With so much happening on the AI/HPC hardware front lately, I wonder where the prior-generation servers end up once well-funded companies upgrade. I'm interested in acquiring a GPU-laden machine with enough VRAM for local fine tuning, training, and batch inference. I could build a workstation with current or older-gen GPUs, but it seems like a used professionally built rack-mount machine could be higher value.
With so much happening on the AI/HPC hardware front lately, I wonder where the prior-generation servers end up once well-funded companies upgrade. I'm interested in acquiring a GPU-laden machine with enough VRAM for local fine tuning, training, and batch inference. I could build a workstation with current or older-gen GPUs, but it seems like a used professionally built rack-mount machine could be higher value.