The electrical requirements of NVIDIA's next-generation GPUs, RTX 5090 is 575W, and RTX 5080 is 360W? - A maximum of 27% increase in power compared to the previous generation.

Detailed information on the next-generation GeForce RTX series has been revealed by multiple reliable sources. The flagship model, RTX 5090, is expected to have a Thermal Design Power (TDP) of 575W, while the RTX 5080 is expected to be 360W, confirming a significant increase in power requirements compared to the previous generation RTX 40 series. NVIDIA is scheduled to make an official announcement at the CES keynote on January 6.

The power requirements of the next-generation flagship GPU have been revealed by the industry's most reliable leakers, kopite7kimi and hongxing2020. The TDP value of 575W for the RTX 5090 represents a 125W increase from the previous generation RTX 4090. This increase carries significant implications beyond just the numerical value, highlighting the challenge of power efficiency in GPU performance improvements.

What is noteworthy is the detailed distribution of Electrical Utilities. In the case of the RTX 5090, the GB202-300-A1 GPU Die itself consumes 575W, and an additional 25W is allocated for the GDDR7 memory and other board components. This power distribution approaches the supply capacity of the 600W 12VHPWR power connector. In actual operation, the additional 75W supplied from the motherboard's PCI Express slot brings the theoretical maximum power supply to 675W.
On the other hand, for the RTX 5080, the GB203-400-A1 chip consumes 360W, requiring an additional 40W for the GDDR7 memory and peripheral components. Interestingly, the memory-related power requirements of RTX 5080 are higher than those of RTX 5090. This is attributed to the fact that the GDDR7 memory of RTX 5080 operates at 30Gbps compared to 28Gbps for RTX 5090.

The RTX 5000 series based on the Blackwell architecture is manufactured on TSMC's 4NP process. This process is an improvement over the 4N process used in the previous generation Ada Lovelace. While the 4NP process has achieved around 30% improvement in transistor density compared to 4N, it is not considered a complete generational shift.
The heart of the flagship model RTX 5090, the GB202-300-A1 die, is equipped with 21,760 CUDA cores, which are distributed among 170 streaming multiprocessors (SMs). The memory configuration adopts 32GB of GDDR7 connected through a 512-bit memory interface.

On the other hand, the RTX 5080 adopts the GB203-400-A1 chip and implements 10,752 CUDA cores (84 SMs). It is configured to connect 16GB of GDDR7 memory with a 256-bit bus width. Noteworthy is that the memory speed of the RTX 5080 is 30Gbps, surpassing the higher model. This is due to constraints related to the characteristics of first-generation GDDR7 memory, requiring additional power for high-speed operation. NVIDIA expects that future improvements in GDDR7 memory will allow for faster operation with less power consumption.

As a result, the gradual evolution of the manufacturing process and constraints in memory technology are major factors pushing up the power requirements of next-generation GPUs. NVIDIA is in a situation where while pursuing performance improvements, they are forced to strike a delicate balance with power efficiency.
The rapid increase in power requirements clearly reflects that the semiconductor manufacturing process is entering a deceleration phase. NVIDIA is pushed into a situation where they have to invest more power in order to pursue performance improvements.
Of particular interest is the point that approaches the 600W limit of the 12VHPWR power connector. Even adding 75W from the PCI Express slot, the theoretical limit becomes 675W. In other words, for future generations of GPUs, there is a high possibility of requiring a new power supply method. For general users, a power unit of over 1000W will be essential, inevitably leading to further overall system cost increases.