Bolt Graphics has achieved a significant hardware milestone by powering on the first silicon for its Zeus GPU project. Although the test chip functioned correctly during initial trials, the final hardware intended for volume production remains several years away from market release. Based in California, the startup confirmed that its 12nm test chip, manufactured by Taiwan Semiconductor Manufacturing Company (TSMC), operated without issues upon activation. To maintain close coordination with its manufacturing partners, the company also established a new office in Taipei.
Measuring six square millimeters, the test chip consumes 1.4 watts of power. This efficiency marks a substantial reduction from the 35 watts required by the field-programmable gate array (FPGA) previously used for prototyping. The primary architectural feature of the hardware is dedicated circuitry designed specifically for electromagnetic simulation. This capability assists engineers in developing diverse optical and electronic systems, ranging from camera lenses to radar sensors.
According to CEO and founder Darwesh Singh, the successful test demonstrates the hardware acceleration of a highly complex physics algorithm. While independent performance metrics are currently unavailable, Bolt reports that the silicon exceeded internal performance estimations. The production-ready Zeus GPU will utilize TSMC’s N5P process node rather than the 12nm technology used for the prototype. Selecting this mature process allows the company to secure manufacturing capacity without competing directly for heavily allocated wafer supplies.
The architecture of the Zeus GPU integrates open-standard RISC-V processor cores with proprietary path-tracing hardware. To maximize system versatility, the upcoming PCIe expansion cards will feature a hybrid memory architecture combining soldered LPDDR5X chips with standard laptop-style DDR5 SO-DIMM slots. This configuration will enable users to scale memory capacity up to 384GB. Initial performance projections released by the company claimed significant advantages over existing hardware, including vastly superior path-tracing throughput and rendering speeds compared to established consumer cards.
Industry adoption is initially targeted at scientific, engineering, and studio environments before expanding into artificial intelligence and consumer gaming markets. Competing in the consumer sector will require extensive software ecosystems to support thousands of existing applications. Despite these upcoming software challenges, the company reported substantial early interest, with thousands of enterprise and individual users registering for early access alongside a sizable prospective product pipeline.
