diff --git a/aria2c-win.exe b/aria2c-win.exe
new file mode 100644
index 000000000..5004e1034
Binary files /dev/null and b/aria2c-win.exe differ
diff --git a/expose.h b/expose.h
index 5f48fe6cf..daea7472a 100644
--- a/expose.h
+++ b/expose.h
@@ -69,6 +69,7 @@ struct load_model_inputs
     const float tensor_split[tensor_split_max] = {};
     const int quant_k = 0;
     const int quant_v = 0;
+    const bool check_slowness = false;
     const bool quiet = false;
     const int debugmode = 0;
 };
diff --git a/gpttype_adapter.cpp b/gpttype_adapter.cpp
index 35b774b57..bcb113968 100644
--- a/gpttype_adapter.cpp
+++ b/gpttype_adapter.cpp
@@ -136,6 +136,7 @@ static std::string concat_output_reader_copy_res = ""; //for gen response
 static std::vector<logit_bias> logit_biases;
 static bool add_bos_token = true; // if set to false, mmproj handling breaks. dont disable unless you know what you're doing
 static bool load_guidance = false; //whether to enable cfg for negative prompts
+static bool check_slowness = false; //will display a suggestion to use highpriority if slow
 
 static int delayed_generated_tokens_limit = 0;
 std::deque<std::string> delayed_generated_tokens; //for use with antislop sampling
@@ -1927,6 +1928,7 @@ ModelLoadResult gpttype_load_model(const load_model_inputs inputs, FileFormat in
     max_context_limit_at_load = clamped_max_context_length;
     add_bos_token = !inputs.no_bos_token;
     load_guidance = inputs.load_guidance;
+    check_slowness = inputs.check_slowness;
 
     if(!add_bos_token)
     {
@@ -4164,6 +4166,14 @@ generation_outputs gpttype_generate(const generation_inputs inputs)
     {
         printf("\n(Draft Results - Success:%d, Failure:%d)",draft_successes,draft_failures);
     }
+    if(check_slowness && ts2<2.0f)
+    {
+        check_slowness = false;
+        if(!is_quiet)
+        {
+            printf("\n======\nNote: Your generation speed appears rather slow. You can try relaunching KoboldCpp with the high priority toggle (or --highpriority) to see if it helps.\n======\n");
+        }
+    }
     fflush(stdout);
     output.status = 1;
     int finaltokcount = (int)current_context_tokens.size()-realnpredict;
diff --git a/koboldcpp.py b/koboldcpp.py
index f36fb22f0..f771f4592 100644
--- a/koboldcpp.py
+++ b/koboldcpp.py
@@ -190,6 +190,7 @@ class load_model_inputs(ctypes.Structure):
                 ("tensor_split", ctypes.c_float * tensor_split_max),
                 ("quant_k", ctypes.c_int),
                 ("quant_v", ctypes.c_int),
+                ("check_slowness", ctypes.c_bool),
                 ("quiet", ctypes.c_bool),
                 ("debugmode", ctypes.c_int)]
 
@@ -1234,6 +1235,7 @@ def load_model(model_filename):
     inputs.load_guidance = args.enableguidance
     inputs.override_kv = args.overridekv.encode("UTF-8") if args.overridekv else "".encode("UTF-8")
     inputs.override_tensors = args.overridetensors.encode("UTF-8") if args.overridetensors else "".encode("UTF-8")
+    inputs.check_slowness = (not args.highpriority and os.name == 'nt' and 'Intel' in platform.processor())
     inputs = set_backend_props(inputs)
     ret = handle.load_model(inputs)
     return ret
@@ -5577,60 +5579,6 @@ def setuptunnel(global_memory, has_sd):
         print(str(ex))
         return None
 
