Official DOSBox Settings Discussion Topic

Edit history:

LotBlind: 2023-03-27 12:55:35 pm

LotBlind: 2023-03-27 12:50:02 pm

2013-08-19 11:27:48 am

As recently announced, DOSBox is generally accepted for SDA submissions for all games that don't run on modern PCs. The default settings of DOSBox work for a lot of games, but are not optimized for individual titles or may lead to playback issues. This topic is designed as a central discussion space for DOSBox settings, both for just playing the games and most importantly recording them for SDA.

For games that run too fast at default or max cycles settings a reasonable value needs to be agreed on by people familiar with the game. If game run speed is not affected by the cycles settings, "max" is suggested as it tends to reduce load times and thus increases entertainment when watching the run. Loadtimes will still be cut from SDA timing as per the usual timing rules for PC games.

Some old games run faster with sound disabled. Games that have an audio option are recommended to be ran with sound on, even if it slows the game down a little bit. Whether the sounds can be left off or not is a case-by-case decision so make sure to confirm it first before submitting.

Note that while this is the official discussion topic for DOSBox settings, approval of settings in here are not necessarily binding for the verification process. The final say always lies with the verifiers for any given run submitted.

~~For the time being, this topic will be maintained by Xarthok as he has a lot of experience with running DOS games on DOSBox~~.

The new unofficial person in charge of this topic is tigrou, who has written this post below for a detailed breakdown of how DOSBox settings affect game speeds and fairness of competition. If you have any input or differing opinions, ask away.

Thread title:

As [url=https://forum.speeddemosarchive.com/post/virtualization_software_and_dosbox_allowed_for_old_games.html]recently announced[/url], DOSBox is generally accepted for SDA submissions for all games that don't run on modern PCs. The default settings of DOSBox work for a lot of games, but are not optimized for individual titles or may lead to playback issues. This topic is designed as a central discussion space for DOSBox settings, both for just playing the games and most importantly recording them for SDA.

For games that run too fast at default or max cycles settings a reasonable value needs to be agreed on by people familiar with the game. If game run speed is not affected by the cycles settings, "max" is suggested as it tends to reduce load times and thus increases entertainment when watching the run. Loadtimes will still be cut from SDA timing as per the usual timing rules for PC games.

Some old games run faster with sound disabled. Games that have an audio option are recommended to be ran with sound on, even if it slows the game down a little bit. Whether the sounds can be left off or not is a case-by-case decision so make sure to confirm it first before submitting.

Note that while this is the official discussion topic for DOSBox settings, approval of settings in here are not necessarily binding for the verification process. The final say always lies with the verifiers for any given run submitted.

[s]For the time being, this topic will be maintained by Xarthok as he has a lot of experience with running DOS games on DOSBox[/s].

The new unofficial person in charge of this topic is tigrou, who has written [url=https://forum.speeddemosarchive.com/post/official_dosbox_settings_discussion_topic_23.html]this post[/url] below for a detailed breakdown of how DOSBox settings affect game speeds and fairness of competition. If you have any input or differing opinions, ask away.

Xarthok

LotBlind

Edit history:

