>>908145
The point was to get rid of all the complex scheduling silicon that tries to extract parallelism from inherently sequential code and move all that complexity to software. It was expected that this way more silicon could be dedicated to actually doing shit instead of planning to do shit, and therefore higher performance could be obtained from the same transistor budget.
Of course, like all VLIW power fantasies, this performance never materialised. Optimal scheduling depends on too many dynamic variables to be achieved ahead of time. However, Transmeta found out that their shit-performing failed abortion of a performance chip sipped far less power than competing processors. So they quickly regrouped and started marketing it as a mobile part. The rest is history.