Print

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

[Llamageddon]

This is a bit hypothetical, I have an i5-4690K (3.5 GHz) so I assume there is not much to be gained by a new CPU, but I am curious, as given the price of graphics cards at the moment I am getting into a lot of CPU intensive games like DF.

What would be the best CPU on the market for DF if money is no concern?

What would be the best upgrade from mine if on a budget?

And I suppose for others reading this, is there an even better budget option lower than my i5?

Are there any specifics to keep in mind for a game like DF? Would things like faster ram, SDD  etc make a bigger difference?

As I say, kind of hypothetical, but any insight into how DF makes use of these resources and processes things would also be quite interesting to me.

[Bumber]

http://www.bay12forums.com/smf/index.php?topic=105554.0
http://www.bay12forums.com/smf/index.php?topic=148409.0
http://www.bay12forums.com/smf/index.php?topic=157926.0

[Llamageddon]

Ah, thanks. I searched the forums but I get so many results that are completely unrelated I can never find anything I am looking for.

How did you find those posts so easily? Are there search parameters I don't know about?

[Bumber]

How did you find those posts so easily? Are there search parameters I don't know about?

I only found them because I knew they existed. Had to try a couple different terms.

[Rose]

Best is probably 8086, but if you can't get it, 486 will do in a pinch.

[wierd]

I somehow doubt that a 16bit cpu running between 5 and 10mhz is going to play DF that great.

[Llamageddon]

Hah, great recommendations Japa. I'll dust off the old 486 then. I'm sure everything will be fine... Does a 486 have a 1gb cache? Must do.

[Rose]

Close.

It has 8kb. Should be enough.

[Codyo]

In the end, I'd go with a CPU that has the highest clock speed possible. With the current market and overclocking with intel CPU's you should be able to get 4.5-5.0 Ghz But that's probably overkill for this game. In this day and age if you're getting the latest hardware, you shouldn't have to worry about the size of the caches people will mention.
I agree with what others say on this topic whenever it's brought up. Having the fastest RAM should make a difference as the world has to be loaded in the background for constant calculations each month. If you didn't play before the 0.40 versions, you wouldn't have seen what a titanic hit on performance "real time" history generation made during Fortress mode. RAM would help with this aspect of the game.

[wierd]

*Contemplates seeing if DF will run on a 486 using the linux version.

[mikekchar]

This might be a way to try it: http://bochs.sourceforge.net/  I haven't played with it at all, but I've heard good things about it.  I think the biggest problem will be RAM, but with enough virtual memory you might be able to do it.

[thompson]

General advice is a modern CPU with low latency RAM and a fast CPU to RAM memory bus.

The precise CPU isn't so straightforward. Naively you may expect a CPU with a large cache to be better, but this isn't necessarily true. Modern CPUs can actually anticipate the memory the data they need to fetch from the RAM and prefetch it to the cache before it is needed. The microarchitecture can also vary between CPUs with similar clock speeds, which can mean some CPUs can perform calculations with fewer clock cycles than others. Unfortunately, I don't know what sorts of optimizations Toady has applied (or which compiler and the settings of said compiler), so a general all-rounder CPU could be the safest bet. Edit: I should add that larger caches are slower as connections are physically larger. Whether a large slow cache works well for DF depends if CPU prefetching is effective. I have no idea if that is the case for DF.

One of the biggest microarchitectural differences is the length of the pipeline. It determines how far in advance the CPU had to fetch information. Deep pipelines are faster, but if you get a branch prediction failure there will be longer delays. It would be good to do some science on this to see if there is an optimal microarchitecture.

A good science project would be to create a standardized fort (let's call it DFSF2018), save it, then compare FPS across different CPU/RAM combinations. If we could get enough data we might be able to work out if any specific CPU architectures are better than others, and whether the need for low RAM latency can be mitigated by a CPU with better prefetching, if deep pipelines are better than shallow ones, and how important cache size is. Note: to do all that we'll need an absolute ton of data.

[Werdna]

There's likely to be some serious AMD vs Intel uP performance differences after things settle from Meltdown (Intel) and Spectre (both) fixes and OS workarounds.  I haven't been keeping up to date with it, maybe someone else here is better informed and can comment. 

I mention it because if you guys are going to do science with CPU/memory keep in mind that there are Meltdown and Spectre patches for the OS (and I think Windows already has issued several patches of the patches) that will likely change performance, so you may see different results depending on where you are in your updates.  I believe Spectre will result in mobo patches, so ditto there.

[Larix]

Just random empiric observation:
An Athlon 64 X2 with 3 GHz clock (pub. 2007) ran DF adequately, somewhat but not notably faster than a 2 GHz (mobile) Core2Duo. A mobile 7th gen. i3 gets about 40% higher FPS at 2.4 GHz clock. The important differences in favour of the i3 are: more cache (3,5 MB vs. 1 MB (nominally 2, but only one per core) on the AMD), more RAM bandwidth (DDR4/2133 vs. DDR2/800), something like 80% higher IPC. A fairly recent Intel Atom with 1 GB (DDR3/1600?)RAM can execute the game, but i wouldn't call it "run". All this to me suggests that a modern CPU with good single-thread performance, backed up with fast RAM should work well.

For all those joking about running DF on old systems: i tried out an older, probably 0.28.xx, version on a Celeron 633, running Windows Me (yes, Me; worked alright for me ) with all of 64 MB of PCI 66 MHz RAM, shared video memory. It ground horribly trying to show the intro movie, so i escaped out of that after five minutes of sporadic disorganised letters flying across the screen. The game itself did start and reached something like 2FPS at embark. I didn't play it for long. I've no idea whether that was just the issue of having to do tons of memory accesses at 66 MHz or if it went over the built-in memory and kept swapping stuff on and off the harddisk (a.k.a. pagefile).

[mikekchar]

I expect Spectre to chew a good 10% off of DF performance, unfortunately.  But who knows.  It really depends on the code.

It's interesting about the system comparison.  For people who are serious about FPS, I think you'll get better gains from being very careful about fortress design that you will from processor or RAM upgrades (assuming you are already on a reasonably modern CPU).  Or putting it differently, I think an expensive computer won't really appreciably save you from bad fortress design.  It would be interesting to see if I'm right, though.  I'd love to see a challenge for maxing FPS on a productive fortress of 100 dwarfs, opened caverns, full vegetation on the surface and temperature turned on.  I'll bet there would be examples with almost an order of magnitude difference.