Running H2O with 32-bit Java is really not recommended or supported. It will also limit you to very small datasets.
But, if necessary, you can start H2O at the command line manually, and then connect to the running H2O process using the web interface (or R or Python).
CMD> cd \path\to\h2o
CMD> java -Xmx1g -jar h2o.jar -ip localhost
04-07 15:59:30.483 127.0.0.1:54321 58927 main INFO: ----- H2O started -----
04-07 15:59:30.503 127.0.0.1:54321 58927 main INFO: Build git branch: rel-tutte
04-07 15:59:30.503 127.0.0.1:54321 58927 main INFO: Build git hash: e27e9002802e76b3ebb56a77f5bb82b827d5b810
04-07 15:59:30.503 127.0.0.1:54321 58927 main INFO: Build git describe: jenkins-rel-tutte-2
...
04-07 15:59:30.922 127.0.0.1:54321 58927 main INFO: Registered: 136 REST APIs in: 368mS
04-07 15:59:31.475 127.0.0.1:54321 58927 main INFO: Registered: 201 schemas in 552ms
04-07 15:59:31.476 127.0.0.1:54321 58927 main INFO:
04-07 15:59:31.476 127.0.0.1:54321 58927 main INFO: Open H2O Flow in your web browser: http:
04-07 15:59:31.476 127.0.0.1:54321 58927 main INFO:
Now open your web browser on localhost: 54321, as said at the bottom.
-Xmx1g , , .
[ "-ip" - H2O , H2O , , . ]