NVIDIA GeForce RTX 5070 vs NVIDIA GeForce RTX 4070 Ti

The Nvidia GeForce RTX 5070 is a fast desktop graphics card utilizing the GB205 chip based on the Blackwell architecture. The model offers 6144 CUDA cores, and 12 GB GDDR7 graphics memory with a 192-bit memory bus (672 GB/s). The base clock of the GPU (reference model) is specified at 2.16 GHz, while the boost clock can reach 2.51 GHz.

Architecture

Nvidia GeForce RTX 5070 features new Tensor (5th Gen) and RT-cores (4th Gen), with 988 AI TOPS and 94 TFLOPS performance 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 managed to improve 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 and Frame Warp.

Performance

Thanks to the new Tensor and RT-cores, as well as some bump in CUDA cores, the card is promised (by Nvidia) to offer as much as 2x better performance compared to the previous RTX 4070 (thanks to the new DLSS 4 MFG feature). According to Nvidia's presentation, the card can even reach the performance level of the RTX 4090. The RTX 5070 is expected to be among the faster graphics cards of 2025 and, presumably, will deliver good performance in games at high resolution.

Our own tests show that the card is 4% faster than the RTX 4070 Ti in the synthetic benchmarks and almost identical to the Radeon RX 7900XT in games (RTX 5070 is 1% slower at 4K, 1080p, and 2K). 

Power consumption

According to Nvidia, the GB205 chip is manufactured using the modern 4N FinFET process at TSMC and is specified with a TGP (Total Graphics Power) of 250 watts (50 W higher than RTX 4070). Nvidia now also specifies maximum GPU temperature, which is stated at 85 ℃ for RTX 5070. The model also requires at least 650W PSU.