Print

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

[vomov]

I've been running DF on my work PC (from my dropbox), and it runs awesomely fast compared to my beast of a home PC. I've got a year-old fort with a standard embark and 80 dwarfs.

Comparison:

Code: [Select]

PC 1 (home PC): x64 windows 8.1
AMD A8-6500 APU @ 3.8GHz, quad core (and a Radeon 5800 as the videocard, I think)
8GB RAM (fsb of 2400)
Running DF of an OCZ Agility
Aside from windows defender and Chrome, nothing is active.
FPS: 30


Code: [Select]

PC 2 (work PC): windows 7
Intel Pentium 4 @ 2.4GHz, single core
2GB RAM (fsb of 800)
Running DF of an generic hard drive
Note: also running matlab, which is happily churning away at some large datasets. DF while waiting.
F-secure virusspammer checks everything twice, Chrome is active, and a dozen weird IT-utilities that are required on our stupid corporate network
FPS: 110

I've found earlier threads about the AMD/ATI openGL options, and fiddled with the drawing options on both sides. I don't think these options make that much difference (+2 for one option, -1 for another).

So, this issue seems to be related to the type of processor; DF quite simply runs better on Intel.
If this is correct, it would mean that I'm not the only one seeing this, and I can find some old threads hinting to it.
It would be better to have some more recent reports, so this thread asks you to share your setups and your framerates (ideally the same fort on different PC's), so we can find out if Intel is really better for !!FUN!!

[Thief^]

Really you'd need a standard savegame and settings if you want to compare performance across PCs, people posting framerates from different saves really won't help.

[vomov]

A single savegame is not a proper indication of DF; the game is too varied. I'd rather have people work with their own savegames, which could provide a more broad approximation of DF performance.

The idea is for people who have access to two (or more) PC's to move their own savegame between them, and reporting any odd differences. I'd love for someone to confirm the AMD/Intel discrepancy.

[Orange Wizard]

Mine seems to cap out at two hundred and forty-four thousand on a crappy Intel laptop.

Such are the virtues of Thief's ascertation.

[utunnels]

Also limit gfps to a reasonable low value, turn off vsync, do not use graphics pack, etc.

[vomov]

Before we get a boatload of single pc configurations with a fps, and no way to figure out what causes the difference; the idea is to move your whole DF (the whole folder, including the save, settings, etc.) to a different PC, and report the performance and specs of both.

[Bloax]

Grab 7-Zip, compress the folder

and upload it for us to compare.

[vomov]

A single savegame is not a proper indication of DF; the game is too varied. I'd rather have people work with their own savegames, which could provide a more broad approximation of DF performance.

Ideally, one would like a single 'test-fortress', but due to the massive capabilities of DF, I would argue this is not only unpractical, but impossible. A more reliable measure would be to take actual forts by actual players, and then vary the hardware. I'm willing to upload my fort, but then we're substituting a theoretical 'standard fortress' with my crappy fortress, and we'll still only have a single datapoint.

Best practice would be to upload hundreds of test fortresses, and then methodically varying the hardware, but I don't think that we'll get a lot of people loading fort after fort to note the FPS.

Am I the only one why has access to multiple PC's or something?

[Footkerchief]

A single test fortress won't work, but a set of test fortresses seems necessary.  Letting people run their own fortresses introduces possible confounding factors, e.g. people on slower computers might compensate by running less demanding fortresses.

I don't have any AMD PCs.

Honestly, it might be hard to get any rigor in the test unless you can create a standardized benchmarking package (maybe using DFhack to capture FPS information?)  Other games have that (with a variety of preset scenarios), DF could too.  Otherwise, there are just confounding variables all over the place (if the methodology is just 1. run game 2. "quickly" write down your impression of average FPS).

[Bloax]

Footkerchief: You do know that DF has a built-in FPS counter, right?

