ASUS demos 48V GeForce RTX 5090 set to 1000W
Yes, one cable.
ASUS is not only showing a generic 48V power architecture. The demo system uses a modified GeForce RTX 5090, identified in GPU Tweak III as ASTRAL-RTX5090. ComputerBase says ASUS raised the card from the usual 600W, or 800W on the ROG Matrix RTX 5090, to 1000W at 48V. The GPU Tweak III screenshot shows 48.1V input voltage and 1030.3W total power consumption.
The demo still uses a 12V-2×6-style connector , but the important part is that it is no longer carrying 12V. At 48V, the current is much lower. ComputerBase calculates that 1000W at 48V needs about 20.8A total , or around 3.5A per wire across six wires. The photo of GPU Tweak III shows per-pin readings between 3.56A and 3.71A , which lines up with that calculation.
Source: ComputeBase
For comparison, ComputerBase says a 600W RTX 5090 at 12V would need 50A total, or 8.3A per wire. At 48V, the same 600W load would need only 12.5A total, or 2.1A per wire. This is why ASUS can run a 1000W card through the same physical connector in the demo without stressing the cable in the same way. The photos show a liquid-cooled RTX 5090 test system, not a normal retail graphics card.
ROG 48V GPU Power Architecture
This ROG concept demo delivers next-level power and protection for extreme PC enthusiasts, featuring a custom PSU and 12V-2×6 cable capable of supplying up to 1200W through a single connection. Designed for the growing power demands of AI-era GPUs, the 48V architecture reduces transmission current, minimizes power loss and heat generation, and improves overall power efficiency and reliability. Integrated 48V / 12V intelligent detection logic enables automatic power rail switching for compatibility with both next-generation 48V and traditional 12V systems.
Source: ComputeBase
ComputerBase also notes why PCs still use 12V. Existing components are designed around it, GPU voltage regulation is simpler, and 12V is safer for users. So they call the ASUS setup a thought experiment , not a product launch. However, ASUS does mention in the official press release that this is something they are exploring for next-gen AI-era GPUs, whatever that means.
Source: ASUS
