NVIDIA GeForce 256 · NV10

The photographer identifies this optical die specimen as NVIDIA GeForce 256, also labelled NV10. The specimen record should remain linked to the GeForce 256 design while preserving its own identity and image credit. Its visible surface is evidence of a photographed piece of silicon; it does not independently establish clock speed, memory fitted to a graphics card, or a functional boundary around every repeated-looking structure.

NVIDIA’s retrospective places transformation, lighting, setup, and rendering in separate engines within GeForce 256. It describes the GPU using a 220nm process label and a 120 MHz clock. The architectural change is the placement of geometry work: transforming vertices and calculating lighting could be performed by dedicated graphics hardware. Those responsibilities should not be described using modern CUDA-core or tensor-core terminology, which belongs to later architectures.

NVIDIA’s 2004 historical pipeline diagram connects application work on the CPU to the graphics processor through AGP. On the GPU side it shows transform and lighting, rasterization, texturing, combining, and raster operations, with textures and the framebuffer in video memory. VLSI Museum interpretation: incoming scene data and outgoing display pixels are different streams. Moving geometry calculations into the GPU does not make texture and framebuffer traffic disappear.

NVIDIA’s later architecture brief explicitly identifies GeForce 256 as an example of DirectX 7 fixed-function transform and lighting, followed by programmable vertex shaders in the DirectX 8 era. It explains the classic sequence from vertices to assembled primitives, rasterized fragments, and framebuffer pixels. This historical description supplies an architectural vocabulary; the surrounding document’s GeForce 8800 shader machinery must not be imported into an NV10 exhibit.

VLSI Museum interpretation: a fixed pipeline invites stage-by-stage reasoning. Increasing geometry detail asks more of vertex processing, while a change in fragment coverage or texture use shifts work elsewhere in the pipeline. A triangle-rate claim therefore cannot stand in for a measured frame rate in every game. NVIDIA’s conceptual diagram supports that distinction, but provides neither workload-specific bottleneck measurements nor coordinates for annotating this specimen.

NVIDIA’s demo archive identifies GeForce 256 as the original target of Bubble and Grass. Bubble demonstrates reflections through cube environment mapping; Grass combines many blades of grass with cloud effects and interactive weather. These preserved examples show how software authors presented graphics capabilities through particular scenes. They are useful historical companions to the die, but demonstrations of selected techniques are not independent, comparable performance benchmarks.