diff --git a/include/types.h b/include/types.h
index 489820f7b..c87bb27b2 100644
--- a/include/types.h
+++ b/include/types.h
@@ -1048,6 +1048,7 @@ typedef struct hc_device_param
   u32     speed_pos;
   u64     speed_cnt[SPEED_CACHE];
   double  speed_msec[SPEED_CACHE];
+  bool    speed_only_finish;
 
   hc_timer_t timer_speed;
 
diff --git a/src/dispatch.c b/src/dispatch.c
index 7c1621001..4f0170a01 100644
--- a/src/dispatch.c
+++ b/src/dispatch.c
@@ -269,7 +269,7 @@ static int calc_stdin (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_par
 
     if (status_ctx->run_thread_level1 == false) break;
 
-    if (user_options->speed_only == true) break;
+    if (device_param->speed_only_finish == true) break;
   }
 
   device_param->kernel_accel_prev = device_param->kernel_accel;
@@ -363,7 +363,7 @@ static int calc (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param)
 
       device_param->pws_cnt = 0;
 
-      if (user_options->speed_only == true) break;
+      if (device_param->speed_only_finish == true) break;
 
       if (status_ctx->run_thread_level2 == true)
       {
@@ -668,7 +668,7 @@ static int calc (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param)
         */
       }
 
-      if (user_options->speed_only == true) break;
+      if (device_param->speed_only_finish == true) break;
 
       if (status_ctx->run_thread_level2 == true)
       {
diff --git a/src/interface.c b/src/interface.c
index fc0262bda..decb8cc51 100644
--- a/src/interface.c
+++ b/src/interface.c
@@ -26007,6 +26007,24 @@ u32 hashconfig_get_kernel_threads (hashcat_ctx_t *hashcat_ctx, const hc_device_p
       if (device_param->kernel_preferred_wgs_multiple_loop2)  kernel_threads = MIN (kernel_threads, device_param->kernel_preferred_wgs_multiple_loop2);
     }
   }
+  else
+  {
+    if (hashconfig->attack_exec == ATTACK_EXEC_INSIDE_KERNEL)
+    {
+      if (hashconfig->opti_type & OPTI_TYPE_OPTIMIZED_KERNEL)
+      {
+        if (device_param->kernel_wgs1) kernel_threads = MIN (kernel_threads, device_param->kernel_wgs1);
+      }
+      else
+      {
+        if (device_param->kernel_wgs4) kernel_threads = MIN (kernel_threads, device_param->kernel_wgs4);
+      }
+    }
+    else
+    {
+      if (device_param->kernel_wgs2) kernel_threads = MIN (kernel_threads, device_param->kernel_wgs2);
+    }
+  }
 
   // we'll return a number power of two, makes future processing much more easy
   // kernel_threads = power_of_two_floor_32 (kernel_threads);
diff --git a/src/opencl.c b/src/opencl.c
index 78bd55700..5a3c4ab1c 100644
--- a/src/opencl.c
+++ b/src/opencl.c
@@ -1322,7 +1322,9 @@ int choose_kernel (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param,
           {
             device_param->speed_pos = 1;
 
-            return -2; // special RC
+            device_param->speed_only_finish = true;
+
+            return 0;
           }
         }
       }
@@ -1528,7 +1530,7 @@ int run_kernel (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, con
 
   kernel_threads = MIN (kernel_threads, device_param->kernel_threads);
 
-  kernel_threads = power_of_two_floor_32 (kernel_threads);
+  // kernel_threads = power_of_two_floor_32 (kernel_threads);
 
   while (num_elements % kernel_threads) num_elements++;
 
@@ -2468,10 +2470,7 @@ int run_cracker (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, co
        * benchmark was aborted because too long kernel runtime (slow hashes only)
        */
 
-      if (user_options->speed_only == true)
-      {
-        if (rc == -2) break;
-      }
+      if (device_param->speed_only_finish == true) break;
 
       /**
        * speed
@@ -2549,6 +2548,8 @@ int run_cracker (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, co
 
           device_param->speed_pos = 1;
 
+          device_param->speed_only_finish = true;
+
           break;
         }
       }
@@ -2569,7 +2570,7 @@ int run_cracker (hashcat_ctx_t *hashcat_ctx, hc_device_param_t *device_param, co
       device_param->outerloop_msec += device_param->speed_msec[0] * m * hashes->salts_cnt;
     }
 
-    if (user_options->speed_only == true) break;
+    if (device_param->speed_only_finish == true) break;
 
     //status screen makes use of this, can't reset here
     //device_param->innerloop_pos  = 0;
@@ -6317,6 +6318,8 @@ void opencl_session_reset (hashcat_ctx_t *hashcat_ctx)
     memset (device_param->speed_cnt,  0, SPEED_CACHE * sizeof (u64));
     memset (device_param->speed_msec, 0, SPEED_CACHE * sizeof (double));
 
+    device_param->speed_only_finish = false;
+
     device_param->exec_pos = 0;
 
     memset (device_param->exec_msec, 0, EXEC_CACHE * sizeof (double));