Ray Tracing Architecture: How NVIDIA RTX and AMD RDNA 4 Implement Hardware Ray Acceleration

Ray Tracing Architecture: How NVIDIA RTX and AMD RDNA 4 Implement Hardware Ray Acceleration

Hardware-accelerated ray tracing is now standard in any discrete GPU released after 2020, but the silicon architectures implementing it differ substantially between NVIDIA and AMD — with direct consequences for how games using path tracing and RTGI perform across product stacks.

NVIDIA RT Cores: Dedicated BVH Traversal Hardware

NVIDIA introduced the first dedicated ray tracing hardware — RT Cores — with Turing (RTX 2000 series, 2018). Each SM cluster contains one RT Core responsible for BVH traversal and ray-triangle intersection testing. Blackwell (RTX 5000 series, 2025) brings 3rd-generation RT Cores with approximately 191 TFLOPS traversal throughput on the RTX 5090, plus compressed BVH nodes, displaced micro-mesh geometry support, and enhanced opacity micromap support.

AMD RDNA 4: Ray Accelerators on Compute Units

AMD implements Ray Accelerators co-located within each Compute Unit’s SIMD32, coupling ray traversal throughput to CU count. RDNA 4 (RX 9000 series, 2025) doubles Ray Accelerator throughput per CU versus RDNA 3. The RX 9070 XT achieves approximately 97 TFLOPS of ray traversal, competitive with RTX 4070 but trailing RTX 5070 Ti by roughly 35%.

API Implementation

Games access ray hardware through DirectX Raytracing (DXR) and Vulkan Ray Tracing (VK_KHR_ray_tracing_pipeline). Both expose a pipeline state object model with five shader types: ray generation, intersection, any-hit, closest-hit, and miss shaders. Acceleration structures operate at two levels: BLAS (per-mesh geometry) and TLAS (per-instance transforms).

Conclusion

The architectural divergence between NVIDIA’s dedicated RT Core approach and AMD’s integrated Ray Accelerator model reflects two valid engineering philosophies — both capable of supporting modern path-traced content, but with different performance profiles. For GPU purchases targeting path tracing, workload-specific benchmarks across Cyberpunk 2077 RT Overdrive and Alan Wake 2 remain the most reliable performance indicators.