As a computational physicist working in materials/condensed matter, I'm either highly biased or well-placed to comment on this.
Physics, in practice, is divided into three overlapping approaches: experimental, theoretical, and computational. (The highest impact research papers usually include a combined effort from all three.)
If you plan to go into computational research then you will have to do a fair amount of programming. However, I don't know anyone who has made use of Raspberry Pi's for physics research (that's not to say that no one has, but it's a novelty rather than something that is commonly done). In computational physics, your code will almost exclusively be executed either on standard desktop machines or supercomputers (where you use message-passing systems like MPI to exploit huge parallelism). Virtually all universities have their own supercomputers, but you may also be granted access to some larger national or even international supercomputers (such as ARCHER, Jaguar, and so on).
Graphics cards have also become quite popular for physics research in recent years due to the rise of CUDA, and most supercomputers now include several nodes packed with high-end graphics cards. So GPGPU programming is a nice skill to have but by no means a necessity.
It's also worth mentioning programming languages. Mainly for historical reasons, most academic code is actually written procedurally in Fortran (which is so archaic it still has functionality left over from the punch-card era). C/C++, Java, and Python are also widely used, along with the Unix shell (most academic machines run Linux). Those who do a lot of statistical modelling mostly use R or IDL. And those who are too lazy to do real programming - mostly mathematicians and engineers - use MATLAB or Mathematica (okay, I'm being a bit harsh on that one).
Let me finish by discussing theoretical and experimental physics. Virtually every theorist I know does much of their work on computers - programming code to numerically solve, or test something, for instance. And many of their 'theories' are aimed at advancing computational methodologies. A classic example of this is the Hohenberg-Kohn theorems which laid the foundation for density functional theory, and there are now many theorists trying to extend this by developing linear-scaling and real-space DFT.
It has also become common for experimentalists to program. Whether that be microcontrollers like Arduinos (as pointed out by Emilio Pisantry below), scripts to analyse data, or even employ standard simulation techniques to better understand their experimental observations.