Well I don't play fortress mode and all, but I'd gladly do performance measurements on my PC.
Speaking of which:

Code: [Select]

Win 7 x64
Intel Core 2 Duo E8600 @ 3.66GHz, GeForce 9800 GT (yeah this is a 2008 PC)
4.00GB Dual-Channel DDR2 @ 365MHz (4-4-4-10) [Thanks Speccy]
Running DF on a RAM drive.
No notable background processes.

[Footkerchief]

Footkerchief: You do know that DF has a built-in FPS counter, right?

Yes, and it fluctuates over time.  It has to be collected for a period of time and then averaged.  How many samples over how long a period?  These questions matter, because unless your method for collecting that data is 100% consistent, you'll introduce bias.  For example, one might subconsciously pick a lower number as the "average" because they're already expecting the system to be slow.  Also, fortress events can change the FPS, which reinforces the need for a consistent measurement period.

[Bloax]

Ah yes, it was just an issue of vague formulation.

[vomov]

Does the built-in FPS counter have accuracy issues?

When working on large, complicated data, one generally tries to minimize influence of unknown factors.
For example: when trying to figure out the side-effects of a new drug, you'd want to feed the drug to only healthy people. Ideally, you'd even test these people for conditions etc. they themselves might not be aware of. (sidenote; ethics do not normally allow this)
You'll get very clean data, and produce some very pretty clear pictures of what's going on.

The problem of this type of experimentation is that you'll only find results that are valid for the group of healthy individuals, which can mean your results don't apply to any other group. There've been multiple situations in which this has become quite clear.

In testing DF, the definition of 'healthy'... doesn't really apply in the dictionary way. Even worse, the variability between playstyles and the resulting fortresses (and the nature of the computational loads etc.) makes a single test fortress impossible, and would require a lot of test fortresses.

This is sometimes seen in medical research; variation between subjects is too big, and comparing them becomes impossible. In some cases there's a very easy way to compensate for this using difference scores. You expose a subject to two conditions, like a medication and a placebo (not at the same time), and note the differences between the two outcomes.

Since a computer usually doesn't display a placebo-effect, I would say this is a valid and easy way to track some big issues, like processor architecture, OS, RAM speed, etc. Simply said: you simply load the same save on both PC's, and note the FPS on both.

Yes, and it fluctuates over time.  It has to be collected for a period of time and then averaged.

Ideally, yes, but then you're introducing the time as a factor, which needn't be one. FPS doesn't fluctuate that much if a fort is stable (assuming you let the game run for a bit after loading), and big changes (breaching the caverns, releasing the clowns, etc) require a re-measurement anyway. I would even say that you can treat them as separate forts, for testing purposes, since the load is of a different variety.

However, I'm starting to think that setting up a set of 'volunteer testing fortresses' and 'volunteer testing systems' is starting to get inevitable, if I want to see issues like these brought to light...

[Footkerchief]

Even worse, the variability between playstyles and the resulting fortresses (and the nature of the computational loads etc.) makes a single test fortress impossible, and would require a lot of test fortresses.

Yes.

Ideally, yes, but then you're introducing the time as a factor, which needn't be one. FPS doesn't fluctuate that much if a fort is stable (assuming you let the game run for a bit after loading), and big changes (breaching the caverns, releasing the clowns, etc) require a re-measurement anyway. I would even say that you can treat them as separate forts, for testing purposes, since the load is of a different variety.

The thing is, unless you're taking precisely timed measurements, time is already a factor.  "Assuming you let the game run for a bit" means, in experimental terms, "let your silly human brain find the moment that best satisfies its biases."

[miauw62]

A single savegame is not a proper indication of DF; the game is too varied. I'd rather have people work with their own savegames, which could provide a more broad approximation of DF performance.

Yes, but a higher FPS can indicate two things: a better CPU or more efficient fort design. Using a standard fortress is completely valid when measuring FPS. Then again, at that point you could basically compared the clock rates of a single core of your cpus. actually theres probably other factors that come into play too like cache size (and what other programs you are running in the background and your OS etc)