Most efficient way to find nearest positive semi-definite matrix?

Currently working on a GAN, for which I need to calculate the nearest positive semi-definite matrix to the covariance matrix of the data distribution. The smallest embedding space that works is with a dimension of 60416, meaning that my covariance matrix is 60416x60416.

I have made an extensive search, which led me to this question: https://www.mathworks.com/matlabcentral/answers/471045-how-to-find-the-nearest-positive-definite-matix. I have tried the script found at https://www.mathworks.com/matlabcentral/fileexchange/42885-nearestspd, but has crashed the cluster where the experiments are being conducted and even if it eventually runs, the current memory usage is still not desirable.

Is there any way to more efficiently make this computation? The script is almost 10 years old and I find it very odd that there is not a more recent/optimized version. Thanks in advance for any help provided.

[Matt J]

The recency of the code is likely not the problem. The size of your matrix is. Can I assume your matrix is not sparse?

[Torsten]

The script is almost 10 years old and I find it very odd that there is not a more recent/optimized version.

Good things last ...

[Pedro Louro]

@Matt J The matrix is not sparse at all.

[Matt J]

Well, a 60416*60416 non-sparse matrix consumes about 13GB of storage in single floats, 27GB in double. You likely do not have any RAM left over for the computations.

[Walter Roberson]

you can make some minor changes to the code such as

H = V*Sigma*V';

After that you could clear V. A couple of lines later you can clear H.

These minor changes might allow you handle slightly larger matrives, but I doubt it will make much difference.