diff --git a/comfy/model_management.py b/comfy/model_management.py
index bcf1399c4..4b78e27cb 100644
--- a/comfy/model_management.py
+++ b/comfy/model_management.py
@@ -1446,7 +1446,7 @@ def pytorch_attention_enabled():
     return ENABLE_PYTORCH_ATTENTION
 
 def pytorch_attention_enabled_vae():
-    if is_amd():
+    if is_amd() and not SUPPORT_FP8_OPS:  # exclude RDNA4 (gfx1200, gfx1201) and CDNA4 (gfx950) that support fp8
         return False  # enabling pytorch attention on AMD currently causes crash when doing high res
     return pytorch_attention_enabled()
 
diff --git a/comfy/sd.py b/comfy/sd.py
index e573804a5..96c41ee8d 100644
--- a/comfy/sd.py
+++ b/comfy/sd.py
@@ -440,7 +440,7 @@ class VAE:
         if 'decoder.up_blocks.0.resnets.0.norm1.weight' in sd.keys(): #diffusers format
             sd = diffusers_convert.convert_vae_state_dict(sd)
 
-        if model_management.is_amd():
+        if model_management.is_amd() and not model_management.SUPPORT_FP8_OPS:  # exclude RDNA4 (gfx1200, gfx1201) and CDNA4 (gfx950) that support fp8
             VAE_KL_MEM_RATIO = 2.73
         else:
             VAE_KL_MEM_RATIO = 1.0