Existing (not optimally parallel) EDA tools can still exploit the operating system to benefit from parallelism. Examples abound, like the pattern-based DRC; in the TCAD area, computational lithography, etc.
FPGA-based accelerators enable to run chip designs at MHz speeds and to debug system-level scenarios in the lab, but they are not simulators. It is just a different product category.
- You can reach 1-10MHz speeds with them and therefore debug your driver and even your application in embedded systems
- They are very expensive.
- Require significant ramp-up time, and then if you change your code or libraries you are not really debugging your real silicon design
- Does not work alongside your existing test-bench (verification environment), and if it does you cannot reach MHz speeds.
- Limited in capacity (to scale you need to add more FPGAs/boxes but then you trade-off with speed)
- Lack support for non-synthesize-able code
- No support for 4-state logic
- Lack full visibility
- Long compilation time (require to synthesys and place-and-route)