DATA-ORIENTED DESIGN The hardware will thank you.
If you wish to submit an article, please contact email@example.com for details.
Stream processing can be described as having your transforms restricted to an input stream, output stream, and a kernel. This way of transforming data is well proven as a high throughput technique. It is very high latency, but for graphics transforms the latency is unimportant compared to throughput, so it's a massive win.
But, there are times when stream processing latency becomes a problem, such as when you know there is only one element in the pipe.
When data oriented design talks about knowing the numbers, the values, and the shape of your data, it often suggests you turn your serial code that does one at a time processing into the stream processing model. But the important point is that it is only often, not always, so you still have to look at your data.
The fastest stream processing engine in the world, is still going to be slower if there is only ever one element in the pipe. The cutoff is not even two items, it's some number that relies on a combination of different things. Always profile before you make the switch, and then after, unless the numbers are massive, or minuscule.