推理速度达英伟达 GPU 的 10 倍,刚获 6.4 亿美元融资的 Groq 是什么来头?

深潮Published on 2024-08-08Last updated on 2024-08-08

以快著称的 AI 芯片独角兽 Groq,官宣融资 6.4 亿美元,最新估值也暴涨到 28 亿美元。

撰文:元宇宙之心

8 月 5 日,美国人工智能芯片初创公司 Groq 官宣获得 6.4 亿美元最新融资,公司估值也来到了 28 亿美元。

此轮融资由华尔街资管巨头贝莱德的 BlackRock Private Equity Partners 领投,思科的 Cisco Investments 和三星电子的 Samsung Catalyst Fund 也参与其中。Groq 背后的技术创新实力和财力支持可见一斑。

01.1 分钟项目速览

1.项目名称:Groq

2.成立时间:2016 年

3.产品简介

Groq 推出的新型 AI 加速芯片 LPU,专为大语言模型量身定制,具有高速推理性能。其芯片性能表现比常规的 GPU 和 TPU 提升 10 到 100 倍,推理速度达到了 NVIDIA GPU 的 10 倍。

4.创始人团队

  • Jonathan Ross:创始人兼 CEO,谷歌 TPU 项目核心研发人员

  • Yann LeCun:图灵奖得主、深度学习三巨头之一,担任技术顾问

5.融资情况

  • 2017 年,Groq 在成立初期获得了种子轮融资 1030 万美元;

  • 2018 年,Groq 在 A 轮融资中筹集了 5230 万美元,由 Social Capital 领投;

  • 2020 年,Groq 在 B 轮融资中筹集了 1.5 亿美元,由 Tiger Global Management 领投,参与者还包括 D1 Capital Partners 和 The Spruce House Partnership;

  • 2021 年,Groq 在 C 轮融资中筹集了 3 亿美元,由 Tiger Global Management 和 D1 Capital Partners 继续支持;

  • Groq 在最新一轮融资中筹集了 6.4 亿美元,由 BlackRock Inc.基金领投,思科、三星跟投。

02.创新驱动的 AI 处理器先锋

Groq 的创立过程可以被视为创新和技术突破的一个典范

Ross 在创办 Groq 之前,曾在谷歌担任工程师,负责推动深度学习和计算架构方面的研究。在谷歌工作期间,Ross 发现传统计算架构在处理现代 AI 任务时存在显著的性能瓶颈,特别是在深度学习和大规模数据分析方面。

传统的 CPU 和 GPU 无法满足这些任务对并行计算和低延迟的高要求。这种认知促使 Ross 离开谷歌,创办一家专注于突破传统计算限制的公司——Groq。

Groq 的创始团队在公司创立初期集中精力于硬件设计和技术研发。团队成员来自于高性能计算和半导体领域的顶尖人才,具备丰富的经验和技术背景。早期的工作包括处理器架构的设计、原型的开发以及性能测试。

Groq 的技术理念围绕着一种创新的处理器架构展开,旨在提供比现有处理器更高的计算性能和效率。公司的目标是设计一种能够突破 CPU 和 GPU 限制的硬件平台,以满足 AI 和 HPC 领域对计算能力的不断增长的需求。

随着技术的成熟和市场需求的增加,Groq 开始扩展其业务范围,进入了数据中心、云计算和边缘计算等领域。目前,该公司已与多个行业领先的技术合作伙伴建立了战略合作关系,推动其产品在全球范围内的应用。

03.重塑高性能计算

自创立以来,Groq 凭借其突破性的技术和卓越的产品,迅速成为人工智能(AI)和高性能计算(HPC)领域的先锋。

Groq 的技术核心在于其创新的处理器架构。与传统的 CPU 和 GPU 处理器不同,Groq 的设计理念着眼于提升计算能力和效率,特别是针对现代 AI 和数据密集型应用的需求

  • 高度并行化设计:Groq 的处理器架构采用了极高的并行化设计,集成了大量计算单元,能够同时处理大量的数据。这种设计不仅提升了计算能力,还降低了数据处理的延迟。

  • 简化数据路径:传统处理器常常面临数据传输瓶颈,Groq 的架构通过优化数据路径和高速缓存设计,显著减少了数据传输延迟。这样一来,处理器能够以更高的效率处理大规模数据集,满足 AI 训练和推理过程中的高性能要求。

  • 灵活的配置选项:Groq 提供了多种配置选项,允许用户根据具体应用需求调整计算资源。这种灵活性使得其处理器可以广泛应用于不同的计算场景,从数据中心到边缘计算环境。

Groq 的 AI 加速器是其产品阵容中的核心组成部分,专为加速深度学习模型的训练和推理而设计。其优势主要体现在以下三个方面:

  • 高吞吐量和低延迟:AI 加速器通过提高处理能力和降低延迟,显著加快了机器学习模型的训练速度。这对于需要处理海量数据的应用,如图像识别和自然语言处理,尤其重要。

