ComfyUI: Why --cache-classic drastically improves memory management on RTX 5090
EasternAd8821 · reddit · 2026-08-22
User reports severe performance issues with a 60GB BF16 model on an RTX 5090 (32GB VRAM, 128GB RAM) using default dynamic memory settings, causing maxed out VRAM/RAM and SSD swap slowdowns. Switching to the --cache-classic startup flag reduces VRAM usage to 25GB and system RAM to 35GB, restoring acceptable speeds (42s/step). The post questions why the legacy cache mode outperforms modern management on high-end hardware.
More from Infra
- Nvidia uses linear algebra to fix cross-model KV cache transfer, 25x faster — bendee983 · 2026-08-22
- Anthropic hires former Google TPU head Amir Salek to boost in-house chip development — nmasc_ · 2026-08-22
- Lambda runs All-Reduce across 10,368 NVIDIA GB300 GPUs — TheZachMueller · 2026-08-22
- Running 284B DeepSeek on 64GB MacBook: Lossless quality confirmed — cowboy-bebob · 2026-08-22
- Dev Debate: Apple Silicon Remains Best for On-Device LLMs — walkingriver · 2026-08-22
- Nvidia Earnings Preview: Rubin Architecture and Competitive Moat Analysis — ryanshrout · 2026-08-22