Bolt Graphics, a US-based GPU startup founded in 2020, aims to challenge Nvidia by focusing on content creation and professional workstations with innovative, scalable GPUs featuring a unique architecture that integrates RISC-V CPU cores and prioritizes ray tracing performance. Planning mass production by next year, Bolt targets affordability and flexibility through modular designs and larger memory capacities, positioning itself in underserved market segments before expanding into gaming.
Bolt Graphics is a new US-based GPU company aiming to challenge industry giant Nvidia by targeting market segments Nvidia has recently deprioritized, such as content creation and professional workstations. Founded in 2020 by Darw Sin, Bolt has developed prototype GPUs with a unique architecture and plans to enter mass production by the end of next year. Their approach focuses on making GPUs accessible and affordable, avoiding ultra-high-end exotic cards that few can buy. Bolt’s initial strategy is to build a foothold in creative and professional markets before expanding into gaming, addressing the significant challenges associated with software and driver support in the gaming ecosystem.
Bolt’s GPU design is unconventional, featuring a combination of LPDDR5X soldered memory and user-expandable DDR5 SO-DIMM slots, allowing flexible memory configurations from 8GB up to 48GB or more. The card includes dual PCIe Gen 5x6 connectors and an Ethernet RJ45 port for remote management, reflecting its design as a single-board computer capable of running Linux with integrated CPU cores. This modular design supports scalability through multi-GPU configurations connected via passive ribbon cables, aiming for cost-effective expandability and flexibility uncommon in traditional GPUs.
Architecturally, Bolt’s GPU differs from conventional models by integrating a high-performance out-of-order RISC-V CPU core with vector cores analogous to shader processors, alongside specialized hardware accelerators for ray tracing, physics simulation, and complex math functions. This design prioritizes ray tracing performance over traditional rasterization, betting on the future of computer graphics moving toward real-time path tracing. While rasterization performance is lower compared to existing GPUs, Bolt’s architecture is optimized for workloads that demand high memory capacity and efficient data handling, supported by significantly larger on-chip cache to mitigate bandwidth limitations from using more affordable memory types.
Bolt has demonstrated functional prototypes using FPGA platforms running Blender with real-time path tracing plugins, showcasing the feasibility of their hardware and software integration. Their path tracing performance claims suggest substantial improvements over current GPUs like Nvidia’s RTX 5090 in ray-triangle intersection throughput, targeting both offline rendering and real-time applications. Despite skepticism about a new entrant challenging Nvidia, Bolt’s focus on specific performance niches and their chiplet-based design with plans for future optical interconnects highlight a long-term vision emphasizing scalability, power efficiency, and cost-effectiveness.
Looking ahead, Bolt plans to release emulator access to early adopters by the end of the current year and aims for mass production and sales of their GPUs by the end of next year. They intend to price their products competitively below Nvidia’s top-tier offerings to appeal to a broader market. Bolt’s roadmap includes continued hardware iterations and software development, with a keynote planned at Siggraph to share further details. Their strategy reflects a pragmatic approach to entering a difficult market by targeting underserved segments and leveraging innovative architecture to differentiate themselves from established players.