Featuring a PCIe 5.0 x16 interface, the NVIDIA RTX 5090 GPU features 21, 760 shading units, 680 texture mapping units, and 176 raster output units or ROPs. Improving the speed of machine learning applications, the RTX 5090 also delivers 680 tensor cores, plus 170 ray tracing cores for improved specular lighting and reflections. The GeForce RTX 5090 provides the lowest latency of all of the Blackwell architecture NVIDIA RTX family. Compared to the RTX 5080 GPU’s 960 GB/s memory bandwidth, the RTX 5090 delivers 1792 GB/s, not to mention 3352 AI TOPS compared to just 1801 AI TOPS on the RTX 5080 GPU. Also, 3x NVIDIA encoders compared to only 2x on the RTX 5080.
All in, the card has 92,200 million transistors and gamers will be happy to know it does offer DirectX 12 for support of the latest games and modern games to be released. Outfitted with Deep Learning Super Sampling 4 or DLSS4 it renders photo-realistic images and offers multi-frame generation to boost frames per second or FPS when gaming. There are 3x DisplayPort 2.1b and an HDM 2.1b connector for support of up to 4x monitors. It required a 16-pin power connector to support a 575 W maximum power draw. The recommended PSU for the host system is at 1000 W. Base clock is listed at 2017 MHz with boost up to 2407 MHz.
