It took me a few days to build a high-performance computing machine. Haven’t done this for 20 years. Now the beast is ready, and I am proud it is running. Main assets: a 64 core AMD Threadripper PRO processor, and two Nvidia RTX 3090 graphic cards. This computer works like a cluster and allows me to run up to 128 calculations in parallel.
256 GB memory and 6 TB fast SSD storage complement the powerful processor and GPUs. Several operating systems are installed and managed by Proxmox. First I installed water cooling, but it featured noisy fans, so I switched to a reliable Noctua air cooler. This works: even at 100% CPU usage, the temperature does not go beyond 75 degree Celsius on the CPU and the noise from the 8 fans is tolerable.
This computer enables the usage of various software packages, and all computation for which my laptop is too slow, like docking of >10’000 molecules.
If you consider building a similar computer, the following tutorials were helpful for me:
- Ultimate-ULTIMATE 3D Rendering Workstation Build [$19000] | AMD 3995WX + ASUS 2x RTX 3090: this tutorial was very helpful – I have the same CPU, motherboard, air cooler and case
- Fractal Meshify 2 Step-by-Step PC Build Guide: this guide was fantastic for some nitty-gritty details
- Alphacool Eisbaer Aurora 360 Cooler – overview, install and performance: everything about the water cooler setup
- Proxmox 7.0 installation & basics [Simple and understandable]: For Proxmox setup (German)
GPUs for AI and MD
For AI applications and MD calculations, I have two Nvidia GPUs (ASUS GeForce TUF RTX 3090 and Palit GeForce RTX 3090 GamingPro OC). The 24 GB VRAM give me excellent performance for the bucks invested. They are powerful enough for 95% of my work and have full support for PyTorch, TensorFlow, and other AI frameworks. For the 5% of time when I need more power, I continue to rent exclusive cards in the cloud.
