RAM usage

[by...@byronknoll.com]

The rules state there is a limit of "10GB RAM". I have a couple questions about this:

1) How much physical RAM does the test machine have? The geekbench page says 8GB. Why is the contest limit 10GB if the test machine only has 8GB? Will the amount of RAM available for the entry be even lower than this, due to memory usage from the OS and other running programs?

2) Can submitted programs use over 10GB of memory and rely on swap disk usage? For example, using 20GB of memory, filling up the 8GB of physical RAM, and using 12GB of swap space? The swap space could count towards the 100GB disk usage limit.

[Matt Mahoney]

1. I have a new test machine. 

Lenovo 82HT, core i7-1165G7, 2.79 GHz, 4 cores, 8 threads, 16 GB RAM, 512 GB SSD, Windows 10. Geekbench score 1427 single core, 4667 multi core. I plan to make it dual boot Linux (probably Ubuntu).

Old machine from 2010: Dell Latitude E6510, core i7-620M, 2.67 GHz, 2 cores, 4 threads, 4 GB RAM, hard disk, Ubuntu. Geekbench 5 score 566 single 1104 multi.

I ran the geekbench 5 tests yesterday. Apparently this is different than geekbench 4 from 2016 so I plan to retest. The single thread test is the one that matters.

https://browser.geekbench.com/v4/cpu/145066

2. You will need to manage your own swap space so that the Linux time command or Windows timer program shows 10 GB memory allocated. Otherwise swapping will drastically slow down the program on a machine that actually had 10 GB of free memory. I realize most machines you would be running it on would have at least 12 or 16 GB.

--
You received this message because you are subscribed to the Google Groups "Hutter Prize" group.
To unsubscribe from this group and stop receiving emails from it, send an email to hutter-prize...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/hutter-prize/7add2fbe-7d50-4337-b365-2bc3cf1921bcn%40googlegroups.com.

[by...@byronknoll.com]

Thanks for the information! A couple of follow-up questions:

1) The rules state: "100 hours on our test machines". Does this mean the competition becomes easier as the test machines are upgraded? i.e. is the only requirement that the program needs to run in <100 hours on the slowest test machine? Or will the time limit be scaled by the geekbench result of the original test machine (550)?

2) Would it be possible for you to post the run time of phda9v1.8 on the new test machine? I think that would allow contestants a more accurate way to estimate the time usage of their entries (i.e. more accurate than the geekbench result).

[Matt Mahoney]

I will need to consult with the Hutter prize committee with the rules
regarding time requirements. The rules say 100 hours on the old
machine (which is not possible because it only has 4 GB of memory), or
50,000/(geekbench 5 score) hours. Also, one core = 2 threads.

I tested both machines with both Geekbench 4 and 5. Scores: (single
core, multi core):

Old Dell (Ubuntu): GB4: 2505, 4399. GB5: 566, 1104.
https://browser.geekbench.com/v4/cpu/16191052

New Lenovo (Win10): GB4: 6638, 20384. GB5: 1427, 4667.
https://browser.geekbench.com/v4/cpu/16191053

> To view this discussion on the web visit https://groups.google.com/d/msgid/hutter-prize/5d38f4df-b01e-4c53-8736-5b9dfcd1ba3bn%40googlegroups.com.



--
-- Matt Mahoney, mattma...@gmail.com