Floating point is already incredibly imprecise. The thought of debugging any piece of software on this system sends shivers down my spine.<p>I would much rather see something like UNUMS (<a href="https://news.ycombinator.com/item?id=9943589" rel="nofollow">https://news.ycombinator.com/item?id=9943589</a>), which provide improved accuracy AND energy efficiency over IEEE floating point.