  • 优化算法支持:Groq 的加速器优化了各种机器学习算法,包括卷积神经网络(CNN)、循环神经网络(RNN)等。通过硬件层面的优化,加速器能够高效执行这些复杂的计算任务,提高模型的训练效率和推理性能。

  • 扩展性和可配置性:加速器设计支持高扩展性,能够应对不同规模的数据处理任务。用户可以根据需要配置多个加速器,构建高性能计算集群,以满足大规模 AI 应用的需求。

凭借高度并行化的设计、优化的数据路径、以及强大的 AI 加速能力,Groq 为数据中心、云计算和边缘计算等多个领域提供了卓越的技术支持

随着技术的不断发展和产品的持续优化,Groq 正在为现代计算需求提供前所未有的解决方案,推动计算科技向更高水平迈进。

作为 AI 芯片领域黑马的 Groq,其 LPU 芯片的高速推理能力毋庸置疑。但随着公司估值的暴涨和市场的期待,Groq 也在面临着一系列的挑战与问题。

04.挑战与机遇并存

首先便是 LPU 的容量问题,虽然在处理大语言模型时展现出了卓越的性能,但内存较小导致在实际部署时可能需要大量的硬件资源。

据分析,LPU 在运行 LLaMA 70B 等大模型时,Groq 的硬件需求和成本可能远高于预期,这无疑增加了公司在大规模部署时的经济压力。

其次,LPU 的专用性虽然在特定任务上提供了优势,但也限制了其在更广泛 AI 任务中的应用。与 GPU 相比,LPU 的通用性不足可能影响其在多样化 AI 场景中的竞争力。

作为初生产品与初创公司,Groq 在技术成熟度、市场认可度以及生态系统建设方面还有很长的路要走。它需要不断优化产品,扩大研发团队,并与行业伙伴建立合作关系,以加速技术的商业化进程。

展望未来,Groq 的发展机遇同样巨大。随着 AI 技术的不断进步和应用场景的拓展,相关企业对高效能 AI 芯片的需求日益增长。Groq 若能有效解决成本和通用性问题,其 LPU 芯片有望在 AI 推理市场占据重要地位。

Groq 公司计划到 2025 年 3 月底前推出 10.8 万个 LPU,这一宏伟目标若能实现,将进一步巩固其在行业中的领导地位。Groq 如何在激烈的市场竞争中不断突破自我,实现技术创新与商业成功的双重飞跃,值得我们持续关注。

Trending Cryptos

Related Reads

Options Don't Work in DeFi? Vitalik Might Not Agree

For years, the prevailing view has been that options struggle to gain traction in DeFi due to complexity, fragmented liquidity, and lack of natural demand compared to products like perpetual futures. However, a recent algorithmic stablecoin design proposed by Vitalik Buterin presents a different perspective, using options not as a standalone trading product, but as foundational infrastructure for other financial instruments. In this design, one unit of ETH is split into two components: a "stable" side (P) that retains value up to a specified strike price, and an "upside" side (N) that captures all appreciation above that strike. Combined, they always equal one ETH, eliminating debt, margin, and liquidation risks inherent in typical collateralized debt position (CDP) stablecoins. The stable component essentially mimics the payoff of a covered call option. To function as a stablecoin, this structure requires continuously rolling deep in-the-money calls, which introduces challenges like rollover slippage, predictable transaction flow vulnerable to front-running, and persistent liquidity needs. A core hurdle is finding consistent buyers for the leveraged ETH upside exposure (N). While it offers leverage without funding rates or liquidation, it must compete with simpler alternatives like direct call options or perpetuals. The system's scalability depends on a sustained demand for this specific form of leverage. The author draws parallels to their experience with Rysk, where earlier versions of DeFi options protocols struggled. The breakthrough came with Rysk V12, which aligns incentives: asset holders generate yield by selling covered calls against their holdings, while market makers efficiently acquire the desired option exposure. This demonstrates that options can find product-market fit when embedded as a risk distribution and pricing engine within structured products, stablecoins, or yield-generating assets, rather than marketed as a complex direct trading instrument. Vitalik's proposal reinforces this architectural approach—using fully collateralized, non-custodial, and physically settled options as a fundamental building block. The real opportunity for options in DeFi may lie not in becoming the next perpetual swap, but in powering the next generation of on-chain financial products.

marsbit8m ago

Options Don't Work in DeFi? Vitalik Might Not Agree

marsbit8m ago

Conversation with Investor Zheng Di: MicroStrategy's Coin Sale Experiment, AI Economy, and Opportunities in US Stocks

