Why does this code run (20 to 100)% slower on R2025a compared to R2024b?

clc
clear
close
syms t x1(t) y1(t) x2(t) y2(t) x3(t) y3(t)
G = 6.6743 * 10^-11;
%epsilon = 1e-4
m1 = 10^12;
m2 = 1*10^12;
m3 = 1*10^12;
r1 = [x1(t),y1(t)];
K1 = 1/2 * m1 * (diff(x1(t),t)^2 + diff(y1(t),t)^2);
r2 = [x2(t),y2(t)];
K2 = 1/2 * m2 * (diff(x2(t),t)^2 + diff(y2(t),t)^2);
r3 = [x3(t),y3(t)];
K3 = 1/2 * m3 * (diff(x3(t),t)^2 + diff(y3(t),t)^2);
L1x = diff(diff(K1,diff(x1(t),t)) , t);
L1y = diff(diff(K1,diff(y1(t),t)) , t);
L2x = diff(diff(K2,diff(x2(t),t)) , t);
L2y = diff(diff(K2,diff(y2(t),t)) , t);
L3x = diff(diff(K3,diff(x3(t),t)) , t);
L3y = diff(diff(K3,diff(y3(t),t)) , t);
r12 = r2 - r1;
r13 = r3 - r1;
r23 = r3 - r2;
dlugosc_r12 = sqrt(r12(1)^2 + r12(2)^2);
dlugosc_r13 = sqrt(r13(1)^2 + r13(2)^2);
dlugosc_r23 = sqrt(r23(1)^2 + r23(2)^2);
Q12 = G * m1 * m2 / dlugosc_r12^2 * (r2-r1)/dlugosc_r12;
Q13 = G * m1 * m3 / dlugosc_r13^2 * (r3-r1)/dlugosc_r13;
Q23 = G * m2 * m3 / dlugosc_r23^2 * (r3-r2)/dlugosc_r23;
Q21 = -Q12;
Q32 = -Q23;
Q31 = -Q13;
Q1 = Q12 + Q13;
Q2 = Q21 + Q23;
Q3 = Q31 + Q32;
eqn_1_x = L1x == Q1(1);
eqn_1_y = L1y == Q1(2);
eqn_2_x = L2x == Q2(1);
eqn_2_y = L2y == Q2(2);
eqn_3_x = L3x == Q3(1);
eqn_3_y = L3y == Q3(2);
syms X1 Y1 X2 Y2 X3 Y3
Q1_num = subs(Q1,[x1(t), y1(t), x2(t), y2(t), x3(t), y3(t)],[X1, Y1, X2, Y2, X3, Y3]);
Q2_num = subs(Q2,[x1(t), y1(t), x2(t), y2(t), x3(t), y3(t)],[X1, Y1, X2, Y2, X3, Y3]);
Q3_num = subs(Q3,[x1(t), y1(t), x2(t), y2(t), x3(t), y3(t)],[X1, Y1, X2, Y2, X3, Y3]);
syms vx1 vy1 vx2 vy2 vx3 vy3
state_dot = [
    vx1;
    vy1;
    vx2;
    vy2;
    vx3;
    vy3;
    Q1_num(1)/m1;
    Q1_num(2)/m1;
    Q2_num(1)/m2;
    Q2_num(2)/m2;
    Q3_num(1)/m3;
    Q3_num(2)/m3
    ];
f = matlabFunction(state_dot, 'Vars', {sym('t'), [X1; Y1; X2; Y2; X3; Y3; vx1; vy1; vx2; vy2; vx3; vy3]});
u0 = [-1e5;         %x1
    0;              %y1
    1e5;            %x2   
    0;              %y2
    0;              %x3
    sqrt(3)*1e5;        %y3    
    -11/2 * 1e-3;        %vx1
    11/2*sqrt(3)*1e-3;       %vy1
    -11/2 * 1e-3;        %vx2
    -11/2*sqrt(3)*1e-3;      %vy2
    11e-3;           %vx3
    0];         %vy3
tspan = [0, 1e9];
%options = odeset('RelTol', 1e-15, 'AbsTol', 1e-20);
[t_sol, u_sol] = ode45(f, tspan, u0);
t_anim = linspace(t_sol(1), t_sol(end), 5000);
u_anim = interp1(t_sol, u_sol, t_anim);
%% 
% figure;
tor_1 = plot(u_anim(:,1), u_anim(:,2), 'r', 'LineWidth',1.5); hold on;
tor_2 = plot(u_anim(:,3), u_anim(:,4), 'g', 'LineWidth',1.5);
tor_3 = plot(u_anim(:,5), u_anim(:,6), 'b', 'LineWidth',1.5);
% xlabel('x [m]');
% ylabel('y [m]');
% legend('Ciało 1', 'Ciało 2', 'Ciało 3');
% title('Trajektorie ciał w układzie trzech ciał');
% axis equal
% grid on;
pozycja_1 = plot(u_anim(1,1),u_anim(1,2),'ro','markersize',10,'markerface','r'); hold on
pozycja_2 = plot(u_anim(1,3),u_anim(1,4),'go','markersize',10,'markerface','g');
pozycja_3 = plot(u_anim(1,5),u_anim(1,6),'bo','markersize',10,'markerface','b');
% xlim([-2e5,2e5])
% ylim([-2e5,2e5])
axis equal
for i = 1 : 1 : length(t_sol)
    set(pozycja_1,'XData', u_anim(i,1),'YData', u_anim(i,2));
    set(pozycja_2,'XData', u_anim(i,3),'YData', u_anim(i,4)); 
    set(pozycja_3,'XData', u_anim(i,5),'YData', u_anim(i,6)); 
    set(tor_1,'XData', u_anim(1:i,1),'YData', u_anim(1:i,2));
    set(tor_2,'XData', u_anim(1:i,3),'YData', u_anim(1:i,4));
    set(tor_3,'XData', u_anim(1:i,5),'YData', u_anim(1:i,6));
    drawnow;
    % pause(0.001);
end

This is an animated solution of the three body problem. On 2024b it runs perfectly fine. When we tried it on 2025a, the animation constantly hitches, the CPU usage is almost double and the runtime is much slower. The curves also look less detailed and jagged in some places. When we run in without drawing anything it seems comparable, but still slightly slower. All of this behavior persists across different hardware. Anybody else having this kind of problem with the new release?

[Princeton]

I am also having this problem. My n-body planetary simulation is also having constant hitches and slowdowns, with the planets being a bit shaky/vibrating throughout the program. As my animation goes longer and longer, the hitches only get worse and more frequent, to a point where it is unwatchable and barely progresses. This only happened in 2025a, and works completely fine in 2024b.

[Matthew]

Yes, I have some real time plotting in my robotics path planning simulations, and now it's slow to load and stutters. It's practically unusable now and it's not worth my time to try and figure out if it's something in my code. I'm considering reverting back to 2024b.

Answer 1

[Adam Danz]

Hello @Mateusz, I ran your demo in R2024b and R2025a and can confirm a bit of a slowdown in 25a.

Please contact tech support and include your demo above along with your system information. Feel free to include the URL of this thread.

[Mateusz]

Apparently, the new desktop is to blame, as support was able to reproduce my issue, not only on 2025a, but also on the demo desktop on 2024b. Unfortunately, I am disappointed to report that a MathWorks employee, having looked at the issue, responded with the link to an article about animation best practices...