diff --git a/comfy/cli_args.py b/comfy/cli_args.py
index dbaadf723..5cb2794c8 100644
--- a/comfy/cli_args.py
+++ b/comfy/cli_args.py
@@ -75,6 +75,7 @@ fpvae_group.add_argument("--fp32-vae", action="store_true", help="Run the VAE in
 fpvae_group.add_argument("--bf16-vae", action="store_true", help="Run the VAE in bf16.")
 
 parser.add_argument("--cpu-vae", action="store_true", help="Run the VAE on the CPU.")
+parser.add_argument("--cpu-model-sampling", action="store_true", help="Run the model sampling on the CPU.")
 
 fpte_group = parser.add_mutually_exclusive_group()
 fpte_group.add_argument("--fp8_e4m3fn-text-enc", action="store_true", help="Store text encoder weights in fp8 (e4m3fn variant).")
diff --git a/comfy/model_management.py b/comfy/model_management.py
index 0eebf1ded..e3b682d16 100644
--- a/comfy/model_management.py
+++ b/comfy/model_management.py
@@ -1064,6 +1064,11 @@ def vae_device():
         return torch.device("cpu")
     return get_torch_device()
 
+def model_sampling_device():
+    if args.cpu_model_sampling:
+        return torch.device("cpu")
+    return get_torch_device()
+
 def vae_offload_device():
     if args.gpu_only:
         return get_torch_device()
diff --git a/comfy/sd.py b/comfy/sd.py
index f331feefb..a12bcb022 100644
--- a/comfy/sd.py
+++ b/comfy/sd.py
@@ -1715,6 +1715,10 @@ def load_state_dict_guess_config(sd, output_vae=True, output_clip=True, output_c
         if inital_load_device != torch.device("cpu"):
             logging.info("loaded diffusion model directly to GPU")
             model_management.load_models_gpu([model_patcher], force_full_load=True)
+            #damcclos: move the model_sampling back to the CPU. The work needed for this is not worth the gpu.
+            model_sampling_device = model_management.model_sampling_device()
+            if model_sampling_device == torch.device("cpu"):
+                model_patcher.model.model_sampling.to(model_sampling_device)
 
     return (model_patcher, clip, vae, clipvision)