-snip-
You can´t check 1600 passwords a second, sorry to dissapoint you.
Its not an estimate. Btcrecover[1] achives ~1.68kP/s on my GPU (GTX 970) vs. bitcoin core wallet files. I cant get it to work with UTF encoding, which slows it down significantly for certain passwords, but that might just be a problem on my end. The CPU (4th gen i5) does ~360kP/s vs. multibit key files (no GPU support) using the same set of tools. Things like the saber cluster[2] will rip weak passwords a new one and thats a "cheap" 1 year old cluster.
Most properly written encryption software will allow 1 per second if the iterations are set high. In your case, you may need the help of a professional. If you had a passphrase with a high bit rate you are going to have to remember a good portion of your password, and then a script can be written to attack it. Again though, if the encryption for the wallet was written propertly, you may only be able to try 1 password per second. I used a script written by someone else to crack my LUKS encryption password to a flashdrive. I knew what the pw was. just didn´t know what symbols I added to the beginning and end. I was able to incorporate the symbols I thought were used in the script and after several days i got it. however, i knew most of the passphrase. LUKS encryption only allows you to try 1 password per second because of high iterations.
Yes its possible to crank up the number of rounds to slow things down or even use a nested scheme of several encryption mechanisms like e.g. VeraCrypt or TrueCrypt, however most bitcoin wallets dont seem to use it to an extent that it slows attacks down to 1 try per second.
IIRC bitcoin core sets the number of rounds when creating the wallet for the first time. Its aiming at 1 attempt per second per core when it does it by itself. Its however possible to attack the wallet.dat directly or rather a certain string from it and not use core for it which can speed things up even on a CPU.
Just give btcrecover a try.
[1]
https://github.com/gurnec/btcrecover[2]
http://blog.cr.yp.to/20140602-saber.html