NVIDIA GeForce RTX 5090 vs NVIDIA GeForce RTX 5060 Ti

The Nvidia GeForce RTX 5090 is a high-end desktop graphics card utilizing the GB202 chip based on the Blackwell architecture. The model offers 170 SMs out of 192 (21,760 CUDA cores), and 32 GB GDDR7 graphics memory with a 512-bit memory bus (1,792 GB/s). The base clock of the GPU (reference model) is specified at 2.01 GHz, while the boost clock can reach 2.41 GHz. RTX 5090 offers 170 RT cores (+42 cores compared to the previous RTX 4090) and 680 5th Gen. Tensor cores.  

Architecture

The Nvidia GeForce RTX 5090 features a new Tensor (5th Gen) and RT-cores (4th Gen), with 3352 AI TOPS and 318 TFLOPS respectively. Blackwell architecture is the first to feature the new DLSS 4 Multi Frame generation, which, thanks to enhanced AI capabilities can generate more frames per second while using less memory and delivering better picture quality. Nvidia also improved the performance and output quality of DLSS, Frame Generation, and Ray Reconstruction features (also available for the previous RTX 4000 series). Additionally, the new models now support Reflex 2 low latency mode with Frame Warp.

Performance

Thanks to the new Tensor and RT-cores, as well as a massive bump in CUDA cores, the card is promised (by Nvidia) to offer as much as 2x better performance compared to the previous RTX 4090 (thanks to the new DLSS 4 MFG feature). The RTX 5090 is expected to be the fastest graphics card of 2025 and, presumably will deliver good performance even at 8K.

According to our own benchmarks, the RTX 5090 is 27% faster than the previous RTX 4090 in applications and games (without accounting for RT and DLSS). With the Ray Tracing (Cyberpunk 2077 2.1, DLSS Transformer), gaming performance can be 25% higher compared to the previous flagship model, and DLSS Frame Generation (2x) can deliver even 28% higher FPS. Thanks to the MFG 4x mode the frame counter can be even 2x-4x higher.  

Power consumption

According to Nvidia, the GB202 chip is manufactured using the modern 4N FinFET process at TSMC and is specified with a TGP (Total Graphics Power) of 575 Watt (175W higher than RTX 4090). Nvidia now also specifies maximum GPU temperature, which is stated at 90 ℃ for RTX 5090. The model also requires at least 1000 W PSU.

The Nvidia GeForce RTX 5060 Ti is a mid-range desktop graphics card utilizing the GB206 chip based on the Blackwell architecture. The 5060 Ti offers 16 GB GDDR7 graphics memory with a 128-bit memory bus. Notoriously, the model has more VRAM compared to the higher-end RTX 5070, although with a lesser bus width of 128 bits. 

Thanks to the bump in dedicated RT/CUDA cores and new architecture, the card is expected to offer better performance compared to the previous RTX 4060 Ti. The RTX 5060 Ti is expected to be among the faster graphics cards in 2025 and, presumably will deliver good performance at elevated resolution.

According to rumored specs, the GB206 chip is manufactured in the modern 4N FinFET process at TSMC and specified with a TGP (Total Graphics Power) of 200 Watts (40 W higher than RTX 4060 Ti 16 GB).