Enthusiast says DLSS 5 pushed RTX 5090 past 600W and melted the 16-pin connector — Nvidia's neural rendering tech adds up to 50% more power draw in testing
- ID
- 22068
- Status
- summarized
- Published
- 07 Sep 2026, 6:00 PM
- Fetched
- 07 Sep 2026, 6:38 PM
- Provider
- Tom's Hardware
- Category
- technology
- Original URL
- https://www.tomshardware.com/pc-components/gpus/enthusiast-says-dlss-5-pushed-rtx-5090-past-600w-and-melted-the-16-pin-connector-nvidias-neural-rendering-tech-adds-up-to-50-percent-more-power-draw-in-testing
- Source URL
- https://www.tomshardware.com/feeds/all
Summary
- Score
- 5.5
- Created
- 07 Sep 2026, 6:39 PM
- Tags
- Audience
- developersai_ml_learnersvibe_coders
What happened
An enthusiast reports that enabling DLSS 5 on an RTX 5090 pushed power draw past 600W and melted the 16-pin power connector, with testing showing Nvidia's neural rendering technology adding up to 50% more power consumption. This echoes the earlier RTX 4090 connector melting issues, suggesting the problem persists in the new generation.
Why it matters
If you're running local AI inference or fine-tuning on RTX 5090s, DLSS 5's neural rendering layer can push the card well beyond its rated TDP and risk connector damage. Anyone provisioning high-end GPU workstations for local ML workloads should factor in the real-world power overhead of AI-enhanced rendering features and verify their PSU and cabling headroom accordingly.
Discussion angle
For builders running local models on consumer GPUs: does the 5090's real-world power behavior change the calculus for local AI infrastructure vs. cloud, especially given that AI rendering features can add 50% power overhead on top of already extreme baselines?