I have been testing the new CU7 AXBuild compile feature. It seems to work quite well. Comparing the compile time of the AXBuild to the standard AX compile time is as follows:
Standard system X++ compile: 196 minutes
|1 Core (min)||2 Core (min)||4 Core (min)|
Microsoft states that 1.4 workers per core is the optimal. If the worker control is not specified at runtime it will multiply the number of cores by 3 and divide that by 2. This means it is running at 1.5 times core count.
My testing above shows there is a significant improvement over the standard AX compile, even with 1 worker. As the core count increases the overall time for compile improves as the worker count increases. Looking at the numbers the improvement after 2 times the core count is minimal, suggesting running at 2 time core count is really the maximum you would want to specify.
My testing is of course subjective. It is possible as I increased the number of workers and cores, but would simply move the bottleneck to another part of the system, either network IO or SQL IO. It is difficult to fully evaluate this with the test performed, so the results might be a little skewed. I have read how people on a quad core have been able to get compile times of 15 minutes. They did state they were using Hyper-V and SSD drivers, so this might correlate fairly well as SSD drivers are a lot faster.
The end result is, the increase off cores and workers result in an improved compile time.