A 3D graphics accelerator operates in parallel with a host central processing unit (CPU). Software executing on the host CPU performs transformation and lighting operations on 3D-object primitives such as triangles, and generates gradients across the triangle for red, green, blue, Z-depth, alpha, fog, and specular color components. The gradients for texture attributes are also generated and sent to the graphics accelerator. Both the graphics accelerator and the CPU software perform triangle edge and span walking in synchronization to each other. The CPU software walks the triangle to interpolate non-texture color and depth attributes, while the graphics accelerator walks the triangle to interpolate texture attributes. The graphics accelerator performs a non-linear perspective correction and reads a texture pixel from a texture map. The texture pixel is combined with a color pixel that is received from the CPU software interpolation of non-texture attributes. Once the texture pixel from the graphics accelerator and the color pixel from the CPU software are sent to a blender in the graphics accelerator, both continue to interpolate the next pixel in the horizontal-line span, or move to a pixel in the next span. Both the CPU software and the graphics accelerator interpolate the same pixel at the same time. Using both the CPU and the graphics accelerator improves performance since both operate in parallel on the same pixel at the critical interpolation bottleneck.