Nvidia’s H100 and B200 GPUs have revolutionized the AI hardware market. These advanced chipsets power AI chatbots, making complex tasks like asking ChatGPT “how to eat a mango without spilling it” possible. Nvidia’s advancements have propelled its market capitalization to $3 trillion, surpassing tech giants Microsoft and Apple last month.
However, a new contender has emerged in the AI hardware space. Etched, a California-based startup founded by two Harvard dropouts aims to disrupt the market with their specialized ASIC (Application Specific Integrated Circuit) chip called Sohu. According to Etched’s claims, Sohu is 20 times faster in running transformers like ChatGPT compared to Nvidia’s H100.
Limitations and Potential
The startup claims that Sohu is 20 times faster in running transformers like ChatGPT, which could revolutionize AI performance. While Sohu excels in running transformer models, it cannot run other AI models like Convolutional Neural Networks (used for image recognition). However, this specialization opens new possibilities for AI developers. For instance, Sohu could enable real-time translation between languages like Hindi, Gujarati, Tamil, French, English, and German. It also holds potential for integrating visual and language processing, allowing for simultaneous understanding of text and images.
Etched raised $120 million on June 25 to bring Sohu to market, though the release timeline remains unclear. The company claims to have secured “tens of millions of dollars” in preorders. Etched has also partnered with TSMC (Taiwan Semiconductor Manufacturing Company) to fabricate the 4-nanometer chip, which will help ramp up production in the first year.
Nvidia’s AI Hardware Leadership
Nvidia’s H100 and B200 GPUs have set a high standard in the AI hardware market. These GPUs power advanced AI chatbots and have contributed to Nvidia’s impressive market capitalization of $3 trillion, surpassing Microsoft and Apple.
Nvidia’s GPUs are versatile, handling various computational tasks, including real-time graphics rendering and AI model training. This versatility has cemented Nvidia’s leadership in the AI hardware space.
Etched’s Bold Claims and Future Prospects
Etched, a young startup from California has entered the AI hardware market with ambitious claims about its Sohu chip. Early tests suggest that Sohu is 20 times faster in running transformers like ChatGPT, promising a significant leap in AI capabilities. These claims, if proven, could significantly disrupt the market. Sohu’s design focuses exclusively on transformer AI models, unlike Nvidia’s multi-purpose GPUs.
The specialization of Sohu in transformer models could be a double-edged sword. On one hand, it promises unprecedented speed for specific AI applications like ChatGPT, OpenAI’s Sora, and Google’s Gemini. This specialization could enable new AI capabilities, such as real-time language translation and multimodal integration of text and images. However, Sohu’s inability to handle other AI models, such as Convolutional Neural Networks, limits its versatility compared to Nvidia’s GPUs.
Market Impact and Challenges
Etched has secured $120 million in funding and has preorders worth “tens of millions of dollars.” This financial backing, along with a partnership with TSMC to manufacture the 4-nanometer chip, positions Etched well for a strong market entry. However, the actual release timeline of Sohu remains uncertain, and the real-world performance of the chip is yet to be validated.
Etched’s approach to creating a specialized chip for transformer models could fill a niche market, but it faces significant challenges. Nvidia’s established presence and proven versatility give it a strong advantage. Moreover, Etched’s performance claims are based on emulation tests, which may not fully reflect real-world conditions.
The future of AI hardware innovation will likely see a blend of specialization and versatility. Etched’s Sohu chip represents a bold attempt to disrupt the AI hardware market with a specialized, high-performance product. While the potential benefits of Sohu are significant, its success will depend on delivering on its performance claims and overcoming the limitations of its specialized design. Nvidia’s established dominance and versatile GPUs will continue to pose a formidable challenge.
Also Read: Sam Altman Said That Brian Chesky Helped Him Grow: The Secret Behind ChatGPT’s Success.