LotBlind: 2020-07-17 03:48:32 pm

LotBlind: 2018-04-19 12:40:32 pm

UraniumAnchor: 2014-04-04 01:34:24 pm

Xarthok: 2013-08-26 09:31:56 am

Xarthok: 2013-08-19 01:39:34 pm

Xarthok: 2013-08-19 01:39:31 pm

Xarthok: 2013-08-19 01:39:30 pm

Xarthok: 2013-08-19 01:03:29 pm

Xarthok: 2013-08-19 12:57:44 pm

Xarthok: 2013-08-19 12:56:16 pm

Xarthok: 2013-08-19 12:54:08 pm

Xarthok: 2013-08-19 12:51:29 pm

Xarthok: 2013-08-19 12:49:18 pm

Xarthok: 2013-08-19 12:47:44 pm

Xarthok: 2013-08-19 12:36:54 pm

Xarthok: 2013-08-19 12:29:17 pm

Xarthok: 2013-08-19 12:27:11 pm

Xarthok: 2013-08-19 12:17:33 pm

Xarthok: 2013-08-19 12:13:45 pm

Xarthok: 2013-08-19 12:09:40 pm

Xarthok: 2013-08-19 12:02:35 pm

Xarthok: 2013-08-19 12:01:58 pm

Xarthok: 2013-08-19 11:55:58 am

Xarthok: 2013-08-19 11:54:44 am

LLCoolDave: 2013-08-19 11:29:00 am

2013-08-19 11:28:05 am

DOSBox <3

I'll start with a list of games I'm familiar with and DOSBox settings to use with them. Usually the only setting that needs alteration if any is the "cycles=" value in the conf file. In most cases max cycles will work best but even in those cases I want to list those games here for people who want the acceptance of that clarified. In rare cases we would also have to tweak audio settings for better emulation. Any additions to the list are more than welcome.

Alien Carnage - max cycles (the game runs significantly slower on auto cycles so this can be debated)
Alone in the Dark - 11.000 cycles (GOG)
Arctic Adventure - auto (3000) cycles
Bio Menace - max cycles
Commander Keen I, II, III - max cycles
Commander Keen IV, V, VI, Dreams - max cycles (for faster load times)
Crystal Caves - max cycles
Dangerous Dave I, II, III, IV - max cycles
ElectroMan - max cycles
Guimo - "fixed 20000" cycles seems to be most playable and close to the Windows port
Jazz Jackrabbit - auto (3000) cycles
Jetpack - auto (3000) cycles
Jill of the Jungle - auto (3000) cycles (sound seems to be screwed up on max)
Little Big Adventure - "fixed 80000" cycles (according to GOG)
Prehistorik 2 - auto (3000) cycles (cutscenes are slightly longer but overall emulates better than on max)
Secret Agent - max cycles
Time Gate: Knight's Chase - fixed 80000 cycles
Titus the Fox/Moktar - auto (3000) cycles (game runs slightly slower but sound sounds much better than on max)

The "core" setting should probably be left to "auto" unless this causes instability. In practice, this means the game will be run on the "dynamic" core if possible, which makes it run faster without conferring an unfair advantage to those with faster computers.

I'll start with a list of games I'm familiar with and DOSBox settings to use with them. Usually the only setting that needs alteration if any is the "cycles=" value in the conf file. In most cases max cycles will work best but even in those cases I want to list those games here for people who want the acceptance of that clarified. In rare cases we would also have to tweak audio settings for better emulation. Any additions to the list are more than welcome.

[b]Alien Carnage[/b] - max cycles (the game runs significantly slower on auto cycles so this can be debated)
[b]Alone in the Dark[/b] - 11.000 cycles (GOG)
[b]Arctic Adventure[/b] - auto (3000) cycles
[b]Bio Menace[/b] - max cycles
[b]Commander Keen I, II, III[/b] - max cycles
[b]Commander Keen IV, V, VI, Dreams[/b] - max cycles (for faster load times)
[b]Crystal Caves[/b] - max cycles
[b]Dangerous Dave I, II, III, IV[/b] - max cycles
[b]ElectroMan[/b] - max cycles
[b]Guimo[/b] - "fixed 20000" cycles seems to be most playable and close to the Windows port
[b]Jazz Jackrabbit[/b] - auto (3000) cycles
[b]Jetpack[/b] - auto (3000) cycles
[b]Jill of the Jungle[/b] - auto (3000) cycles (sound seems to be screwed up on max)
[b]Little Big Adventure[/b] - "fixed 80000" cycles (according to GOG) 
[b]Prehistorik 2[/b] - auto (3000) cycles (cutscenes are slightly longer but overall emulates better than on max)
[b]Secret Agent[/b] - max cycles
[b]Time Gate: Knight's Chase[/b] - fixed 80000 cycles
[b]Titus the Fox/Moktar[/b] - auto (3000) cycles (game runs slightly slower but sound sounds much better than on max)

The "core" setting should probably be left to "auto" unless this causes instability. In practice, this means the game will be run on the "dynamic" core if possible, which makes it run faster without conferring an unfair advantage to those with faster computers.

