Monthly deep learning workstation builds

June 10, 2025

A future post will detail the multitude of considerations to think about when selecting parts for deep learning work.

Disclaimer: These hardware configurations have not been tested. Although I have done ample research on the components and therefore I expect them to work well together, there will likely be variables that I did or could not have taken into consideration. I do not make any guarantees about the selections below. I simply hope that these lists offer a strong starting point for you to do your own research and part selection.

Monthly part selections

June 2025

ComponentSelectionSKULink#$/UnitTotal
CPU

AMD EPYC 7502 (32C/64T)

n/a

Ebay

1

$170

$170

Mobo

Supermicro H12SSL-NT

n/a

Ebay

1

$670

$670

RAM

Crucial Bundle 64GB (2 x 32GB) DDR4-2666MHz RDIMM

CT32G4RFD4266

Amazon

2

$90

$180

PSU

EVGA Supernova 1600 G+

220-GP-1600-X1

Amazon

1

$350

$350

SSD

Micron 7450 PRO 960 GB

MTFDKBA960TFR-1BC1ZABYY

Amazon

1

$230

$230

CPU cooler

Noctua NH-U14S TR4-SP3

n/a

Amazon

1

$110

$110

Case

Fractal Design Define 7 XL

n/a

Amazon

1

$225

$225

Total (without GPUs)$1935

Key features and rationale

  • Air-cooled desktop workstation
  • 2nd generation AMD EPYC CPU (32 cores, 64 threads) and Supermicro server motherboard
    • While the 7002 series of CPUs was released in 2019 with a relatively low clock speed of 2.5 GHz, the huge number of threads make the CPU perfect for data processing workloads.
  • Enterprise SSD for high reliability (>10000 TBW lifetime)
  • Enough physical slots, space, power, and PCIe slots for 4 dual-slot GPUs
    • The large number of PCIe lanes connected directly to the CPU (and physically supported by the motherboard) supports using 4 GPUs at their full capabilities.
    • Make sure that your GPU power requirement in the system stays below 1200 W total to avoid overdraw on the PSU and/or tripping your circuit breaker.

TODO and additional notes

  • Blower-style GPUs are recommended for optimized case airflow.
  • Add more case fans to ensure effective airflow.