi-domain synthetic dialogue dataset covering areas like healthcare and law. They trained a model by replacing 5% of standard RL data with "beneficial trait" data. This model outperformed the baseline in 83% of 53 evaluations, showing average gains of 9.1% in alignment, safety, and helpfulness. Crucially, improvements generalized: a model trained only on healthcare "good behavior" data also performed better in 17 out of 19 non-healthcare alignment tests. The paper also tests "alignment persistence." When subjected to adversarial prompts or harmful fine-tuning, the beneficial trait model showed greater resilience, with smaller performance drops and less "spillover" of bad behavior to unrelated tasks. While not a complete solution, this work suggests a shift from post-hoc correction to proactively shaping robust, principled AI behavior, a critical step for deploying models in high-stakes, complex decision-making scenarios.

marsbitHá 30m

OpenAI's New Paper: How to Train an AI that "Doesn't Deteriorate Under Pressure"?

marsbitHá 30m

Semiconductor Stock Rebound: Is the Technical Correction Over or a Trend Reversal?

The core of recent semiconductor stock volatility is not about daily price swings, but rather the market questioning whether AI-driven semiconductor pricing has entered a new phase. Following a sharp sell-off in Korean stocks on June 23rd, led by Samsung and SK Hynix, a subsequent rebound is seen more as a technical positioning adjustment rather than a confirmed trend reversal. The key variable is HBM (High Bandwidth Memory), essential for AI chips. Its supply-demand imbalance granted memory makers significant pricing power. The current market focus is on whether this dynamic remains strong enough to justify elevated valuations. All eyes are on Micron's upcoming earnings report. The critical factor is not whether results meet already high expectations, but whether the company's guidance confirms that AI memory pricing power, order visibility, and future margins are still expanding. Micron's outlook will serve as a crucial test for the broader AI semiconductor chain, including Samsung, SK Hynix, and other infrastructure players. The recent bounce appears to be a pre-earnings positioning repair. For it to evolve into a sustained uptrend, concrete evidence is needed that the AI infrastructure expansion cycle's fundamentals—particularly for high-end memory—remain robust and can continue to surpass elevated market expectations. The risk is that strong demand alone may not be sufficient if future guidance hints at peaking momentum or increasing supply-side pressures.

marsbitHá 1h

Semiconductor Stock Rebound: Is the Technical Correction Over or a Trend Reversal?

marsbitHá 1h

Trading

Spot
Futuros
活动图片