钛媒体:引领未来商业与生活新知 前天 13:26
Intel Launches AI-Focused GPU Lineup to Take on Nvidia
index_new5.html
../../../zaker_core/zaker_tpl_static/wap/tpl_guoji1.html

 

在COMPUTEX 2025上,英特尔发布了一系列AI芯片,包括Arc Pro B60和B50 GPU,分别提供24GB和16GB显存,专为开发者和专业用户设计。CEO Lip-Bu Tan承认英特尔面临挑战,并强调公司正在进行文化改革,鼓励直接沟通和反馈。英特尔还推出了Project Battlematrix,这是一个模块化的Xeon平台,最多可运行八个Arc Pro B60 GPU。此外,Gaudi 3 AI加速器也推出了PCIe和机架级配置。英特尔正通过加强AI硬件、文化透明度和全球合作,力图重塑其在半导体行业的地位。

🚀 **AI芯片发布:** 英特尔推出了Arc Pro B60和B50 GPU,分别配备24GB和16GB显存,针对架构、工程和AI推理工作负载进行了优化,B60尤其支持多GPU扩展。

🗣️ **领导层坦诚:** CEO Lip-Bu Tan承认英特尔部分产品缺乏竞争力,并强调公司正在做出改变以解决这些问题,同时强调了内部文化改革的重要性,鼓励直接沟通和反馈。

⚙️ **Project Battlematrix平台:** 英特尔推出了Project Battlematrix,这是一个基于Xeon的模块化平台,能够运行多达八个Arc Pro B60 GPU,总共提供192GB的VRAM,适用于中型AI模型。

💡 **Gaudi 3 AI加速器:** 英特尔的Gaudi 3 AI加速器现在提供PCIe和机架级配置,每个机架最多支持64个加速器和8.2TB的高带宽内存,尽管无法出口到中国,但英特尔计划将其与Arc Pro GPU和Xeon CPU配对,以满足国内AI计算需求。

Intel CEO Lip-Bu Tan

TMTPOST — While Intel skipped a keynote speech at the 2025 Taipei International Computer Show (COMPUTEX), the U.S. chipmaker made waves with a major product unveiling and candid leadership remarks that underscore its ambition to reclaim competitive ground in a rapidly evolving semiconductor landscape.

On May 19, Intel introduced a new suite of AI-focused graphics processing units (GPUs) and accelerator chips tailored for developers and professional users. The launch includes the Arc Pro B60 and Arc Pro B50, offering 24GB and 16GB of memory, respectively. The B60 model, in particular, supports multi-GPU scalability, and both units are optimized for architecture, engineering, construction (AEC), and AI inference workloads.

During an Intel-hosted dinner in Taipei, CEO Lip-Bu Tan acknowledged Intel's recent struggles. "Some of Intel's products currently lack competitiveness," he said, noting that the company is "making changes to address these shortcomings." Despite facing challenges, Tan pointed out Intel still commands 68% of the PC and client processor market and 55% of the data center CPU segment.

Tan emphasized cultural reform inside Intel, calling for direct communication and truth-telling. "Sometimes too many layers in the hierarchy can distort communication," he said. To address this, Tan often consults engineers "seven or eight levels down" the corporate ladder. He also restructured the engineering division to report directly to him.

Execution is Intel's top priority, he stressed. "The best way to revitalize Intel is through execution, execution, and more execution," Tan said. That includes a no-nonsense approach to customer and internal feedback: "Tell me the bad news first. Tell me where things went wrong."

Intel's newly launched Arc Pro B series supports both consumer and professional drivers on Windows and enables containerized deployments on Linux—an increasingly important feature for AI workloads. The company also introduced Project Battlematrix, a modular Xeon-based platform capable of running up to eight Arc Pro B60 GPUs with 192GB of combined VRAM. This configuration is suited for mid-sized AI models with up to 150 billion parameters.

In a press briefing, Intel representatives claimed the new GPU lineup offers "the highest configuration in the mainstream price segment." For comparison, NVIDIA's RTX A1000 and RTX 2000 Ada feature 8GB and 16GB of memory, respectively.

The Arc Pro B60 delivers up to 197 TOPS (Int8), while the B50 clocks in at 170 TOPS, both compliant with U.S. export control regulations—clearing them for sale in mainland China. Intel expects B60 samples to reach partners like ASRock and Gunnir by June, with the B50 to be distributed through authorized channels starting in July.

The company also shared updates on the Intel Gaudi 3 AI accelerator, which now comes in both PCIe and rack-level configurations. Designed to support scalable AI inference across models like Llama, Gaudi 3 features up to 64 accelerators per rack and 8.2TB of high-bandwidth memory. Though Gaudi 3 cannot be exported to China, Intel plans to pair Arc Pro GPUs with Xeon CPUs to meet domestic demand for AI compute power.

In a further nod to the developer ecosystem, Intel launched the AI Assistant Builder, an open-source software stack for building AI agent products on Intel platforms.

The announcements come at a pivotal time for Intel. Tan, who also spoke at the dinner, acknowledged recent missteps. "To be honest, it's a bit embarrassing," Intel CFO David Zinsner said last week, referring to a string of underwhelming product launches.

Still, Gelsinger painted an optimistic picture. Marking Intel's 40-year partnership with Taiwan's technology ecosystem, he credited the region with playing a foundational role in the global success of the x86 architecture.

Gelsinger also shared personal reflections on his time in Taiwan, including a stint at Acer that stretched from a planned three months to 15 years. "Don't casually say three months," he joked, "because three months can turn into fifteen years."

Now, as he returns to Taipei as Intel's CEO, Tan emphasized a renewed commitment to collaboration and transformation: "Together, we can make the world a better place," he said, outlining a future centered on data centers, PCs, edge computing, and the cloud.

As Intel doubles down on AI hardware, cultural transparency, and global partnerships, Computex 2025 may mark an inflection point in the company's efforts to regain its footing against rivals like Nvidia and TSMC—and to write a new chapter in its storied semiconductor legacy.

更多精彩内容,关注钛媒体微信号(ID:taimeiti),或者下载钛媒体App

Fish AI Reader

Fish AI Reader

AI辅助创作,多种专业模板,深度分析,高质量内容生成。从观点提取到深度思考,FishAI为您提供全方位的创作支持。新版本引入自定义参数,让您的创作更加个性化和精准。

FishAI

FishAI

鱼阅,AI 时代的下一个智能信息助手,助你摆脱信息焦虑

联系邮箱 441953276@qq.com

相关标签

英特尔 AI芯片 COMPUTEX 2025 Gaudi 3
相关文章