EM simulation is time consuming, so we want to use the fastest available (affordable) CPUs.
This speed benchmark compares
- Intel Core i9-7940X (14 cores, 3.1GHz/4.3 GHz, 19MB cache, released Q3’17), 4 modules DDR4-2400
- AMD Ryzen 9 5950X (16 cores, 3.4GHz/4.9 GHz, 64MB cache, released Q4’20), 4 modules DDR4-3200
Compared to the 3 year old Intel Core i9, the AMD data does not look much different, but cache is much larger and RAM faster.
Empire XPU
The Empire XPU solver is very sensitive to CPU architecture, cache size and memory speed. It compiles a solver on-the-fly for the actual CPU used, and supports the latest CPU architecture enhancements for floating point processing.
Difference in simulation speed is amazing. We have a factor 2 in speed for most testcases, and 2.8x faster simulation for one large RFIC model.
The Intel system is running Windows 10 Pro, the AMD system is running Ubuntu 20.04 LTS.
Summary
The new AMD Ryzen 9 5950X looks very promising to speed up EM simulation with desktop computer hardware. It really shines when used with Empire XPU and its “solver-on-fly” concept in Empire XPU that takes advantage of all the latest processor-specific features.
Cost of the complete machines was only 2000€ (Intel 7940X in 2018) and 1600€ (AMD 5950X in 2021).