Bay 12 Games Forum

Please login or register.

Login with username, password and session length
Advanced search  
Pages: 1 [2]

Author Topic: DF and Quad-Channel Ram  (Read 2899 times)

wierd

  • Bay Watcher
  • I like to eat small children.
    • View Profile
Re: DF and Quad-Channel Ram
« Reply #15 on: December 18, 2016, 12:42:38 am »

The random generator driven nature of DF events means that running the actual game for benchmarking is not a good solution.

A better solution, IMO, would be a surrogate test suite, created using the data collected from the dfhack team.

EG, a bunch of true benchmark computation suites that test performance based on a number of canned computational load patterns, that produce and iterate through big memory structures like DF does.  The data in the structures would always be the same, the computations done on them would always be the same, etc.  As long as it does things reasonably faithfully to the way DF does them, it would make a good baseline to work with.
Logged

BauxiteProcessor

  • Bay Watcher
    • View Profile
Re: DF and Quad-Channel Ram
« Reply #16 on: December 18, 2016, 04:28:07 am »

The random generator driven nature of DF events means that running the actual game for benchmarking is not a good solution.

A better solution, IMO, would be a surrogate test suite, created using the data collected from the dfhack team.

EG, a bunch of true benchmark computation suites that test performance based on a number of canned computational load patterns, that produce and iterate through big memory structures like DF does.  The data in the structures would always be the same, the computations done on them would always be the same, etc.  As long as it does things reasonably faithfully to the way DF does them, it would make a good baseline to work with.

That would be quite super interesting to see. Of course actual game benchmarks could be used to sanity check things in cases where results seem weird.
Logged

redivider

  • Bay Watcher
    • View Profile
Re: DF and Quad-Channel Ram
« Reply #17 on: December 18, 2016, 07:00:06 am »

If you run it for a small time interval and repeat the run more times and avarage it, the random wont have a big impact, in other words the error will be minimal/not seen in practice. You would actually get a practical, functional "real world" result instead of artificial benchmarks of, say an ssd being 0.003s faster.
Logged

Max™

  • Bay Watcher
  • [CULL:SQUARE]
    • View Profile
Re: DF and Quad-Channel Ram
« Reply #18 on: December 19, 2016, 02:20:33 pm »

If you want a computer cruncher, get down to hell in a packed dark fortress and deconstruct it.
Logged

Fleeting Frames

  • Bay Watcher
  • Spooky cart at distance
    • View Profile
Re: DF and Quad-Channel Ram
« Reply #19 on: December 19, 2016, 04:32:28 pm »

Would have to be same save every time, but no idea how well caveins correlate with other stuff.
Pages: 1 [2]