TinyGPU v2.0 Renders 3D Graphics Using Just 240,000 Transistors

Developer Pongsagon Vichit has successfully validated TinyGPU v2.0, a miniature GPU silicon that demonstrates the absolute minimum hardware required for functional 3D rendering. The project marks a significant milestone in amateur ASIC design, transitioning a functional graphics processor from FPGA prototypes to physical silicon.

Produced via the Tiny Tapeout community as part of the SKY 25b shuttle, the TinyGPU v2.0 contains approximately 240,000 transistors. Technical reports indicate the chip renders 3D graphics at frame rates between 7.5 and 15 FPS. The hardware supports a specific set of legacy-style graphics features, including transformation and lighting (T&L), backface culling, flat shading, and affine texture mapping. It can handle a maximum of 1,000 triangles on screen and supports one dynamic directional light during rasterization.

The Path to Silicon

The TinyGPU project is primarily an educational endeavor designed to explore the boundaries of GPU architecture. Vichit previously developed the project using FPGA prototypes, such as the Basys3, before transitioning to an Application-Specific Integrated Circuit (ASIC). This transition was made possible through the Tiny Tapeout initiative, a community-driven platform that allows independent designers to share silicon space on a production shuttle, drastically lowering the barrier to entry for custom chip fabrication.

Implications for Hardware Education

While TinyGPU v2.0 is not commercially competitive with modern hardware, its value lies in its role as a practical educational tool. By stripping a GPU down to its most basic functional components, the project provides a transparent look at how rasterization and T&L operate at the transistor level. It serves as a proof of concept for the accessibility of modern ASIC fabrication, proving that individuals can now design and manufacture specialized hardware without the backing of a major semiconductor corporation.

Future Iterations

Development has already shifted toward the next generation of hardware. TinyGPU v3.0 is expected to be delivered between December 2026 and January 2027. This upcoming version will introduce a programmable pixel shader inspired by DirectX 8 PS 1.1, utilizing SIMD architecture to process four pixels in parallel. Vichit noted the necessity of this rapid iteration, jokingly remarking that the GPU business is "cut-throat" and requires fast release cycles.