Xarthok

LLCoolDave

Edit history:

LLCoolDave: 2013-08-19 11:29:16 am

2013-08-19 11:28:09 am

DOSBox <3

Reserved

CrimsonxxMana

2013-08-23 05:33:18 pm

GOG puts Little Big Adventure at 80,000 cycles. The cycles on the gameplay really only change responsiveness on the controls. As such changing between the different states the character can use will be faster or slower depending on how many cycles you are using. I have just used the default cycles of 80,000 GOG provided. I have no idea what cycles the segmented run was ran on. Or if it was the original DOS mode. But I think they can't be much different from the GOG settings provided.

Phoenix3568

Edit history:

Phoenix3568: 2013-08-24 12:30:21 am

2013-08-24 12:26:00 am

Procrastination Nation!

@Xarthok thank you. I've been wanting to run JotJ for a while (Jill) and am glad that this is now possible.

I believe Jazz Jackrabbit runs well on auto (3000) cycles as well. Please confirm.

Other games that could be run:
Duke Nukem 1/2 + episodic content

foob

2013-09-26 01:31:31 pm

could i ask whether or not i matter which platform you run DOSBox on?

I ask because i have recently posted a thread regarding runs using a Raspberry Pi for emulation (which allows gamepads and direct SCART output given the right cabling) and the method used to build my raspberry pi console (retropie) has DOS emulation built in. I'm fairly certain it achieves this by using an ARM port of DOSBOX. It was never something i originally intended to use but seeing as you are correct in your statement that finding dedicated machines capable of running old dos games is very difficult nowadays, Raspberry Pi's could offer a modern solution to that problem, with very little cost. Taking it a step further, you could easily produce an official "config" for download use on it, and regulate settings that way.

Just a thought.

ShadowWraith

2013-09-29 11:08:20 am

.

The problem is finding machines that can run DOS effectively, not DOSBox. I'm not sure how close the emulation is on the Raspberry Pi, but if it's as accurate as PC DOSBox is, I don't see why you couldn't use it.

Ash

2014-01-19 07:37:42 pm

Is there a rule or consensus about using DOSBox with games that have an "unbound" speed setting? I'm planning a run of Police Quest 1 (the original AGI version), which has four in-game speed settings; the fastest of these is semi-unbound ("semi" because the speed setting only affects on-screen action; it has no effect on timed event triggers, of which there are many in this game). I've been practicing with the fastest in-game speed at the default DOSBos setting (3000 cycles). However, members on this thread have hinted that "unbound" speed settings might not be permitted. I'm hoping to get some confirmation one way or the other about this.

SamStrife

2014-01-20 01:00:02 am

Trying so hard

Quote from CrimsonxxMana:

GOG puts Little Big Adventure at 80,000 cycles. The cycles on the gameplay really only change responsiveness on the controls. As such changing between the different states the character can use will be faster or slower depending on how many cycles you are using. I have just used the default cycles of 80,000 GOG provided. I have no idea what cycles the segmented run was ran on. Or if it was the original DOS mode. But I think they can't be much different from the GOG settings provided.

As someone running the original Little Big Adventure currently, I've come to learn the the GOG version of the game runs markedly slower. I couldn't for the life of me work out why I was having flawless sections be so much slower than WR pace. When I was forced to use the original due to a bug, that would have cost nearly 30 seconds, in the GOG version, I realised that even with sloppy play, I was beating my PB's handily.

I need to look to the case as to why this is. Is it a GOG problem? Is the original version of the game with the Windows XP patch a better version or is there a default setting that I haven't looked into making it seem way better for me? I will investigate.

[quote="CrimsonxxMana"]GOG puts Little Big Adventure at 80,000 cycles. The cycles on the gameplay really only change responsiveness on the controls. As such changing between the different states the character can use will be faster or slower depending on how many cycles you are using. I have just used the default cycles of 80,000 GOG provided. I have no idea what cycles the segmented run was ran on. Or if it was the original DOS mode. But I think they can't be much different from the GOG settings provided.[/quote]

As someone running the original Little Big Adventure currently, I've come to learn the the GOG version of the game runs markedly slower. I couldn't for the life of me work out why I was having flawless sections be so much slower than WR pace. When I was forced to use the original due to a bug, that would have cost nearly 30 seconds, in the GOG version, I realised that even with sloppy play, I was beating my PB's handily.

