NVIDIA announced its Vera CPU on May 21, 2024. The company delivered the processors to Anthropic, OpenAI, SpaceX, and Oracle on May 19. This shipment marks a direct entry into the high-performance server chip market.
New processor uses self-developed architecture with high memory bandwidth for AI workloads
The Vera CPU uses NVIDIA's self-developed Olympus Arm architecture. Each processor contains 88 cores and supports 176 threads. The system integrates 1.5TB of LPDDR5X memory with a bandwidth capacity of 1.2TB per second. This configuration makes it the first data center CPU to utilize this specific memory standard.
NVIDIA claims the chip delivers industry-leading performance per watt. A rack containing 256 Vera CPUs can support over 22,500 concurrent agent instances under full load conditions. CFO Colette Kress stated the company aims to become a leading global CPU supplier and expects $20 billion in revenue from this segment this year.
The launch positions NVIDIA as a major player in enterprise computing infrastructure. The company reported Q1 FY2027 revenue of $81.6 billion, representing an 85 percent increase year over year. This financial growth underscores the strategic importance of expanding beyond graphics processing units into general-purpose central processing.
NVIDIA has not disclosed specific pricing or broader commercial availability for the Vera CPU. The company continues to focus on delivering these processors directly to select enterprise and AI research partners.
Discussion
0 comments
Log in to join the thread with a thoughtful take, question, or correction.