From GeForce 256 to RTX 5090

These four NVIDIA reference cards span twenty-six years, from the GeForce 256 in 1999 to the GeForce RTX 5090 in 2025. The nominal GPU-clock and theoretical memory-bandwidth rows trace separate hardware properties. Neither produces a performance score. The first column is the 32 MiB DDR reference configuration; the earlier 32 MiB SDR configuration remains available in the comparison builder.

The later cards establish clocks and theoretical frame-buffer bandwidths. Their shared methods put the figures in the same rows, but the GeForce 256 DDR bandwidth is a Tier C figure. It stays provisional, so this view does not turn the 2006, 2016, and 2025 figures into ratios against it. The rows remain a sourced record of changing hardware quantities across the full period.

The GeForce 256 table gives a provisional 960 MFLOPS FP32 figure. Its raster rates use different metrics. The later cards state theoretical FP32 peaks, but those describe one calculation at a specified clock, not every generation’s games, rendering operations, memory behavior, or fixed-function work.

Power declarations remain separate. The RTX 5090 lists rated power consumption, while the 8800 GTX and GTX 1080 list thermal-design power; no checked source states a board-power rating for either GeForce 256 reference configuration. A thermal design target and a rated-power figure are different manufacturer claims. The catalog keeps their rows apart and does not present either as performance.