I need to look to the case as to why this is. Is it a GOG problem? Is the original version of the game with the Windows XP patch a better version or is there a default setting that I haven't looked into making it seem way better for me? I will investigate.

LotBlind

2014-01-24 11:55:05 am

...while we're at it, someone could fix the "Little Big Planet" in Xarthok's list.

SamStrife

2014-01-29 02:22:16 am

Trying so hard

Quote from LotBlind:

...while we're at it, someone could fix the "Little Big Planet" in Xarthok's list.

Haha far too many people make that mistake when it comes to LBA. It's a real shame.

CapnClever

2014-04-04 10:12:12 am

I've been running Commander Keen games for a while now, and when I looked into cycles I decided that it made sense to find a "fixed" cycle value in order to ensure that gameplay is reproducible. The problem I have with "max" cycles is that it's hardware-dependent: even if most games have a limit on how well they'll end up performing (for the sake of running my only concern is minimizing load times), finding a value that reaches that limit is more consistent than assuming max cycles will always reproduce the same results.

So with that in mind, I have the Keen games when I play set to "fixed 60000". As far as I can tell, Steam's recommended settings (auto, which is a fixed 3000 for Keen games) are exactly the same when it comes to gameplay, and upping this value only makes the levels load faster (and only for games after Keen 1-3, I could probably leave it at 3000 for those and no timing differences would occur). I'm just wondering what people think of it: I already have a few good runs but I didn't think to actually discuss the cycles I decided upon, so if I have to redo 'em that's on me anyway.

I've been running Commander Keen games for a while now, and when I looked into cycles I decided that it made sense to find a "fixed" cycle value in order to ensure that gameplay is reproducible. The problem I have with "max" cycles is that it's hardware-dependent: even if most games have a limit on how well they'll end up performing (for the sake of running my only concern is minimizing load times), finding a value that reaches that limit is more consistent than assuming max cycles will always reproduce the same results.

So with that in mind, I have the Keen games when I play set to "fixed 60000". As far as I can tell, Steam's recommended settings (auto, which is a fixed 3000 for Keen games) are exactly the same when it comes to gameplay, and upping this value only makes the levels load faster (and only for games after Keen 1-3, I could probably leave it at 3000 for those and no timing differences would occur). I'm just wondering what people think of it: I already have a few good runs but I didn't think to actually discuss the cycles I decided upon, so if I have to redo 'em that's on me anyway.

UraniumAnchor

2014-04-04 01:34:23 pm

Not a walrus

If it's just load times then SDA discounts those anyway, so you should still just set it to 'max', unless I'm misunderstanding something.

ShadowWraith

2014-04-05 05:33:57 pm

.

I'd prefer if you used the Steam values for that, as while having a better CPU doesn't cause ridiculous increases in gamespeed in the Keen series like some other DOS games, it makes it easier for us to process (hence why the rules are specific here), and also since there's no timing difference there's no point in running it at higher values than the Steam ones other than being a special snowflake.

CapnClever

Edit history:

CapnClever: 2014-04-08 07:05:22 pm

2014-04-08 07:04:46 pm

Okay finally got around to checking Steam values, which are as follows:

Keen 1-3: auto (equivalent to fixed 3000 in this case)
Keen 4-5: fixed 30000

I think I switched to 60000 because it was sliiiightly faster (like maybe three frames and only when entering levels) in terms of load times, but I have no qualms switching back. I'm just glad it didn't end up being auto for Keen 4-5 because that would've ended up with rather large loading time changes.

The Steam releases are run using DOSBox 0.71, which became obsolete before DOSBox/VMs were acceptable on SDA (current version is 0.74). That said, should I be running it exactly from the Steam copy, or is it enough to replicate the configuration files and use the latest? Also, Keen6 and Keen Dreams aren't part of the Steam collection: should I use the Keen 4-5 settings (including DOSBox version 0.71) for those since they're basically the same engine?

ShadowWraith

2014-04-08 09:44:36 pm

.

You're allowed to use the latest version of DOSBox, just make sure the settings are consistent. Yes, you can use the same settings for Keen6/Keen Dreams. (you can use whichever you want for those two as long as the game behaves properly, as there's no precedent set there.)

