LM Studio user probes rig limits

Local LLM sizing is mostly a VRAM, quantization, and context-length problem wearing a hardware costume. LM Studio’s newer model-picker and RAM/VRAM estimates are exactly the kind of tooling this user is looking for.

• –In practice, the Linux box will usually be the cleaner inference host, but the 10GB GPU ceiling still matters more than the 64GB system RAM.
• –The app’s own docs point users toward model variants and quantization choices, which is the right place to start instead of hunting for a universal “best rig” chart.
• –The two machines cannot pool memory for one model in the simple case, so each workstation needs to be benchmarked on its own.
• –For cognition or research work, the best move is a small ladder of models and quantizations, then compare throughput, quality, and context headroom directly.