Print

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

[Dirst]

I'm not sure where I heard this, but the bottleneck is supposed to be memory bandwidth.  Footkercheif's profiler probably has the answer.

Memory bandwidth itself has a couple different opportunities for bottlenecks.  The most obvious is the speed of the RAM sticks in your machine, but the memory bus and the caches come into play as well.  CPU would become binding only if someone built a ginormously overpowered L2 cache that held the entire world in it.

[guessingo]

on chip cache increases with every new generation of chips... but in rather small increments. so it may be a while. Eventually I'd think the whole game world will fit in cache, but eventually is a long time.

[Putnam]

Maybe we will get lucky and cpu clock speeds will increase again...

Clock speeds are worthless as a measure of CPU speed.

[malvado]

Maybe we will get lucky and cpu clock speeds will increase again...

Clock speeds are worthless as a measure of CPU speed.

Well, in a Modern Cpu they can at least give an indication of overall performance if it comes from the same producer and has the same architecture , example Intel 4xxx series or Amd FX +83xx series. Sure you also have to put into consideration IPC etc , but clock speeds are essentially an easy way to give you an indication of the performance standpoint of a cpu in a given architecture.
So I wouldnt go as far as to call them worthless , in fact whats worthless is probably the user who doesnt understand how they fit inn with the overal performance of that line of cpu's.

[Putnam]

I'm not sure where I heard this, but the bottleneck is supposed to be memory bandwidth.  Footkercheif's profiler probably has the answer.

Memory bandwidth itself has a couple different opportunities for bottlenecks.  The most obvious is the speed of the RAM sticks in your machine, but the memory bus and the caches come into play as well.  CPU would become binding only if someone built a ginormously overpowered L2 cache that held the entire world in it.

Yep. Probably the reason for the slowness of the game.

[smjjames]

I'm not sure where I heard this, but the bottleneck is supposed to be memory bandwidth.  Footkercheif's profiler probably has the answer.

Memory bandwidth itself has a couple different opportunities for bottlenecks.  The most obvious is the speed of the RAM sticks in your machine, but the memory bus and the caches come into play as well.  CPU would become binding only if someone built a ginormously overpowered L2 cache that held the entire world in it.

Yep. Probably the reason for the slowness of the game.

So, it's not so much the CPU or RAM (unless it's an older computer) than it's the way the algorithm and data structures are, well, structured?

No idea how much Toady One knows about how to do that stuff, but it might require a second party to help out, and we know how much he dislikes giving the code to someone else to work on. Theres definetly various programmers on the forums that could volunteer, or maybe he could just use that profiler/compiler thing.

[Putnam]

No, definitely the RAM. It's less about the data structure talk and more about the memory statistics, heh. I'm fairly sure there isn't much use of discontiguous data structures in DF's code.

Besides that, the profiler is your friend. If you must optimize, trust it over youtube videos or whatever.

[smjjames]

Speaking of profiler, Dirst said that Footkerchief's profiler probably has the answer, but I don't see any profiler log anywhere.

[Dirst]

Speaking of profiler, Dirst said that Footkerchief's profiler probably has the answer, but I don't see any profiler log anywhere.

Footkerchief mentioned "trust the profiler" on a couple occasions, so I was gently prodding him for a link to or digest of the results.

[Ives]

The world simulation SEEMS pretty segregated from fort / adventure mode, and you may be able to offload that to a thread.

This assumption will become increasingly unrealistic as Toady adds more ability for the fortress to interact with the world.

I don't think so, because it wouldn't make sense for the fortress to be able to interact with the world at any given time and at random intervals. If a single day is the basic "frame" of world calculations, then it could absolutely be done in a different thread and communications between the world and the fortress would only happen at the start of each day. Since it's pretty much a given that world calculations for one day on core 1 would finish before fortress calculations for one day on core 2, you should only have a negligible amount of "lag" when the two threads touch bases.

I don't think world simulation is contributing a lot to FPS drops though, the way it is right now. It's definitely a good idea to do it in a separate thread, because it's going to get a whole lot more complex before actually making a meaningful impact on the fortress mode.

[Footkerchief]

Footkerchief mentioned "trust the profiler" on a couple occasions, so I was gently prodding him for a link to or digest of the results.

I don't have access to profiler results.  When I say "trust the profiler," what I mean is that intuition and speculation are nearly worthless.