CapnClever

2014-04-09 12:24:38 am

Okay, first thing, I'm an idiot who forgot he changed those cycle settings for Keen4/5 back when he originally downloaded them. Which means those games are also set to 3000 cycles, which means they're a great deal slower than I would want.

Second, to show you what I mean, I quickly made two videos (both are shorter than five minutes and I recommend looking at least halfway through on both) to show the difference between the Steam-default (auto/3000) and max. (Forget that I mentioned 60000 cycles, I've been using max since I first asked and haven't had any issues I thought I had before.)

Keen5 any% w/3000 cycles
Keen5 any% w/max cycles

Other than the obvious, I noticed that the 3000cycle run seemed to contain occasional visual desyncing that carried over to the recording, which isn't good quality-wise. There are some unexpected differences in buffering inputs, as well. I own a 20-year-old PC that runs DOS on it and I can tell you it's not nearly as slow playing Keen 5 as the 3000cycle video.

So I guess what I'm asking is if we can make max cycles the de-facto standard for Keen, or at least Keen games with the later engine. I'm almost positive all the current runners use max (or similar, like I was doing with 60000) in their runs based on what loading times look like. Maybe I should've just cited the runs already here on SDA, which are clearly closer to max cycles than they are 3000.

PS sorry for making a mess out of getting the right settings, I haven't kept my config files especially tidy.

Okay, first thing, I'm an idiot who forgot he changed those cycle settings for Keen4/5 back when he originally downloaded them. Which means those games are also set to 3000 cycles, which means they're a great deal slower than I would want.

Second, to show you what I mean, I quickly made two videos (both are shorter than five minutes and I recommend looking at least halfway through on both) to show the difference between the Steam-default (auto/3000) and max. (Forget that I mentioned 60000 cycles, I've been using max since I first asked and haven't had any issues I thought I had before.)

[url=http://youtu.be/jIEdhSDMnEg]Keen5 any% w/3000 cycles[/url]
[url=http://youtu.be/v3HjJmlC8xk]Keen5 any% w/max cycles[/url]

Other than the obvious, I noticed that the 3000cycle run seemed to contain occasional visual desyncing that carried over to the recording, which isn't good quality-wise. There are some unexpected differences in buffering inputs, as well. I own a 20-year-old PC that runs DOS on it and I can tell you it's not nearly as slow playing Keen 5 as the 3000cycle video.

So I guess what I'm asking is if we can make max cycles the de-facto standard for Keen, or at least Keen games with the later engine. I'm almost positive all the current runners use max (or similar, like I was doing with 60000) in their runs based on what loading times look like. Maybe I should've just cited the runs already here on SDA, which are clearly closer to max cycles than they are 3000.

PS sorry for making a mess out of getting the right settings, I haven't kept my config files especially tidy.

Bigmanjapan

Edit history:

Bigmanjapan: 2014-10-03 09:34:56 am

Bigmanjapan: 2014-10-03 05:13:35 am

Bigmanjapan: 2014-10-03 04:37:18 am

Bigmanjapan: 2014-10-02 07:59:23 am

Bigmanjapan: 2014-10-02 05:19:37 am

Bigmanjapan: 2014-10-02 05:11:13 am

2014-10-02 05:09:41 am

Is there any standard which resolution game should be played on when using DOSBox?
The game is Kyrandia 2. Original resolution is 320x200, but I'm comfortable playing with the game upscaled to 1280x800. So the output video is 1280x800 also.

presjpolk

2014-10-03 09:15:02 am

HELLO!

Is it stretched at all at that resolution? If you strictly upscaled it 4x it would be 1280x800.

Bigmanjapan

2014-10-03 09:33:48 am

My bad, 1280x800 indeed it is. Confused it with another game DOSBox settings.

Crow!

2014-11-07 06:00:20 pm

What's that gemma?

I've been running Quest for Glory 1 (VGA Remake). This is a game whose speed is very, very greatly changed by the CPU speed, and I'd like some input in picking Dos Box's cycles setting.

The game lists a 286 processor as the system minimum requirement; faster processors than those were certainly contemporary with the game's release. This places reasonable selections anywhere between 1k - 30k cycles.

