2023b update 5 is slower

Anyone else having the same issue?

[Matt J]

How would it be noticed? What timing tests show this?

[psousa]

The code running is slower. Also my laptop is slower. Both of these are very obvious.

[Dyuman Joshi]

"The code running is slower."

How do you know? Did you run any timing tests, as Matt has asked before? If yes, then please share the results.

"Also my laptop is slower. "

Then most likely the problem is with your Laptop and not MATLAB.

A lot of people have upgraded to R2023b and there have been (almost) no reports of a downgrade in code performance.

You will need to specify and provide more information for us to be able to figure out the issue and provide a suggestion/solution.

[psousa]

Doesn't everyone record runtimes for their runs? Come on now... it's slower, if I tell you run times doesn’t really change anything about the issue.

And this issue only started with update 5. It was fine after update 4.

This seems to me an issue with memory allocation with parpool.

[Walter Roberson]

What happens if you use R2023a ?

[psousa]

That will definitely be slower. Adam solver got substantially faster with 2023b.

[psousa]

Update: downloaded the 2024a pre-release and performance is back to 2023b update 4 levels.

[Sulaymon Eshkabilov]

You can test your computer's performance (along with your updated MATLAB) using builtin benchmarking function from the command window:

>> bench

Answer 1

[Benjamin Kraus]

Update releases include primarily small bug fixes, and as such if they have any impact on the performance of MATLAB, they are likely to only impact specific features or functions, and are very unlikely to cause MATLAB in general to be slower or faster, unless there is some kind of issue with the MATLAB installation.

One thing you could try is a clean installation of MATLAB R2023b Update 5 (rather than an having patched the previous version). It is possible something went wrong with the update process, and a clean installation will resolve the issue.

If a clean installation doesn't help, and you are experiencing an issue that makes MATLAB R2023b Update 5 slower than R2023b Update 4, then it is very likely an issue specifically with something that your code is doing that is running slower due to the update. In order to identify exactly what that is, we would need to see more specific example code, or perhaps profiler results that show the difference between the two MATLAB releases. I recognize that once you've upgraded, getting profiler results from Update 4 would be impossible, but perhaps just showing profiler results for Update 5 would help in identifying what is slower.

You also mentioned that R2024a pre-release resolved the issue, so a profiler comparison between R2023b Update 5 and R2024a pre-release would be informative in identifying any differences.

I think it would also help to give some kind of more quantitative estimate of what you mean by "slower". It doesn't have to be exact, but, for example: "Something that used to take approximately 10 seconds is now taking 100 seconds." vs. "Something that used to take 60 seconds is now taking 61 seconds." Such a statement would help in understanding the issue.

Some additional information is essential if you want to resolve the slowness you are experiencing.

Another option: If you contact technical support, they have ready access to both Update 4 and Update 5. If you can provide them with runnable code that is reliably slower in Update 5, they can easily compare the profiler results between Update 4 and Update 5 to see if the issue is reproducible by someone at MathWorks, and if so they can help to identify the source of the performance regression.