AIBRIEF
Pongsagon Vichit's TinyGPU v2.0 came back working from a community fabricated chip run, with a programmable pixel shader queued for 2026.
A single designer sent a 240,000-transistor GPU to a community-fabricated silicon shuttle and got it back rendering 3D on the first try. What closed this week is the design-to-silicon loop, not the frame rate.
TinyGPU v2.0 was fabricated on Tiny Tapeout's SKY 25b shuttle, a community service that ships small custom chip designs to a real foundry at prices a hobbyist can afford. Designer Pongsagon Vichit confirmed the milestone himself: "ASIC TinyGPU v2.0 is alive and working!!!" The v2.0 design files are open on GitHub, as is the planned v3.0.
The chip carries about 240,000 transistors, above the 200,000 figure floated in pre-tapeout coverage, and runs a fixed-function pipeline: transformation and lighting, rasterization, one dynamic directional light, flat shading, backface culling, affine texture mapping, a 4-bit double buffer, and an 8-bit depth buffer on external QSPI RAM. In a demo scene capped at 1,000 triangles on a 320×240 framebuffer, it hits 7.5 to 15 FPS. That matches the prior Basys3 FPGA test, not a speedup, and is not a benchmark against commercial GPUs.
Vichit had already submitted v3.0 to the next Tiny Tapeout run before v2.0 came back from the fab. The new version adds a programmable pixel shader, a DirectX 8 PS 1.1-inspired design with SIMD-4 execution, 26 instructions, and branching via masked execution, with a target of late 2026. As Tom's Hardware notes, the designer's self-aware line on shipping cadence: "we all know that GPU is a cut-throat business. Must release new models fast."