MATLAB/Simulink performance on Framework Desktop 128GB?

Framework Desktop
1

Hi,

Considering the Framework Desktop (Ryzen AI Max+ 395, 128GB) for MATLAB R2025b / Simulink work – mainly MPC simulations and parallel computing. No GPU acceleration needed.

Has anyone run MATLAB on this hardware? Interested in:

  • General performance

  • Parallel workers (8-16 simultaneously)

  • Linux compatibility

  • Thermal behavior during long runs

Thanks!

2

UPDATE: Performance Showdown — Framework Desktop (128GB) vs. Framework 13 (96GB)

I’ve finished a head-to-head comparison between the new Framework Desktop (Ryzen AI Max+ 395, 128GB) and my Framework 13 (Ryzen 7 7840U). Both systems are running openSUSE Tumbleweed and MATLAB R2026a.

Here are the results showing exactly where the Desktop’s workstation architecture pulls ahead.

1. Standard bench Results (Lower is Better)

Metric Framework 13 (Laptop) Framework Desktop Improvement
LU (Linear Algebra) 0.3555s 0.1810s ~2.0x faster
FFT (Signal Processing) 0.2428s 0.2473s Parity
ODE (Solver Base) 0.0970s 0.0873s ~10% faster
Metric Framework 13 (7840U) Framework Desktop
Sparse 0.4136s 0.6327s
Graphics 0.1906s 1.0113s*

*Note: The Desktop’s Graphics and Sparse scores are influenced by the high-resolution 49" monitor setup and cache management differences. In real-world GPU-accelerated Simulink tasks, the Desktop feels significantly more responsive.

2. Extended Workload Comparison (Custom Stress Test)

Test Framework 13 (8 Workers) Framework Desktop (16 Workers) Delta
Memory Throughput (32k Matrix) 0.2553s 0.0823s 3.1x Faster
Parallel Workload (Full Load) 3.00s 1.47s 2.0x Faster
Memory Bus Width 128-bit 256-bit 2x Width

3. The Comparison Script

For those who want to replicate this on their own hardware:

%% Framework Hardware Benchmark: Desktop (Strix Halo) vs Laptop (7040U)
% Focus: Memory Throughput and Parallel Scaling
clear; clc;

fprintf(‘— Starting Extended Framework Benchmarks —\n’);

%% 1. Memory Bandwidth Test (Massive Array)
% Why: Tests the 256-bit bus (Desktop) vs 128-bit bus (Laptop).
% We use a large 32k matrix (~8GB) to force the data out of the CPU cache.
fprintf('Testing Memory Throughput (Large Matrix Addition)… ');
n_mem = 32000;
A = rand(n_mem, n_mem, ‘single’);
tic;
B = A + A;
t_mem = toc;
clear A B;
fprintf(‘Done: %.4f seconds\n’, t_mem);

%% 2. Parallel Processing & Scaling
% Why: Tests thermal headroom and multi-core efficiency.
fprintf('Testing Parallel Scaling… ');
p = gcp(‘nocreate’);
if isempty(p)
p = parpool(‘local’); % Starts max available workers
end
num_workers = p.NumWorkers;

n_par = 1000;
% Scaling load: 4 tasks per worker
num_tasks = num_workers * 4;

tic;
parfor i = 1:num_tasks
eig(rand(n_par, n_par));
end
t_par = toc;
fprintf(‘Done: %.2f seconds using %d workers\n’, t_par, num_workers);

%% 3. Summary Output
fprintf(‘\n— RESULTS SUMMARY —\n’);
fprintf(‘Memory Throughput:  %.4f s\n’, t_mem);
fprintf(‘Parallel Workload:  %.2f s (%d Workers)\n’, t_par, num_workers);
s = rendererinfo;
fprintf(‘Renderer Device:    %s\n’, s.RendererDevice);

System Specs:

  • Desktop: AMD RYZEN AI MAX+ 395, 128GB RAM, openSUSE Tumbleweed (Kernel 6.19.12)

  • Laptop: AMD Ryzen 7 7840U, 96GB RAM, openSUSE Tumbleweed (Kernel 6.19.5)

If anyone wants me to run specific MATLAB/Simulink tests or toolboxes on this 128GB/Strix Halo setup, let me know! I’ll try to check them when I have some spare time.

1 Like