Arsen Apostolov writes about the actual electrical cost of running local Large Language Models on a single NVIDIA RTX 3090 compared to hosted cloud APIs.
>"I measured the actual GPU electricity for eight local models on one RTX 3090 — and the cheapest wasn't the smallest, nor the priciest the biggest"
Cost of Generating 1 Million Tokens Locally
| MODEL | PARAMS (Billions) | MEAN SPEED (tok/s) | AVG GPU DRAW (W) | € / 1M OUTPUT TOKENS |
| :--- | :---: | :---: | :---: | :---: |
| **gemma3:1b** | 1B | 136 tok/s | 154 W | €0.060 |
| **Qwen3-Coder** | 30.5B | 130 tok/s | 233 W | €0.112 |
| **gemma4:26b** | 26B | 85 tok/s | 246 W | €0.139 |
| **Devstral** | 24B | 49 tok/s | 320 W | €0.321 |
| **gemma3:27b** | 27B | 36 tok/s | 283 W | €0.361 |
| **Seed-OSS** | 36B | 4.5 tok/s | 186 W | €0.946 |
| **GLM-4.5-Air** | 106B | 5.7 tok/s | 141 W | €1.040 |
| **DeepSeek-R1-Distill** | 32.8B | 6.9 tok/s | 155 W | €1.526 |
By measuring real-time GPU power consumption through a custom dashboard, he discovered that token costs are driven by effective wall-clock throughput rather than model parameter size or raw generation speed alone. The results show that while small and fast models can be more economical than cloud services, reasoning-heavy models may actually become the most expensive to run locally due to the time spent "deliberating" between tokens.
* Measurements were performed using HomeLab Monitor, an open-source dashboard that integrates live power data from `nvidia-smi`.
* DeepSeek-R1-Distill emerged as the most expensive model per million tokens because its effective throughput is slowed by reasoning delays.
* The findings focus on marginal electricity costs and exclude total cost of ownership factors like hardware amortization or idle draw.
>"I measured the actual GPU electricity for eight local models on one RTX 3090 — and the cheapest wasn't the smallest, nor the priciest the biggest"
Cost of Generating 1 Million Tokens Locally
| MODEL | PARAMS (Billions) | MEAN SPEED (tok/s) | AVG GPU DRAW (W) | € / 1M OUTPUT TOKENS |
| :--- | :---: | :---: | :---: | :---: |
| **gemma3:1b** | 1B | 136 tok/s | 154 W | €0.060 |
| **Qwen3-Coder** | 30.5B | 130 tok/s | 233 W | €0.112 |
| **gemma4:26b** | 26B | 85 tok/s | 246 W | €0.139 |
| **Devstral** | 24B | 49 tok/s | 320 W | €0.321 |
| **gemma3:27b** | 27B | 36 tok/s | 283 W | €0.361 |
| **Seed-OSS** | 36B | 4.5 tok/s | 186 W | €0.946 |
| **GLM-4.5-Air** | 106B | 5.7 tok/s | 141 W | €1.040 |
| **DeepSeek-R1-Distill** | 32.8B | 6.9 tok/s | 155 W | €1.526 |
By measuring real-time GPU power consumption through a custom dashboard, he discovered that token costs are driven by effective wall-clock throughput rather than model parameter size or raw generation speed alone. The results show that while small and fast models can be more economical than cloud services, reasoning-heavy models may actually become the most expensive to run locally due to the time spent "deliberating" between tokens.
* Measurements were performed using HomeLab Monitor, an open-source dashboard that integrates live power data from `nvidia-smi`.
* DeepSeek-R1-Distill emerged as the most expensive model per million tokens because its effective throughput is slowed by reasoning delays.
* The findings focus on marginal electricity costs and exclude total cost of ownership factors like hardware amortization or idle draw.
>"I Measured Every Watt on Apple Silicon Five models, sustained generation, real wall-socket energy at $0.31/kWh — and the surprise the RTX-3090 numbers predicted, only bigger."
Justin Stewart writes about how the energy cost of running local Large Language Models (LLMs) on Apple Silicon depends more on throughput than parameter count. Using an M3 Ultra Mac Studio, he demonstrates that large Mixture-of-Experts (MoE) models can be significantly cheaper to operate per token than smaller dense models because they only activate a fraction of their parameters during generation. Ultimately, the study reveals that efficiency is driven by how much data must be moved from memory for every token produced.