-def unload_libs():
-    global handle
-    OS = platform.system()
-    dll_close = None
-    if OS == "Windows":  # pragma: Windows
-        from ctypes import wintypes
-        dll_close = ctypes.windll.kernel32.FreeLibrary
-        dll_close.argtypes = [wintypes.HMODULE]
-        dll_close.restype = ctypes.c_int
-    elif OS == "Darwin":
-        try:
-            try:  # macOS 11 (Big Sur). Possibly also later macOS 10s.
-                stdlib = ctypes.CDLL("libc.dylib")
-            except OSError:
-                stdlib = ctypes.CDLL("libSystem")
-        except OSError:
-            # Older macOSs. Not only is the name inconsistent but it's
-            # not even in PATH.
-            stdlib = ctypes.CDLL("/usr/lib/system/libsystem_c.dylib")
-        dll_close = stdlib.dlclose
-        dll_close.argtypes = [ctypes.c_void_p]
-        dll_close.restype = ctypes.c_int
-    elif OS == "Linux":
-        try:
-            stdlib = ctypes.CDLL("")
-        except OSError:
-            stdlib = ctypes.CDLL("libc.so") # Alpine Linux.
-        dll_close = stdlib.dlclose
-        dll_close.argtypes = [ctypes.c_void_p]
-        dll_close.restype = ctypes.c_int
-    elif sys.platform == "msys":
-        # msys can also use `ctypes.CDLL("kernel32.dll").FreeLibrary()`.
-        stdlib = ctypes.CDLL("msys-2.0.dll")
-        dll_close = stdlib.dlclose
-        dll_close.argtypes = [ctypes.c_void_p]
-        dll_close.restype = ctypes.c_int
-    elif sys.platform == "cygwin":
-        stdlib = ctypes.CDLL("cygwin1.dll")
-        dll_close = stdlib.dlclose
-        dll_close.argtypes = [ctypes.c_void_p]
-        dll_close.restype = ctypes.c_int
-    elif OS == "FreeBSD":
-        # FreeBSD uses `/usr/lib/libc.so.7` where `7` is another version number.
-        # It is not in PATH but using its name instead of its path is somehow the
-        # only way to open it. The name must include the .so.7 suffix.
-        stdlib = ctypes.CDLL("libc.so.7")
-        dll_close = stdlib.close
-
-    if handle and dll_close:
-        print("Unloading Libraries...")
-        dll_close(handle._handle)
-        del handle
-        handle = None
-
 def reload_from_new_args(newargs):
     try:
         args.istemplate = False
@@ -5766,10 +5714,23 @@ def downloader_internal(input_url, output_filename, capture_output, min_file_siz
     dl_success = False
 
     try:
-        if shutil.which("aria2c") is not None:
+        if os.name == 'nt':
+            basepath = os.path.abspath(os.path.dirname(__file__))
+            a2cexe = (os.path.join(basepath, "aria2c-win.exe"))
+            if os.path.exists(a2cexe): #on windows try using embedded a2cexe
+                rc = subprocess.run([
+                        a2cexe, "-x", "16", "-s", "16", "--summary-interval=30", "--console-log-level=error", "--log-level=error",
+                        "--download-result=default", "--allow-overwrite=true", "--file-allocation=none", "--max-tries=3", "-o", output_filename, input_url
+                    ], capture_output=capture_output, text=True, check=True, encoding='utf-8')
+                dl_success = (rc.returncode == 0 and os.path.exists(output_filename) and os.path.getsize(output_filename) > min_file_size)
+    except subprocess.CalledProcessError as e:
+        print(f"aria2c-win failed: {e}")
+
+    try:
+        if not dl_success and shutil.which("aria2c") is not None:
             rc = subprocess.run([
                     "aria2c", "-x", "16", "-s", "16", "--summary-interval=30", "--console-log-level=error", "--log-level=error",
-                    "--download-result=default", "--allow-overwrite=true", "--file-allocation=none", "-o", output_filename, input_url
+                    "--download-result=default", "--allow-overwrite=true", "--file-allocation=none", "--max-tries=3", "-o", output_filename, input_url
                 ], capture_output=capture_output, text=True, check=True, encoding='utf-8')
             dl_success = (rc.returncode == 0 and os.path.exists(output_filename) and os.path.getsize(output_filename) > min_file_size)
     except subprocess.CalledProcessError as e:
@@ -6226,7 +6187,7 @@ def kcpp_main_process(launch_args, g_memory=None, gui_launcher=False):
             os_used = sys.platform
             process = psutil.Process(os.getpid())  # Set high priority for the python script for the CPU
             oldprio = process.nice()
-            if os_used == "win32":  # Windows (either 32-bit or 64-bit)
+            if os.name == 'nt':  # Windows (either 32-bit or 64-bit)
                 process.nice(psutil.REALTIME_PRIORITY_CLASS)
                 print("High Priority for Windows Set: " + str(oldprio) + " to " + str(process.nice()))
             elif os_used == "linux":  # linux