In your example, the wire coming out of the init array contains about 400MBytes of data. You split the wire, creating data copies, in several places. I have never been able to get LabVIEW to use more than about 1200MBytes of RAM - this is two extra copies (and yes, I have tried). You probably have somewhere between four and eight copies, depending upon what version of LabVIEW you are using. Read the tutorial I mentioned earlier. You will need that information to survive this programming experience.
I must respectfully disagree with VI Guy about saving using flatten to string. First, flatten to string makes a copy. You have enough copies already. Second, when you read the data, you must know exactly how it was flattened or you lose it. My experience has been that such information gets lost in about three to five months. Third, you can write the data directly to disk using the write file primitive - it takes any data type. You will still need to know how you wrote it, but if you interleave it in a obvious fashion, it will not be as opaque as a flatten to string and you can figure it out again if you need to in the future. Use the example finder to browse and look at all the examples on file I/O, especially the binary ones. You can also open the binary file storage VIs and look at how that was done, then modify for your case.
Is your data being generated all at once or is it coming piecemeal from somewhere? If piecemeal, store it as it is generated. If all-at-once, use the techniques in the tutorial to store it, then save it piecemeal. You want to avoid 400MByte wires at all costs. Dealing with this much data is an advanced topic, so don't be surprised if it takes you some time to figure out. The first real question you should ask yourself is “Do I really need this much data?”
If you do need that much data, I recommend you use HDF5 for this. It will take you two weeks to figure out, but you will be glad you did.
(NOTE: I will be out for vacation until next year.)
