Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[BloodLines]

So I am looking into building a new desktop, so I was wondering what things I should be looking for to make DF run as fast as I can. So I'm not looking for parts as much as just what I need as far as RAM and such. I will most likely be running it on Linux but I will have windows on dual boot as well. Thanks for all inputs

[superbob]

Fast CPU, fast RAM, a motherboard that will let you overclock both.

You want an Intel CPU, i5 is the sweet spot, i7 might be better due to larger l3 cache, best would be the type that runs on a lga2011 socket. Either way you want clock speed and a CPU version that you can overclock. For DF you can disable hyper-threading and all but two CPU cores, which may let you overclock it more.

Aside from that, GPU isn't important, SSD reportedly helps also for just DF you don't need more than 8GB RAM.

Keep in mind you'll be getting diminishing returns in respect to the amount of money you spend and effort you put into overclocking. You still won't be able to run a 16x16 embarks or hugely cluttered forts with lots of dwarves and moving water/magma.

[BloodLines]

Thank you very much you were very helpful, I was actually looking at the i7 so this could work perfectly with my plan. I had no plan of going 16x16 though it would be a little interesting to see that for curiosity sake. But at least getting a large world and have it saving, starting, and running at a good speed would make me very happy. So thank you once again this was exactly the information I was looking for.

[Antikythera]

so i read another thread in this forum from 2012 that talked about RAM latency being almost as important as CPU clock speed. is this still correct?

would lowering RAM latency be more beneficial than increasing its frequency?
i have loads of RAM. how about running the game from a RAM Disk rather than off the SSD?

i have a G3258 which i plan on overclocking heavily. with stock cooling it was at 4.5ghz first attempt.
my main system is a 3930k which i have running @ 4.4 ghz with all six cores. its under an average air cooler.

is the size of the on die cache important? the 3930k is 12M vs the g3258 3M.
i just noticed the 3930k has a big advantage in memory bandwith. so it seems i've answered my own question.

[deknegt]

Well currently Dwarf Fortress is still a 32bit process, so this means DF will use a maximum of 3.4GB of RAM, so in that case RAM that works faster (lower CAS). Higher frequency + Lower CAS = Quicker memory access

[Shonai_Dweller]

Well currently Dwarf Fortress is still a 32bit process, so this means DF will use a maximum of 3.4GB of RAM, so in that case RAM that works faster (lower CAS). Higher frequency + Lower CAS = Quicker memory access

Of course there's a 64 bit test version out sometime during this cycle so you might want to pack a new desktop full of ram just to test drive it and let the rest of us know how much is too much.

[Grimlocke]

I don't know what kind of fortresses you run, but I have only seen DF reserve more than 3.4GB memory when generating huge worlds with huge population caps and no historic figure culling.

If DF is going to be your heaviest workload there is not much sense in going with anything more than the overclockable i5. Benefit of all the more expensive ones is for multithreaded purposes, which DF doesn't benefit from at all (beyond 2 cores anyway).

The memory speeds are somewhat debatable. Wouldn't mind seeing someone actually benchmark this. Speaking of which, has anyone ever made a usable, somewhat standardized benchmark scenario for DF?

[BigD145]

Unless you're playing DF with Oculus I think Core i5 is a good choice for bang for your buck.

[Grimlocke]

Unless you're playing DF with Oculus I think Core i5 is a good choice for bang for your buck.

That is kind of an amusing idea. One could make a stereoscopic tileset and hack DF to display one for each eye.

3D ASCII! Truly we live in the future.

[superbob]

Unless you're playing DF with Oculus I think Core i5 is a good choice for bang for your buck.

Yeah that's pretty much what I meant by saying it's the sweet spot

I7 CPUs tend to have more cache, and the LGA2011 variants have much higher memory bandwidth, which may have some impact on single-core performance in DF. Of course they also have more cores/threads, which will have no impact on DF for the foreseeable future, so for most people that would be a waste of money, considering the cost/benefit ratio. Even more so because AFAIK it's not like you can use all of the CPU cache and memory bandwidth with just two cores.

If someone will put together a benchmark for the latest version, I'm willing to submit some results. I can test if the amount of CPU cache is actually important, I have these three CPUs on hand: http://ark.intel.com/compare/65510,65733,75048
Slowing down the newer i5 to match the E3 should provide some insight - if cache size is important, the performance should be close or better on E3, otherwise the i5 should be faster due to newer architecture. Also, the difference in CPU generations can be compared by matching the speeds of the i5 CPUs, seeing if the game runs about the same on both.

[Grimlocke]

Would be interested in seeing that, and got a 4930K and 4690K to add some results for too.

Question is, how to benchmark DF?

Testing fluids, world generation and general fortress mode mayhem seems sensible. The first two shouldn't be too hard. For fluids a simple arena map that moves a lot of water around should suffice, metric can be average stable framerate.

World generation can be any a pre-seeded world with a set end time, metric should be time until completion.

Fortress mode however is where it gets a bit tricky. You can just use any regular save of an older fortress (though having it called 'benchmarked' would be fun) and see how long it takes to run a season at unlimited framerate, but the RNG might throw it for a loop. Could be tested by just running the same scenario a bunch of times and seeing how much they vary.


All in all could be a fun little project.

[Antikythera]

from what i read the biggest workload is the pathfinding. so perhaps something that generates lots of pathfinding.

i'd also be happy to benchmark on several systems. i have a e5300, e8400, i7 920M, G3258, 3930k. could probably even get a P3 and P4 up and running.

[funkydwarf]

Oh !SCIENCE!....wait a minute..would it just be plane ol' real science? Or maybe !META-SCIENCE!

[BigD145]

Back with 40d an overclocked single-core P4 was running DF better than dual and quad cores within a couple years of 40d's release. Calculations per second per core had either not changed much or had dropped as the workload was being split between cores.

[deknegt]

Back with 40d an overclocked single-core P4 was running DF better than dual and quad cores within a couple years of 40d's release. Calculations per second per core had either not changed much or had dropped as the workload was being split between cores.

Pretty sure it's because DF isn't multi-threaded, this has been an issue for years and people bring it up on a pretty regular basis.
What this means is that every single process is done by one processor core. So that fancy i7 has just as much processing power than a single core pentium (if the ghz per core was identical)