diff --git a/ext/KiteModelsControlPlotsExt.jl b/ext/KiteModelsControlPlotsExt.jl
index 0e82a0a5..8e39f40a 100644
--- a/ext/KiteModelsControlPlotsExt.jl
+++ b/ext/KiteModelsControlPlotsExt.jl
@@ -44,4 +44,4 @@ function ControlPlots.plot(s::SymbolicAWEModel, reltime; kwargs...)
     plot(s.system_structure, reltime; s.set.l_tether, wing_pos, e_z, kwargs...)
 end
 
-end
\ No newline at end of file
+end
diff --git a/src/KPS4.jl b/src/KPS4.jl
index a874bac4..c5dee9c6 100644
--- a/src/KPS4.jl
+++ b/src/KPS4.jl
@@ -389,7 +389,7 @@ Output:
         height = 0.5 * (pos[p1][3] + pos[p2][3])
         rho = calc_rho(s.am, height)
         # if height <= -1000.0
-        #     println("Error: height: $height")
+            # println("Error: height: $height")
         # end
         @assert height > -1000
         if height < 6
@@ -496,13 +496,12 @@ function residual!(res, yd, y::MVector{S, SimFloat}, s::KPS4, time) where S
     if !isnothing(s.sync_speed) && s.sync_speed == 0.0
         use_brake = true
     end
-    # if !isnothing(s.sync_speed)
-    #     res[end] = v_reel_outd - calc_acceleration(s.wm, v_reel_out, norm(s.forces[1]), s.sync_speed)        
-    # else
+    if s.wm isa AsyncMachine
+        res[end] = v_reel_outd - calc_acceleration(s.wm, s.sync_speed, v_reel_out, norm(s.forces[1]), true)       
+    else
         res[end] = v_reel_outd - calc_acceleration(s.wm, v_reel_out, norm(s.forces[1]); set_speed=s.sync_speed, 
             set_torque=s.set_torque, use_brake)
-    # end
-
+    end
     # copy and flatten result
     for i in 2:div(T,6)+1
         for j in 1:3
diff --git a/src/KiteModels.jl b/src/KiteModels.jl
index f55922d3..d6ec848d 100644
--- a/src/KiteModels.jl
+++ b/src/KiteModels.jl
@@ -760,4 +760,4 @@ end
 
 include("precompile.jl")
 
-end
\ No newline at end of file
+end
diff --git a/test/bench4.jl b/test/bench4.jl
index 5fe4761f..1035b1e9 100644
--- a/test/bench4.jl
+++ b/test/bench4.jl
@@ -130,15 +130,16 @@ msg = String[]
     t = @benchmark residual!($res, $yd0, $y0, $kps4, $time)
     push!(msg, ("Mean time residual!:           $(round(mean(t.times), digits=1)) ns"))
     # println("t.memory: ", t.memory)
-    @test t.memory <= 176
+    @test t.memory <= 0
 
     # time using Python/ Numba: 8.94 µs, time using Julia 1.7.2: 1.6µs, Julia 1.8.0: 1.244µs
     # Julia 1.9 on Ryzen:  816.1 ns
     # Julia 1.10 on Ryzen: 787.0 ns 6000 RAM
-    # Julia 1.10 on Laptop on battery: 1047 ns
-    # Julia 1.11 on Laptop on power:   1024..1090 ns
-    # Julia 1.11 on Laptop on battery: 1050..1197 ns
-    # Julia 1.11 on Desktop on main branch: 835..899 ns on June 15, 2025 after 14:12
+    # Julia 1.10 on Laptop on battery: 1047ns
+    # Julia 1.11 on Laptop on battery: 1035ns
+    # Julia 1.11 on Laptop on battery, branch perf2: 1042ns
+    # Julia 1.11 on Desktop, branch perf2: 835..899 ns on June 15, 2025 after 13:12
+
 end
 printstyled("Benchmark results for KPS4:\n"; bold = true)
 for i in eachindex(msg)