Expanding a GPU Rental Fleet: Lessons Learned
The author expanded their GPU rental fleet from 1 to 6 cards, sharing the technical and financial challenges they faced. They discuss the hardware and software setup, earnings comparison, and unexpected insights into the AI inference market.
Why it matters
This article provides practical insights into the technical and financial realities of scaling a GPU rental business, which is relevant for AI developers and researchers looking to leverage on-demand GPU resources.
Key Points
- 1Scaling a GPU rental fleet requires careful planning for PCIe slots, power, and cooling
- 2Earnings do not scale linearly due to unpredictable demand and pricing factors
- 3The author's net passive income is around $220-$280 per month after electricity costs
- 4Key lessons include starting with a proper open-frame rig and testing each card individually
Details
The author had several older GPUs (2 RTX 3070s, 1 RTX 3080, 2 RTX 3060s) sitting unused, so they decided to set up a GPU rental fleet on Vast.ai. This required building an open-air mining frame, using PCIe risers, and daisy-chaining two power supplies to handle the 1,200-1,500W power draw of 6 cards under load. The software setup also had its challenges, with driver conflicts and issues getting all 6 cards detected properly. After the initial setup, the author saw their weekly earnings increase from around $12-18 with 1 RTX 3060 to $65-95 with the full 6-card fleet. However, this was not a linear scaling, as demand and pricing factors played a significant role. The author is now netting around $220-280 per month in passive income after accounting for electricity costs.
No comments yet
Be the first to comment