Multiple Startups Challenge Nvidia from Different Paths: Eliminating DRAM, Interconnect, or Compute-Memory Separation
7x24h News
Multiple Startups Challenge Nvidia from Different Paths: Eliminating DRAM, Interconnect, or Compute-Memory Separation
Multiple next-generation AI chip startups are challenging Nvidia from the perspective of data movement optimization. According to statistics from X user Deedy Das, the technical paths adopted by these companies include: Groq eliminates DRAM, Cerebras eliminates interconnects, d-Matrix eliminates the separation of compute and memory, Majestic eliminates server compute centralization, Etched, Taalas, and MatX eliminate generality, and Substrate eliminates the $400 million lithography machine. All solutions attempt to improve AI chip performance by reducing data movement bottlenecks.
Navigating Web3 tides with focused insightsTechFlow reports, on July 25, multiple next-generation AI chip startups are challenging Nvidia from the perspective of data movement optimization. According to statistics from X user Deedy Das, the technical paths adopted by these companies include: Groq eliminates DRAM, Cerebras eliminates interconnects, d-Matrix eliminates the separation of compute and memory, Majestic eliminates compute centralization in servers, Etched, Taalas, and MatX eliminate generality, and Substrate eliminates the $400 million lithography machine. All solutions attempt to enhance AI chip performance by reducing data movement bottlenecks.
