• Weekly AI News
  • Posts
  • World's first transformer ASIC, Etched Sohu Vs. Nvidia Blackwell, Amazon 750000 Robots

World's first transformer ASIC, Etched Sohu Vs. Nvidia Blackwell, Amazon 750000 Robots

Competition Heats Up in the AI Chip Industry, Amazon Replaces Over 100,000 Jobs with 750,000 Robots, OpenAI Introduces CriticGPT, & More...

Author

Weekly AI News
July 02, 2024 • Estimated Reading Time: 8 minutes

🌐 AI News

  • 🔜 Up Next:
    - July 15-17, 2024 Fortune Brainstorm Tech 2024 (Park City, Utah). Register Now

🚀 Top AI Highlights

The Sohu AI Chip by Etched!

The Sohu AI chip, developed by Etched, was founded by Harvard dropouts Gavin Uberti and Chris Zhu. It is specifically engineered for transformer models, utilizing TSMC’s 4nm process to offer unmatched performance and energy efficiency.

Key Features

  • Performance: Outpaces Nvidia’s next-generation Blackwell B200 GPUs, processing over 500,000 tokens per second on the Llama 70B model.

  • Energy Efficiency: Consumes only 10 watts of power, making it highly efficient for AI operations.

Technological Advancements

  • TSMC’s 4nm Process: Provides high performance and cost-efficiency.

  • Specialized Design: Optimized for transformer models, unlike general-purpose GPUs.

  • ASIC Technology: Custom-designed for specific applications, enhancing efficiency.

  • Transformer Model Optimization: Efficiently executes operations like matrix multiplications and attention mechanisms.

Comparative Performance

  • Throughput: Sohu is more than 10 times faster and cheaper than Nvidia’s Blackwell B200 GPUs.

  • Energy Consumption: Lower than high-end GPUs, with only 10 watts of power usage.

  • Benchmarking: A single 8xSohu server can replace 160 H100 GPUs.

Market Impact

  • AI Product Power: Powers major AI products like ChatGPT, Claude, Gemini, and Sora.

  • Specialization Trend: Indicates a shift towards specialized chips for transformer models, offering substantial performance boosts.

Future Prospects

  • Economic Viability: With AI models costing billions to train and used for inference worth even more, specialized chips like Sohu become economically justified.

  • Industry Shift: As ASICs like Sohu become mainstream, the focus will be on making transformers more efficient, redefining the AI hardware landscape.

Competition Heats Up in the AI Chip Industry!

Nvidia has long been the leader in AI chips, trusted for its high-performance GPUs. However, Etched, a startup, claims its new Sohu chip is significantly faster and more affordable than Nvidia’s Blackwell chips, backed by $120 million in funding.

What Makes Etched’s Sohu Chip Stand Out? Etched’s Sohu chip, designed for large language models, claims to outperform Nvidia’s Blackwell chips by focusing on matrix multiplication efficiency. If true, this could revolutionize the AI chip market with faster and cheaper solutions.

How Valid Are Etched’s Claims? Etched argues that Nvidia's chips are not fully optimized for AI tasks, while Sohu’s architecture is more efficient. However, experts note Nvidia’s extensive optimization efforts and exceptional chip performance, making Etched’s claims challenging to fulfill.

What Is the Competitive Landscape Like? The AI chip market is highly competitive with Nvidia, AMD, Cerebras, and Graphcore. Etched must prove Sohu’s performance and cost benefits and overcome market reliance on Nvidia’s solutions.

What Are Etched’s Chances of Success? Success depends on Sohu delivering superior performance and cost-effectiveness, gaining industry trust, and continuous innovation. Nvidia is expected to continue its heavy R&D investment, leveraging its extensive ecosystem to counter new entrants like Etched.

Who Will Lead the Future of AI Chip Technology? Etched's claims are intriguing, but dethroning Nvidia is challenging due to its expertise, market presence, and continuous innovation. The competition promises to drive advancements, benefiting the AI industry and leading to exciting developments in AI chip technology.

😍 Enjoying so far, share it with your friends!

Amazon Replaces Over 100,000 Jobs with 750,000 Robots!

Amazon, the world's second-largest private employer, has increased its robotic fleet to over 750,000, reducing its human workforce from 1.6 million in 2021 to 1.5 million.

The company's robots, including new models like Sequoia and Digit, perform repetitive tasks to enhance efficiency and safety.

Despite concerns about job displacement, Amazon claims that automation has led to the creation of 700 new skilled job categories. This trend reflects a broader shift towards integrating advanced technologies, with human employees focusing on more complex tasks requiring higher skill levels.

👥 Connect & Feedback!

👉 Join Us:

📧 Advertise In Weekly AI News:

📧 Contact directly at [email protected]

😍 Share with your friends!

Reply

or to participate.