The game just looks silly at 10k+ cycles and the fastest in-game speed setting. Running across a screen takes maybe 0.1 seconds; getting a click in at all in rooms where your character doesn't automatically stop is sometimes difficult.

Speed running the game is very fundamentally different between the two extremes. At ~3k or lower lower cycles settings, it's about carefully setting up to precisely input what you need, and not anything else, each time the game finishes lagging. Each input causes more lag, which gives you some time to set up the next input. At higher cycles settings, you're limited almost exclusively by how fast and accurately you can click, rather than how fast or slow various actions in the game are supposed to be in relation to each other.

I see that Xarthok lists a very large number of games at 3000 cycles. I've been using that amount so far, and it runs a lot like how I remember the game running as a kid before we upgraded our computer to a 486 (at which point it worked more like the 10000 cycles case).

I've been running Quest for Glory 1 (VGA Remake).  This is a game whose speed is very, very greatly changed by the CPU speed, and I'd like some input in picking Dos Box's cycles setting.

The game lists a 286 processor as the system minimum requirement; faster processors than those were certainly contemporary with the game's release.  This places reasonable selections anywhere between 1k - 30k cycles.

The game just looks silly at 10k+ cycles and the fastest in-game speed setting.  Running across a screen takes maybe 0.1 seconds; getting a click in at all in rooms where your character doesn't automatically stop is sometimes difficult.

Speed running the game is very fundamentally different between the two extremes.  At ~3k or lower lower cycles settings, it's about carefully setting up to precisely input what you need, and not anything else, each time the game finishes lagging.  Each input causes more lag, which gives you some time to set up the next input.  At higher cycles settings, you're limited almost exclusively by how fast and accurately you can click, rather than how fast or slow various actions in the game are supposed to be in relation to each other.

I see that Xarthok lists a very large number of games at 3000 cycles.  I've been using that amount so far, and it runs a lot like how I remember the game running as a kid before we upgraded our computer to a 486 (at which point it worked more like the 10000 cycles case).

LotBlind

2018-04-19 12:39:34 pm

For the record, Alone in the Dark is run using the official GOG settings of 11.000 cycles, which is a compromise between speed and playability (running becomes impossible if it runs too fast)... meaning lag-reducing strats are called-for.

Also looks like some questions here may have been left unanswered so feel free to post again if need be.

tigrou

Edit history:

tigrou: 2023-03-28 11:46:02 am

tigrou: 2023-03-28 03:12:04 am

tigrou: 2023-03-28 02:37:16 am

tigrou: 2023-03-27 08:22:29 am

tigrou: 2023-03-27 08:13:23 am

tigrou: 2023-03-27 07:14:04 am

tigrou: 2023-03-27 07:06:57 am

tigrou: 2023-03-27 06:16:12 am

tigrou: 2023-03-27 06:08:47 am

tigrou: 2021-12-31 12:45:33 pm

tigrou: 2021-12-16 08:25:51 am

tigrou: 2021-12-16 08:24:52 am

tigrou: 2021-12-13 03:40:21 pm

tigrou: 2021-12-13 12:10:27 pm

tigrou: 2021-12-13 08:50:18 am

tigrou: 2021-12-13 08:49:22 am

tigrou: 2021-12-13 08:43:35 am

tigrou: 2021-12-12 06:27:18 pm

tigrou: 2021-12-12 06:25:59 pm

tigrou: 2021-12-12 06:24:22 pm

tigrou: 2021-12-12 06:21:29 pm

tigrou: 2021-12-12 06:18:11 pm

tigrou: 2021-12-12 06:16:52 pm

tigrou: 2021-12-12 06:16:31 pm

tigrou: 2021-12-12 06:14:57 pm

2021-12-12 06:04:37 pm

Hi there. Here is some "hard to find" information about DOSBox settings I collected. It's based on some personal research and discussion with LotBlind :

cycles

fixed [number] : how much CPU cycles DOSBox will execute every millisecond.
That value shouldn't be too high (eg: 2000000) as it can create trouble (eg: input lag, choppy sound). For CPU demanding games, auto/max should be used.

auto : it will try to max out the number of CPU cycles used for the game (to put it as high as possible) while keeping some safe margin for DOSBOX itself.
By default, 90% of the CPU time will be spent for running game code, and there is always 10% kept for DOSBox internals (eg: rendering screen, sound, I/O, ...).

