Raymond He

Raymond He

Computational Benchmarks

You can find the time-to-compute of three different CPUs for two computationally intensive tasks:

  1. Reading, and cleaning 615 GB (compressed) of data,1 and
  2. Value function iteration solving the Arellano (2008) model on a fine grid.2

For full transparency, I have included all the relevant specificifications of each platform at the bottom of the page. All code was written on Julia. I am not sponsored by or affiliated with any company in any way.

1. Data processing

image info

2. Value function iteration

image info

Computer Specifications

AMD Ryzen 9 9950X (16 cores, 32 threads - 30 threads used for parallelization)

  • CPU Cooler: Arctic Liquid Freezer III 360 mm
  • Motherboard: Gigabyte X870E Aorus Master
  • GPU: Nvidia GeForce RTX 4080 Super (Founder’s Edition)
  • RAM: 64 GB of G.Skill DDR5 (2 x 32GB) 6000MT/s CL30
  • SSD: 2 TB Samsung 990 Pro Internal SSD PCIe Gen 4x4 NVMe

M4 Max Studio (16 cores - 14 threads used for parallelization)

  • 40-core GPU, 16-core neural engine
  • 64 GB unified memory

M2 Macbook Air 13 inch (8 cores - 6 threads used for parallelization)

  • 8-core GPU, 16-core neural engine
  • 16 GB unified memory

  1. Parallelization was performed across individual files. ↩

  2. The mesh consisted of 101 points for endowment and 951 points for debt. Parallelization was performed across the endowment grid while evaluating the Bellman operator. ↩