So much fail about Garbage Collection.
GC is not about forgetting to free memory. It's about higher level abstraction removing the need for the programmer to do the bookkeeping that the machine can do. Why don't we still program in assembler? Because it's less productive. It's about productivity. As data structures become extremely complex, and get modified over time, keeping track of the ownership responsibility of who is supposed to dispose of what becomes difficult to impossible, and is the source of memory leak bugs. In complex enough programs, you end up re-inventing a poor GC when you could have used one that is the product of decades of research.
The article fails to understand that you can also run out of memory in a program using GC. Just keep allocating stuff without and keeping references to everything you allocate.
Reference Counting is not real GC. Cyclical data structures will never get freed using reference counting.
One of the major, but under-recognized benefits of GC, which the article fails to mention, is that GC allows much simpler ''contracts' in APIs. No longer is memory management part of the 'contract' of an API. It doesn't matter which library or function created an object, nobody needs to worry about who is responsible for disposing of that object. When nobody references the object any more, the GC can gobble it up.
On the subject of Virtual Machines, the article could mention some of the highly aggressive compilation techniques used in JIT compilers. So every method in Java is a virtual call. But a JIT compiler knows when there is only one subclass that implements a particular method and makes all calls to the method non-virtual. If another subclass is loaded (or dynamically created on the fly) the JIT can recompile all methods that call the method such that they are now virtual calls. Yet still, the JIT may be able to prove that certain calls are always and only to a specific subclass, and so they can be non-virtual.
The JIT compiler in JVM can aggressively inline small functions. But if a class gets reloaded on the fly such an the body of an inlined method changed, the JIT will know to recompile every other method that inlined the changed method. Based on the changes to the method, it may or may not now make sense to inline it -- so the decision on whether to inline the method can change based on actual need.
The HotSpot JVM dynamically profiles code and doesn't waste time and memory compiling methods that do not have any significant effect on the system's overall performance. The profiling can vary depending on factors that vary from system to system, and could not be predicted in advance when using a static compiler. The JIT compiler can compile your method using instructions that happen to exist on the current microprocessor at runtime -- something that could not be determined in advance with a static compiler.
All of this may seem very complex. But it's why big Java systems run so darn fast. Not very many languages can have tens or even hundreds of gigabytes (yes GB) of heap with GC pause times of 10 ms. Yes, it may need six times the amount of memory, but for the overall benefits of speed, the cost of memory is cheap.