In 2024 I quit my FAANG job to become an independent researcher. To do this I needed GPUs, so I built “grumbl”, a 6x 6000 Ada GPU server. This blog describes the build, some of the issu…
To calculate money saved, the first step is to use the rental price for each day, and multiply that by the number of GPU hours used for that day, and add it all up. I didn’t have historical provider API logs, so I estimated historical pricing from timestamped references online.
Based on the Wattage records that I had logged, I calculated the electricity cost to be ~$3000, or about $125 per month.
Putting this all together, as of 3/13/26, I calculated rental fees for equivalent compute would have cost $68000 so I saved a total of $17000 so far.
Now the GPUs have paid for themselves, and based on current market rate I’m saving $90-$105 every day after this.
Final Calculation
To calculate money saved, the first step is to use the rental price for each day, and multiply that by the number of GPU hours used for that day, and add it all up. I didn’t have historical provider API logs, so I estimated historical pricing from timestamped references online.
Based on the Wattage records that I had logged, I calculated the electricity cost to be ~$3000, or about $125 per month.
Putting this all together, as of 3/13/26, I calculated rental fees for equivalent compute would have cost $68000 so I saved a total of $17000 so far.
Now the GPUs have paid for themselves, and based on current market rate I’m saving $90-$105 every day after this.
I was just kidding (sorta) that someone would spend almost 50k on graphics cards…
Eh it looks like I worked out https://rosmine.ai/2026/05/18/fixing-llm-writing-with-distribution-fine-tuning/