It's only enabled for games running in protected mode (eg: DOS/4G). For games running in real mode, the number of CPU cycles will be set to 3000. The idea behind this is that games running in real mode are usually old games.
Most of those games are running too fast on fast CPU.

max : same thing as above but it will max out cycles even for games running in real mode.

max [percentage] / auto [percentage] :
It's possible to fine tune how much time is used by the game vs DOSBOX internals by pressing CTRL-F11 / F12
How much CPU is allocated for the game is calculated this way : percentage * 0.9
At 100% game will receive 100 * 0.9 = 90% of CPU (DOSBox has 10% left for itself)
At 105% game will receive 105 * 0.9 = 94.5% (max value)
Minimum value is 1%.

core

normal : game is interpreted instruction by instruction. slow, but accurate.
dynamic : instructions are converted to host CPU instructions before execution which result in a much better performance. less accurate : DOSBox might execute more or less instructions than what you requested. not supported on all platforms. might crash some games.
simple : same as normal but optimized for real mode games. It's faster than normal core because functionalities related to protected mode (eg: memory paging) are not emulated.
auto : switch to dynamic if available and if game is running in protected mode. otherwise normal is used

cputype

should be set to auto. Other settings are used to reproduce some specific CPU models (eg: with 386_slow, some instructions or features will be not emulated).
It might be useful for running some very old games that work only with a specific CPU model.

When speed running old games, it's more fair to force players to use cycles=fixed, core=simple/normal, cputype=auto.
Using cycles=max or auto might result in some players being able to execute some actions faster than some others (eg: player running on fast desktop vs old laptop).
I did some measurements on Time Gate (1995), we are talking about 15 ms difference every screen transition for a CPU twice slower than another one (this is not neglectable).

Additional info :

Quote:

- DOSBox is not a cycle-accurate emulator.
If you set cycles = 3000 this mean DOSBox will execute 3000 instructions per millisecond regardless instruction type.
On a real CPU (eg : 386) some instructions take much more time than some others (factor can be 10x).
So we can't really say that a 386@33mhz is same as 7800 cycles in DOSBox for example (as suggested here).
Some emulators do that correctly (eg : NES emulation on Mesen is cycle-accurate) but not DOSBox.

- DOSBox is single threaded. It's not going to run faster on a 8 core CPU than a 4 cores ones for example.

- Number of CPU cycles guessed out in auto mode will vary over the time. That is because the time/effort needed to emulate the game may vary (eg : you are in game menu vs you are playing the game).

- Number of CPU cycles guessed out in auto mode will always be much higher than what the game really need to be smooth (eg: the game would run perfectly fine with cycles = 30000 but auto will use 700.000 because your CPU is fast enough to execute that amount of instructions every millisecond and still have time for DOSBox to execute internal stuff). cycles = auto will always result in having one of your CPU core maxed out (eg: 25% CPU usage will be reported in task manager on a 4 cores machine, meaning one core is at 100%). This can be an issue on computers that tends to heat up quickly then throttle (eg: old laptops)

- The reason why setting too high CPU cycles make DOSBox laggy is because the way it works : first it emulate as much instructions as you requested, then it ran some internal job (eg: read input, render screen, ...)
If you use a too high value, DOSBox will take lot of time to emulate that amount of instructions you requested, and will read input and do the rest at a very low rate (which will make it laggy).

- Dynamic mode is less precise than simple/normal because it convert instructions in advance by creating blocks of 32 opcodes (or less), which reduce granularity.
Additionally converting instructions in advance might create trouble with games that use self modifying code

Hi there. Here is some "hard to find" information about DOSBox settings I collected. It's based on some personal research and discussion with LotBlind :

[color="PeachPuff"][b]cycles[/b][/color]

[b]fixed [number][/b] : how much CPU cycles DOSBox will execute every millisecond.
That value shouldn't be too high (eg: 2000000) as it can create trouble (eg: input lag, choppy sound). For CPU demanding games, auto/max should be used.

