In the early days of a startup, it is super critical to minimize expenses, especially on things that don’t directly generate new customers. Buying computers is one of the worst kinds of expense, because you have to pay upfront for a depreciable asset.
Five years ago, AccelerEyes faced such a predicament. As a high-performance and technical computing company, we needed to obtain many computers for unit testing our products across different hardware components, operating systems, and bit-count (32- vs 64-bits). We ultimately evolved though the following stages:
- Stage 1: Work almost entirely on personal laptops. Manually test on just 2-4 assembled-from-component machines. Each carried multiple dual-booting hard drives and GPU-types that we manually swapped out to test various combinations. (Very labor intensive and inefficient, but low infrastructure cost)
- Stage 2: Transition most people to company laptops (~$700-$1,500), each with a GPU capable of running our demos. Reduce the manual testing load by buying more assembled-from-component machines, slowly phasing out the need to dual boot systems. (Medium labor, inefficiencies, and infrastructure costs)
- Stage 3: Laptops stay same. Implement an automated test infrastructure to eliminate the need to manually login to the test infrastructure. Jenkins is great for continuous integration testing. (Much improved efficiency, more expensive. Tradeoffs between cost and quality are no longer an issue. Rather, maximizing developer productivity and efficiency is a much higher priority.)
- Stage 4: Laptops stay the same. More machines, including beefy servers, are added to Jenkins, but otherwise stays the same. Also, incorporation of high compute (including GPUs) instances from AWS, Penguin, Peer 1, or others can be useful. (Highly efficient. Now simply adapting to company growth and new hardware vendor components. Predictably stable trajectory.)
Along the way, we’ve received generous donations of computers and components from our hardware partners (e.g. AMD, NVIDIA, and Intel) which has been quite beneficial to our development processes. These donations have in turn had a reciprocal effect of enabling us to better fine tune our software libraries for those platforms.
While most startups don’t need as much processor iron as AccelerEyes, there are certainly ways that many startups can reduce the costs of development within their organization. With today’s commodity computer components, building desktops and servers from scratch is really simple and provides the lowest cost and highest controlled option. For people in Atlanta, GIM Computers is a nice place to go next to Georgia Tech. Micro center is also good.
What computers did you buy at the start of your business? Did you overpay or underpay on infrastructure expenses?
Related articles
Comments on this entry are closed.