Last week, Simon Pickles from the Foundry posted an online survey to gather artist’s views of how the cloud could help them with their pipeline.
He likened the power of the cloud to a tap: you just turn it on and out pours the water. When you don’t need it any more, you can turn it off. Why, he argues, should you spend on up-front expenditure when you can “burst to 1000x power for minutes and iterate like never before; take that bigger job unfettered by a fixed capacity. It’s a future where you can focus on what you do best—making great content.”
It’s a powerful vision (and one I love as it’s perfectly aligned with YellowDog), however, what about the economics of it? When you look at the total cost of ownership of purchasing cloud compute power, how does it stack up to buying your own servers, or even using the power of the crowd?
It’s time to crunch some numbers.
For comparison, I thought I’d look at the total cost of ownership of of buying 60,000 GFLOPS of compute power (that’s a lot – about 500 servers at any one time) to use for rendering over a three-year period.
The costs that need to be examined are:
- Hardware, whether rented from the cloud or purchased.
- Software licenses, which includes any fees for purchasing the software as a service.
- Infrastructure including data centre costs, storage and data transfer fees.
- People you would need in your organisation to manage this.
- Any cost of downtime.
- Hardware replacement costs in that three-year period.
The costs below are illustrative and could be refined. Despite that I hope it shows that there is an economic benefit to cloud- and crowdcompute as well as the flexibility that Simon Pickles described.