12nm Process vs Flagships: Bolt Graphics Tests Zeus Architecture

Bolt Graphics has announced the successful completion of testing for its Zeus GPU. According to the developers, the dual-chip version of this product can outperform the flagship NVIDIA RTX 5090 by five times in path tracing tasks.

Zeus will be released in two form factors: PCIe cards and 2U server configurations. There will be both single-chip models, such as “Bolt Zeus 1c26”, and dual-chip versions.

RISC-V and Architectural Scalability

Unlike traditional solutions from NVIDIA or AMD, the Zeus architecture is built around RISC-V cores and specialized accelerators. The test chip was manufactured by TSMC using a 12nm (FFC) process. While this may seem outdated compared to modern 4nm or 5nm nodes, Bolt Graphics emphasizes the flexibility of the architecture, which can be adapted to 5nm in the future.

Memory Specifications and Compute Performance

Instead of expensive GDDR memory, Bolt Graphics integrates support for LPDDR5X and standard DDR5 SO-DIMM slots. This allows memory capacity to scale up to 160 GB, albeit with a bandwidth of 363 GB/s.

The effectiveness of Zeus in graphics workloads and gaming remains uncertain due to significantly lower FP32 performance compared to modern gaming GPUs. While it delivers strong FP64 (double precision) performance, FP32 (single precision) reaches only around 10 TFLOPS, which is roughly ten times lower than the RTX 5090.

Memory bandwidth limitations may also impact AI workloads.

Market Positioning

Bolt Graphics emphasizes that its solution is not a direct competitor to mainstream gaming GPUs, but rather a specialized tool for scientific computing and simulation.

The total cost of ownership (TCO) is expected to be up to 17 times lower than comparable solutions, which could attract research institutions and cloud providers. However, mass production is only planned for the fourth quarter of 2027. At this stage, the architecture remains experimental.