Bolt Graphics, a US-based GPU startup founded in 2020, aims to challenge Nvidia by focusing on affordable, expandable GPUs optimized for ray tracing and professional content creation, featuring a unique architecture with RISC-V cores, flexible memory options, and remote management capabilities. Currently in prototype testing with plans for mass production by late 2025, Bolt targets mid-to-high-end markets with lower power consumption and cost, positioning itself as a versatile alternative that leverages the industry’s shift toward ray tracing technology.
Bolt Graphics is a new US-based GPU company aiming to challenge Nvidia by targeting markets Nvidia has somewhat neglected, particularly creative professionals and content creators, before expanding into gaming. Founded in 2020 by Darw Sin, Bolt has developed a unique GPU architecture that emphasizes affordability, expandability, and accessibility. Their GPU prototype features unconventional design choices such as user-expandable DDR5 SO-DIMM slots, soldered LPDDR5X memory, dual PCIe slots, and an RJ45 Ethernet port for remote management, reflecting a vision of a versatile, single-board computer-like GPU that can run Linux and support remote telemetry.
The company’s GPU architecture is distinct from traditional designs, incorporating a high-performance out-of-order RISC-V CPU core alongside vector cores analogous to shader processors, and specialized hardware accelerators for ray tracing, physics simulation, and math functions. This design aims to optimize ray tracing performance, a feature Bolt believes is the future of computer graphics, even though it currently comes with trade-offs in rasterization performance. Bolt’s approach focuses on delivering superior ray tracing capabilities, targeting professional workloads like 3D rendering and content creation, with plans to gradually support gaming applications by working closely with developers to port and optimize software.
Bolt’s memory strategy prioritizes cost-effectiveness and capacity over raw bandwidth, using widely available LPDDR5X and DDR5 SO-DIMM modules to offer flexible and expandable memory configurations. This approach contrasts with Nvidia’s use of high-bandwidth, proprietary memory solutions like GDDR6X and HBM, allowing Bolt to keep prices lower and memory more accessible. The company also employs a chiplet design connected via a high-performance UCI fabric, enabling scalability and future integration of advanced technologies like co-packaged optics for improved inter-chip communication.
Currently, Bolt is in the prototype and testing phase, using AMD Xilinx FPGAs to simulate parts of their GPU hardware and running demos such as Blender’s Cycles renderer to validate performance and functionality. They plan to tape out a full 5nm chip by the end of the year and aim for mass production by the end of next year. Bolt has garnered interest from major studios and publishers, particularly for its ray tracing capabilities, and is preparing to release emulators to early adopters soon, with a public product launch anticipated in late 2024 or early 2025.
Strategically, Bolt positions itself as a more affordable alternative to Nvidia’s high-end GPUs, focusing on mid-range to high-end segments but with significantly lower power consumption and cost. By emphasizing ray tracing performance and targeting professional content creation markets first, Bolt hopes to carve out a niche that leverages the industry’s gradual shift toward ray tracing, potentially benefiting from Nvidia’s push in that direction. While challenges remain, especially in software and driver support, Bolt’s innovative architecture and market approach present a fresh and promising contender in the GPU landscape.