2026-11-17 · America/Los_Angeles · 社区动态 · #15
older gpus abandoned by new inference engines. are they still worth it?
、i keep seeing the same problem with local setups. newer inference engines keep raising their minimum hardware limits. older cards fall off the support list. they used to work perfectly fine for small models. plenty of labs have healthy older gpus sitting on shelves. they are not physically dead. they are just hard to use with new tools because the software moved on. we have a few realistic options for these cards. we can run lighter backends like llama cpp. they still support older architectures. we can run small quantized models to fit the vram. or we can use the old cards as low priority…
热度 41.7 / 100;排名与评分保留该期记录。