* The measurements were calibrated against actual wall power using a Shelly Plug US Gen4 meter.
* A custom tool called TokenWatt was used to measure marginal energy consumption via Apple’s IOReport interface.
* In real-world "lumpy" traffic scenarios, the cost of dense models compared to MoE models actually widens even further.
Justin Stewart writes about how the energy cost of running local Large Language Models (LLMs) on Apple Silicon depends more on throughput than parameter count. Using an M3 Ultra Mac Studio, he demonstrates that large Mixture-of-Experts (MoE) models can be significantly cheaper to operate per token than smaller dense models because they only activate a fraction of their parameters during generation. Ultimately, the study reveals that efficiency is driven by how much data must be moved from memory for every token produced.
* The measurements were calibrated against actual wall power using a Shelly Plug US Gen4 meter.
* A custom tool called TokenWatt was used to measure marginal energy consumption via Apple’s IOReport interface.
* In real-world "lumpy" traffic scenarios, the cost of dense models compared to MoE models actually widens even further.
SPAN has announced that its four newest smart panel models are the first to receive UL 3141 certification. This safety standard for Power Control Systems ensures devices can manage electrical loads effectively while protecting consumer safety and device integrity. The certification supports increased home electrification by utilizing existing utility infrastructure, potentially helping homeowners avoid expensive service upgrades.
Abstract:
>tained during most months, resulting in the generation of >400 milliwatts per square meter of mechanical power with a potential for >6 watts per square meter. We further apply this technique for air circulation, achieving >0.3 meters per second with a potential volumetric flow rate that exceeds 5 cubic feet per minute (cfm), which is sufficient for CO2 circulation in greenhouses and for thermal comfort inside residential buildings.
>tained during most months, resulting in the generation of >400 milliwatts per square meter of mechanical power with a potential for >6 watts per square meter. We further apply this technique for air circulation, achieving >0.3 meters per second with a potential volumetric flow rate that exceeds 5 cubic feet per minute (cfm), which is sufficient for CO2 circulation in greenhouses and for thermal comfort inside residential buildings.
Researchers discovered that renewable energy facilities across Central Europe use unencrypted radio signals to control power generation, posing a potential threat to the grid. If intercepted and manipulated, these signals could disrupt grid stability by causing power imbalances, possibly leading to a continent-wide blackout. This raises significant concerns about the security measures currently in place and the need for more secure alternatives like iMSys, which uses encrypted LTE for communication.
- Evolution is seen as a highly path-dependent process due to its historical nature, but outcomes could have varied.
- Convergence and constraints significantly limit evolutionary designs, suggesting that not all possibilities are realized.
- Fundamental constraints are inherent in the logic of living matter, influencing evolutionary outcomes.
- Examples of constraints include thermodynamics in living systems, linear molecular information, cellular composition, multicellularity, cognitive system computations, and ecosystem architecture.
- The study provides evidence for these constraints and proposes pathways for a defined theoretical framework.
- Convergence and constraints significantly limit evolutionary designs, suggesting that not all possibilities are realized.
- Fundamental constraints are inherent in the logic of living matter, influencing evolutionary outcomes.
- Examples of constraints include thermodynamics in living systems, linear molecular information, cellular composition, multicellularity, cognitive system computations, and ecosystem architecture.
- The study provides evidence for these constraints and proposes pathways for a defined theoretical framework.
Physicists and computer scientists are using stochastic thermodynamics to understand the energy costs of computation, with implications for designing more energy-efficient devices.
Using Digital Twins to optimize data center operations and eliminate wasted IT infrastructure can save significant costs and improve sustainability.
Helion, backed by OpenAI, claims to be on track to build its first fusion plant within the next five years, but experts are skeptical of the timeline.
The company's Polaris reactor design uses an electromagnetic coil system to generate a 50-megawatt electrical output, with a planned location in Washington state, USA.
The company's Polaris reactor design uses an electromagnetic coil system to generate a 50-megawatt electrical output, with a planned location in Washington state, USA.
This article details how to use a software-defined radio (SDR) to read data from utility meters, allowing for a more complete understanding of energy usage in a smart home.