Frontier tech investor Zheng "Didier" Di discusses the recent Bitcoin price drop, the financial strategy shift at MicroStrategy, the AI-driven surge in U.S. stocks, and the evolving role of crypto exchanges. Didier posits that the recent BTC decline stems less from macro factors or ETF outflows, and more from market repricing due to MicroStrategy's new financial structure. Following a wave of preferred stock and debt issuance (STRC, STRZ, etc.), MicroStrategy must now manage cash flow to pay dividends, potentially leading to a market expectation of sustained, small-scale BTC sales to maintain its "per-share bitcoin neutral" principle. Didier views this as a financial "experiment" testing market capacity for such recurring sell pressure, which, while creating near-term structural headwinds, likely avoids a true "death spiral" absent major new external shocks. Shifting to AI, Didier argues that tokens are becoming the new form of labor, with AI models and compute (tokenized inputs) increasingly replacing human roles in execution and middle-management. This drives enterprise efficiency and higher margins, fueling the sustained rally in U.S. semiconductor, data center, and infrastructure stocks. He foresees an emerging "machine economy" where automated agents transact and collaborate on-chain. Regarding crypto exchanges offering U.S. equities, Didier sees this as a natural evolution. With few crypto-native assets generating lasting value, exchanges are pivoting towards real-world assets (RWAs) like stocks and bonds. This doesn't necessarily cannibalize crypto but reflects a maturing industry focusing on blockchain's core utilities: decentralized choice and efficient settlement. He notes that trading logic for crypto natives doesn't need to drastically change, as meme-driven and fundamentalist strategies find analogs in U.S. markets. The "1011 event" (likely referring to a major market crash) severely damaged crypto market liquidity, marking a probable end to the altcoin speculative cycle, with capital flowing towards the deeper liquidity of U.S. markets. For the macro outlook, Didier is cautious about near-term market pressure from potential mega-IPOs (e.g., SpaceX) and the U.S. midterm elections, which could bring more regulatory scrutiny. Long-term, he remains bullish on AI's productivity gains and its convergence with blockchain/Web3, predicting a shift from speculative frenzy to a more institutionalized, industrial phase for the crypto sector.

marsbit44m ago

Conversation with Investor Zheng Di: MicroStrategy's Coin Sale Experiment, AI Economy, and Opportunities in US Stocks

marsbit44m ago

Playnance’s $GCOIN Lists on KoinBX Amid Rapid Growth in India

Playnance's native token, $GCOIN, has been listed on the cryptocurrency exchange KoinBX as of June 18. This move aims to enhance accessibility for its rapidly growing community, particularly in India, where the blockchain-powered Web3 iGaming ecosystem has gained significant traction. Over 130 partners in Playnance's "Be the Boss" program have built communities engaging thousands of active players in the region. The "Be the Boss" model allows participants to create and manage their own gaming communities, earning rewards tied to community activity. CEO Pini Peter noted India's high engagement, with community leaders successfully building player networks. One partner, Dr. Nicolas, reported earning over $57,000 through the program in recent months, highlighting both the financial rewards and the opportunity to grow an engaged community. $GCOIN serves as the ecosystem's core utility token, incentivizing participation and aligning the interests of players and community leaders ("Bosses"). The listing on KoinBX is part of Playnance's strategy to expand globally, increasing the token's utility and accessibility by combining community ownership, gamified engagement, and blockchain-based incentives. Founded in 2020, Playnance is a Web3 iGaming infrastructure company focused on creating live, non-custodial, on-chain products to onboard mainstream users. It currently processes approximately one million transactions daily, aiming to simplify the user experience while maintaining full on-chain transparency.

TheNewsCrypto1h ago

Playnance’s $GCOIN Lists on KoinBX Amid Rapid Growth in India

TheNewsCrypto1h ago

STRC Hits Historic Low, Saylor's Perpetual Motion Machine Grinds to a Halt

STRC, the perpetual preferred stock issued by MicroStrategy to fund its Bitcoin purchases, hit a historic low of $85.32, a 17% discount to its $100 par value. Designed as a "digital credit engine" to trade stably near par and enable continuous share issuance for buying Bitcoin, its plunge signals a breakdown in this model. Three key factors drove the decline: 1. Bitcoin's price fell over 50% from its peak, trading around $63,000 amid hawkish Fed signals. 2. MicroStrategy's cash reserves were depleted after a $1.5 billion convertible note repayment, slashing the dividend coverage for STRC's 11.5% yield to ~7 months. The company then sold 32 BTC to cover dividends—Michael Saylor's first Bitcoin sale since 2022—damaging the "never sell" narrative. 3. A competing Bitcoin-backed preferred stock, Strive's SATA, offers a higher yield (~13%) and daily dividends, drawing investors away from STRC. The drop triggers a negative cycle: STRC below par halts ATM share issuances, cutting off a key funding source for Bitcoin buys and potentially forcing more BTC sales for dividends, further eroding confidence. While Saylor argues the model is mathematically sound—needing only 2.3% annual Bitcoin growth to sustain itself—the market is testing the resilience of the leveraged Bitcoin treasury strategy in a bear market. The STRC price now reflects rising skepticism about this financial machinery's durability during downturns.

marsbit1h ago

STRC Hits Historic Low, Saylor's Perpetual Motion Machine Grinds to a Halt

marsbit1h ago

Trading

Spot
Futures

Hot Articles

Discussions

Welcome to the HTX Community. Here, you can stay informed about the latest platform developments and gain access to professional market insights. Users' opinions on the price of AI (AI) are presented below.

活动图片