Given the popularity of UVM (see link ) I think you could make a compelling argument that a UVM testbench is a ‘normal’ testbench
My guess is that in this context ‘normal’ testbench refers to something is using a simpler directed tests and not using an industry standard methodology such as UVM
…compelling argument that a UVM testbench is a ‘normal’ testbench
UVM is gaining popularity, but I would not qualify it as “normal”.
UVM is a transaction-based, class-based, and constrained-randomization of stimulus style of testbench using a library of classes to facilitate some automation (with the cycling through the various phases) and reuse. However, for over 20 years, transaction-based style of architectures were used using Verilog and VHDL, obviously without classes or libraries. In fact, I described those styles in my VHDL and Verilog books. Some engineers, but most likely very few, still used directed tests. Thus, if I were to define a style for “normal”, I would rather say “transaction-based style”, which would then encompass UVM, VMM (still in use), and also the use of Verilog (prior to being SV) and VHDL.