Nodes/IAMCCS-nodes/HW Supporter (auto VRAM/attention/torch knobs)
ComfyUI Node

HW Supporter (auto VRAM/attention/torch knobs)

A ComfyUI node in IAMCCS/HW with 15 inputs and 4 outputs.

By IAMCCS·Created 10 months ago·Updated about a month ago· 109
HW Supporter (auto VRAM/attention/torch knobs)
  • model
  • clip
  • vae
  • model
  • clip
  • vae
  • report_json
profileauto
apply_reserved_vramtrue
reserved_vram_modemanual
reserved_vram_gb1.25
sage_attentionauto
allow_sageattention_torch_compilefalse
torch_compile_modeoff
fp16_accumulationauto
tf32auto
clean_gpu_beforefalse
console_logtrue
include_hardware_reportfalse
CategoryIAMCCS/HW

Inputs (15)

NameTypeDefaultDescription
modelMODEL
profileCOMBOauto5 options: auto, 12GB_VRAM_32GB_RAM, low_vram, balanced, max_speed
apply_reserved_vramBOOLEANtrue
reserved_vram_modeCOMBOmanualmanual: uses reserved_vram_gb | auto_used_plus: sets EXTRA_RESERVED_VRAM = (used_vram + auto_headroom_gb), like ReservedVRAMSetter
reserved_vram_gbFLOAT1.250–24Reserved VRAM in GB (manual mode). If profile=auto and this is 0, the node chooses a conservative value.
sage_attentionCOMBOauto8 options: disabled, auto, sageattn_qk_int8_pv_fp16_cuda, sageattn_qk_int8_pv_fp16_triton, sageattn_qk_int8_pv_fp8_cuda, sageattn_qk_int8_pv_fp8_cuda++, +2
allow_sageattention_torch_compileBOOLEANfalse
torch_compile_modeCOMBOoffauto = try torch.compile with a safe mode (reduce-overhead). On Windows this may still be unstable depending on Torch/driver; if you see crashes, set off.
fp16_accumulationCOMBOauto3 options: auto, on, off
tf32COMBOauto3 options: auto, on, off
clean_gpu_beforeBOOLEANfalse
console_logBOOLEANtruePrint a short summary (applied + warnings) in the server console.
include_hardware_reportBOOLEANfalseIf enabled, probes current hardware and embeds an additional recommendations report into report_json (useful for printing/debug).
clipoptCLIP
vaeoptVAE

Outputs (4)

NameTypeDescription
modelMODEL
clipCLIP
vaeVAE
report_jsonSTRING