The raw compute gap between these two GPUs is enormous. The RTX 5090 delivers 104.9 TFLOPS of floating-point performance against the RX 9070 GRE's 34.3 TFLOPS — roughly a 3× advantage — driven primarily by its massive 21,760 shading units compared to the RX 9070 GRE's 3,072. In practice, this translates directly into substantially higher throughput in compute-heavy workloads like ray tracing, AI-accelerated rendering, and simulation tasks. The RTX 5090 also holds a commanding lead in texture throughput (1,638.8 GTexels/s vs. 535.7 GTexels/s) and pixel fill rate (424.2 GPixel/s vs. 267.8 GPixel/s), meaning it can push far more geometry and pixels per second — a tangible benefit at very high resolutions and in geometry-dense scenes.
Interestingly, the RX 9070 GRE punches back in clock-speed metrics. Its GPU turbo of 2,790 MHz surpasses the RTX 5090's 2,410 MHz, and its memory runs at a faster 2,250 MHz versus 1,750 MHz. Higher clock speeds can improve per-thread latency and single-tile performance, but with only a fraction of the shader and TMU count, the RX 9070 GRE cannot overcome the RTX 5090's parallelism advantage in aggregate throughput. The faster memory speed on the RX 9070 GRE is a genuine strength for bandwidth-sensitive workloads, though without knowing bus width from these specs alone, it cannot fully offset the 5090's overwhelming core advantage.
Both cards support Double Precision Floating Point (DPFP), making them each viable for professional compute tasks that require 64-bit precision. Overall, the RTX 5090 holds a decisive performance edge across virtually every throughput metric in this group. The RX 9070 GRE's higher boost clock and memory speed are noteworthy, but they represent incremental advantages that do not close the wide gap in shader count and compute power. For users prioritizing maximum raw performance, the RTX 5090 is in a different class entirely.