[b]auto[/b] : it will try to max out the number of CPU cycles used for the game (to put it as high as possible) while keeping some safe margin for DOSBOX itself.
By default, 90% of the CPU time will be spent for running game code, and there is always 10% kept for DOSBox internals (eg: rendering screen, sound, I/O, ...).

It's only enabled for games running in protected mode (eg: DOS/4G). For games running in real mode, the number of CPU cycles will be set to 3000. The idea behind this is that games running in real mode are usually old games.
Most of those games are running too fast on fast CPU.

[b]max[/b] : same thing as above but it will max out cycles even for games running in real mode.

[b]max [percentage][/b] / [b]auto [percentage][/b] :
It's possible to fine tune how much time is used by the game vs DOSBOX internals by pressing CTRL-F11 / F12
How much CPU is allocated for the game is calculated this way : percentage * 0.9
At 100% game will receive 100 * 0.9 = 90% of CPU (DOSBox has 10% left for itself)
At 105% game will receive 105 * 0.9 = 94.5% (max value)
Minimum value is 1%.

[color="PeachPuff"][b]core[/b][/color]

[b]normal[/b] : game is interpreted instruction by instruction. slow, but accurate.
[b]dynamic[/b] : instructions are converted to host CPU instructions before execution which result in a much better performance. less accurate : DOSBox might execute more or less instructions than what you requested. not supported on all platforms. might crash some games.
[b]simple[/b] : same as normal but optimized for real mode games. It's faster than normal core because functionalities related to protected mode (eg: memory paging) are not emulated. 
[b]auto[/b] : switch to dynamic if available and if game is running in protected mode. otherwise normal is used

[color="PeachPuff"][b]cputype[/b][/color]

should be set to [b]auto[/b]. Other settings are used to reproduce some specific CPU models (eg: with 386_slow, some instructions or features will be not emulated).  
It might be useful for running some very old games that work only with a specific CPU model.

When speed running old games, it's more fair to force players to use cycles=fixed, core=simple/normal, cputype=auto.
Using cycles=max or auto might result in some players being able to execute some actions faster than some others (eg: player running on fast desktop vs old laptop).
I did some measurements on Time Gate (1995), we are talking about 15 ms difference every screen transition for a CPU twice slower than another one (this is not neglectable).

Additional info : 
[quote]
- DOSBox is not a cycle-accurate emulator.
If you set cycles = 3000 this mean DOSBox will execute 3000 instructions per millisecond regardless instruction type.
On a real CPU (eg : 386) some instructions take much more time than some others (factor can be 10x). 
So we can't really say that a 386@33mhz is same as 7800 cycles in DOSBox for example (as suggested [url=https://www.dosbox.com/wiki/Performance#Emulated_CPU_equivalency]here[/url]). 
Some emulators do that correctly (eg : NES emulation on Mesen is cycle-accurate) but not DOSBox.

- DOSBox is single threaded. It's not going to run faster on a 8 core CPU than a 4 cores ones for example.

- Number of CPU cycles guessed out in auto mode will vary over the time. That is because the time/effort needed to emulate the game may vary (eg : you are in game menu vs you are playing the game).

- Number of CPU cycles guessed out in auto mode will always be much higher than what the game really need to be smooth (eg: the game would run perfectly fine with cycles = 30000 but auto will use 700.000 because your CPU is fast enough to execute that amount of instructions every millisecond and still have time for DOSBox to execute internal stuff). cycles = auto will always result in having one of your CPU core maxed out (eg: 25% CPU usage will be reported in task manager on a 4 cores machine, meaning one core is at 100%). This can be an issue on computers that tends to heat up quickly then throttle (eg: old laptops)

- The reason why setting too high CPU cycles make DOSBox laggy is because the way it works : first it emulate as much instructions as you requested, then it ran some internal job (eg: read input, render screen, ...)
If you use a too high value, DOSBox will take lot of time to emulate that amount of instructions you requested, and will read input and do the rest at a very low rate (which will make it laggy).

- Dynamic mode is less precise than simple/normal because it convert instructions in advance by creating blocks of 32 opcodes (or less), which reduce granularity. 
Additionally converting instructions in advance might create trouble with games that use self modifying code[/quote]