3.1 KiB
| title |
|---|
| Cloud vs. Self-hosting Your AI |
The choice of how to run your AI - on GPU cloud services, on-prem, or just using an API provider - involves various trade-offs. The following is a naive exploration of the pros and cons of renting vs self-hosting.
Cost Comparison
The following estimations use these general assumptions:
| Self-Hosted | GPT 4.0 | GPU Rental | |
|---|---|---|---|
| Unit Costs | $10k upfront for 2x4090s (5 year amort.) | $0.00012/token | $4.42 for 1xH100/h |
- 800 average tokens (input & output) in a single request
- Inference speed is at 24 tokens per second
Low Usage
When operating at low capacity:
| Self-Hosted | GPT 4.0 | GPU Rental | |
|---|---|---|---|
| Cost per Request | $2.33 | $0.10 | $0.04 |
High Usage
When operating at high capacity, i.e. 24 hours in a day, ~77.8k requests per month:
| Self-Hosted | GPT 4.0 | GPU Rental | |
|---|---|---|---|
| Cost per Month | $166 (fixed) | $7465 | $3182 |
Incremental Costs
Large context use cases are also interesting to evaluate. For example, if you had to write a 500 word essay summarizing Tolstoy's "War and Peace":
| Self-Hosted | GPT 4.0 | GPU Rental | |
|---|---|---|---|
| Cost of "War and Peace" | (upfront fixed cost) | $94 | $40 |
Takeaway: Renting on cloud or using an API is great for initially scaling. However, it can quickly become expensive when dealing with large datasets and context windows. For predictable costs, self-hosting is an attractive option.
Business Considerations
Other business level considerations may include:
| Self-Hosted | GPT 4.0 | GPU Rental | |
|---|---|---|---|
| Data Privacy | ✅ | ❌ | ❌ |
| Offline Mode | ✅ | ❌ | ❌ |
| Customization & Control | ✅ | ❌ | ✅ |
| Auditing | ✅ | ❌ | ✅ |
| Setup Complexity | ❌ | ✅ | ✅ |
| Setup Cost | ❌ | ✅ | ✅ |
| Maintenance | ❌ | ✅ | ❌ |
Conclusion
The decision to run LLMs in the cloud or on in-house servers is not one-size-fits-all. It depends on your business's specific needs, budget, and security considerations. Cloud-based LLMs offer scalability and cost-efficiency but come with potential security concerns, while in-house servers provide greater control, customization, and cost predictability.
In some situations, using a mix of cloud and in-house resources can be the best way to go. Businesses need to assess their needs and assets carefully to pick the right method for using LLMs in the ever-changing world of AI technology.