IMST invented a proprietary software acceleration technique for FDTD simulation on modern CPU‘s.

Numerical techniques:

  • Specific C- & assembler-code created for each simulation to fit CPU architecture and simulation model
  • Individual code adaptation for latest CPU’s (AVX, AVX2, AVX 512, …)
  • Efficient caching & compression of FDTD coefficients
  • Speed not limited by RAM access time due to efficient last level cache usage (multiple time step principle)
  • The XPU technique calculates the updates for the E- and H-fields combining multiple time steps of the fields calculated in the cache memory of every core of the CPU
  • IMST invented software acceleration technique for FDTD simulation
    on modern CPU‘s

 

Key Advantages:

  • Ability to access the complete RAM of PC (> 2 TB today)
  • High simulation speed for all simulation model sizes
  • Reduced memory usage compared to standard FDTD
  • Multiple PC’s using the XPU technique can be efficiently clustered using a similar multi time step principle applied between different PC’s
  • Support of cloud computing on Amazon EC2 and Microsoft Azure cluster

XPU Technology vs GPU acceleration hardware resources usage

XPU technology surpasses the simulation speed and simulation size of GPU cards for FDTD simulations