13

On my retail ROG Astral RTX 5080 OC, four single-stream Ollama decode profiles stayed between 265.2 and 343.6 W board power, but efficiency varied nearly 4×:

  • Llama 3.2 3B Q4_K_M: 318.3 tok/s at 265.2 W = 1.20 tok/J
  • gpt-oss 20B MXFP4: 194.4 tok/s at 292.4 W = 0.67 tok/J
  • Qwen 2.5 7B Q4_K_M: 182.2 tok/s at 343.6 W = 0.53 tok/J
  • Qwen 2.5 14B Q4_K_M: 98.8 tok/s at 311.6 W = 0.32 tok/J

Method: 1 Hz nvidia-smi telemetry, loaded samples only (≥50% utilization), with three fresh-prompt iterations per model. These are board-power results from one card and one Windows setup, not whole-system draw or universal RTX 5080 figures.

The same setup measured 52–71 W at the desktop with multi-monitor output holding P0. In a separate model using 18 idle hours/day, ~60 W GPU idle plus a 50 W rest-of-system allowance contributes about 723 kWh/year. Raw CSVs, assumptions, and full method: https://techfuelhq.com/articles/rtx-5080-perf-per-watt-ai-24-7-2026/

What idle draw and P-state does your RTX 5080 settle at once the monitors sleep?

no comments (yet)
sorted by: hot top new old
there doesn't seem to be anything here
this post was submitted on 29 Oct 2026
13 points (93.3% liked)

LocalLLaMA

5571 readers
98 users here now

Welcome to LocalLLaMA! Here we discuss running and developing machine learning models at home. Lets explore cutting edge open source neural network technology together.

Get support from the community! Ask questions, share prompts, discuss benchmarks, get hyped at the latest and greatest model releases! Enjoy talking about our awesome hobby.

As ambassadors of the self-hosting machine learning community, we strive to support each other and share our enthusiasm in a positive constructive way.

Rules:

Rule 1 - No harassment or personal character attacks of community members. I.E no namecalling, no generalizing entire groups of people that make up our community, no baseless personal insults.

Rule 2 - No comparing artificial intelligence/machine learning models to cryptocurrency. I.E no comparing the usefulness of models to that of NFTs, no comparing the resource usage required to train a model is anything close to maintaining a blockchain/ mining for crypto, no implying its just a fad/bubble that will leave people with nothing of value when it burst.

Rule 3 - No comparing artificial intelligence/machine learning to simple text prediction algorithms. I.E statements such as "llms are basically just simple text predictions like what your phone keyboard autocorrect uses, and they're still using the same algorithms since <over 10 years ago>.

Rule 4 - No implying that models are devoid of purpose or potential for enriching peoples lives.

founded 3 years ago
MODERATORS