diff --git a/modules/sd_samplers_common.py b/modules/sd_samplers_common.py
index 03858cdb..065eb714 100644
--- a/modules/sd_samplers_common.py
+++ b/modules/sd_samplers_common.py
@@ -70,9 +70,12 @@ def samples_to_images_tensor(sample, approximation=None, model=None):
 def single_sample_to_image(sample, approximation=None):
     x_sample = samples_to_images_tensor(sample.unsqueeze(0), approximation)[0] * 0.5 + 0.5
 
-    x_sample = torch.clamp(x_sample, min=0.0, max=1.0)
-    x_sample = 255. * np.moveaxis(x_sample.cpu().numpy(), 0, 2)
-    x_sample = x_sample.astype(np.uint8)
+    x_sample = x_sample.cpu()
+    x_sample.clamp_(0.0, 1.0)
+    x_sample.mul_(255.)
+    x_sample.round_()
+    x_sample = x_sample.to(torch.uint8)
+    x_sample = np.moveaxis(x_sample.numpy(), 0, 2)
 
     return Image.fromarray(x_sample)
 
diff --git a/modules/shared_state.py b/modules/shared_state.py
index bdbb4714..699cb0fc 100644
--- a/modules/shared_state.py
+++ b/modules/shared_state.py
@@ -4,8 +4,10 @@ import threading
 import time
 import traceback
 import torch
+from contextlib import nullcontext
 
 from modules import errors, shared, devices
+from backend.args import args
 from typing import Optional
 
 log = logging.getLogger(__name__)
@@ -34,6 +36,10 @@ class State:
 
     def __init__(self):
         self.server_start = time.time()
+        if args.cuda_stream:
+            self.vae_stream = torch.cuda.Stream()
+        else:
+            self.vae_stream = None
 
     @property
     def need_restart(self) -> bool:
@@ -153,12 +159,18 @@ class State:
         import modules.sd_samplers
 
         try:
-            if shared.opts.show_progress_grid:
-                self.assign_current_image(modules.sd_samplers.samples_to_image_grid(self.current_latent))
+            if self.vae_stream is not None:
+                # not waiting on default stream will result in corrupt results
+                # will not block main stream under any circumstances
+                self.vae_stream.wait_stream(torch.cuda.default_stream())
+                vae_context = torch.cuda.stream(self.vae_stream)
             else:
-                self.assign_current_image(modules.sd_samplers.sample_to_image(self.current_latent))
-
-            self.current_image_sampling_step = self.sampling_step
+                vae_context = nullcontext()
+            with vae_context:
+                if shared.opts.show_progress_grid:
+                    self.assign_current_image(modules.sd_samplers.samples_to_image_grid(self.current_latent))
+                else:
+                    self.assign_current_image(modules.sd_samplers.sample_to_image(self.current_latent))
 
         except Exception as e:
             # traceback.print_exc()