VOGONS


First post, by xjas

User metadata
Rank l33t
Rank
l33t

I did some benchmarks to answer this question. Decided to spin this off into its own thread. Feel free to contribute (other games, other systems, etc.)

EDIT: RESULTS UPDATED TO INCLUDE USING FASTVID as suggested by Keropi below.

Machines used:
Celeron 1GHz @ 10x100 (Tualeron) - Socket 370
Gigabyte GA-6VEML / VIA Apollo PLE133T
512MB PC133
Onboard Trident Cyberblade AGP (8MB shared)

P4/3.0HT (800FSB) - Socket 478
Gigabyte 8IPE1000-G / Intel 865PE
3GB PC3200 DDR
Radeon HD3850 AGP

Also tried: 16MB Voodoo Banshee PCI (Guillemot Maxi Gamer Phoneix)

- OS was a fresh install of FreeDOS 1.2 with totally default options, EMS config from the boot menu used
- For FASTVID results, only himemx.exe was loaded and fdconfig.sys/autoexec.bat were cleaned up. (No EMS, no FDAPM, etc.) Fastvid was configured using the suggested settings given when run.
- All games were run with NO sound cards configured
- SOFTWARE RENDERING was used in all cases, otherwise this whole exercise would have been pointless. All tests were run in 8-bit VESA SVGA modes.

Testing procedures:

Duke 3D ver. 1.3D installed straight off the CD
Set resolution by editing duke3d.cfg, started a new game, typed 'DNRATE', waited for the aircraft crash to finish & the FPS to stabilize. Then I walked over and blew up the barrels and tried to note the lowest FPS I saw on the counter.

Quake ver. 1.08 final official DOS patch
Standard 'timedemo demo1.dem'.

Descent 2 1.2 final official DOS patch
Launched with -superhires and started a new game on level 1. Entered the 'FRAMETIME' cheat, and with my ship sitting still at the opening position, I recorded the FPS. (It helps if you start a new game, type 'FRAMETIME', then quit and start again so you don't open the first door and alert the AI.)

Hexen 2 1.28 DOS port running the demo datafiles
Started a new game with the default character and entered 'timerefresh' on the console.

Results

hg3_000.png
Filename
hg3_000.png
File size
8.21 KiB
Views
1369 views
File license
Fair use/fair dealing exception
hg3_003.png
Filename
hg3_003.png
File size
4.7 KiB
Views
1369 views
File license
Fair use/fair dealing exception
hg3_001.png
Filename
hg3_001.png
File size
5.4 KiB
Views
1369 views
File license
Fair use/fair dealing exception
hg3_002.png
Filename
hg3_002.png
File size
5.37 KiB
Views
1369 views
File license
Fair use/fair dealing exception
hg3_005.png
Filename
hg3_005.png
File size
5.23 KiB
Views
1454 views
File license
Fair use/fair dealing exception

(I made these charts in Harvard Graphics 3 for DOS because yo dawg I head you like DOS so I nevermind this joke was tired years ago.)

Conclusions:

The fast P4 is still working hard to run late DOS/software-rendered games at 1280x1024, and a high-end P3 isn't always adequate. Without Fastvid loaded Quake & Hexen 2 never saw "smooth" framerates; Descent 2 was playable at 1280 but it chugged a bit if I went into the map more. With Fastvid loaded every game was playable at 1280x1024 but slowdowns still occurred. Even Descent 2 which initially posted 80FPS with FV on the Radeon slowed down to under 40 in occasional areas on a quick playthrough (lots of combat, complex geometry, etc.)

Tl;dr: A 3GHz P4 is by no means "wasted" running these games at 1280x1024 and the extra speed is definitely still beneficial!

Obviously 3D acceleration would make a big difference but that's not the point of this test.

Feel free to contribute more results!

Last edited by xjas on 2017-09-13, 01:03. Edited 1 time in total.

twitch.tv/oldskooljay - playing the obscure, forgotten & weird - most Tuesdays & Thursdays @ 6:30 PM PDT. Bonus streams elsewhen!

Reply 1 of 9, by keropi

User metadata
Rank l33t++
Rank
l33t++

Don't you need FASTVID or similar to "unlock" the full potential of a VGA on a pentium2 and upwards? Have you ran it before you made your benchmarks?

🎵 🎧 PCMIDI MPU , OrpheusII , Action Rewind , Megacard and 🎶GoldLib soundcard website

Reply 2 of 9, by xjas

User metadata
Rank l33t
Rank
l33t

^^ Wasn't aware of it. I'll go through and try it later (probably only on the P4 stuff). I wanted to start off with a bone-stock configuration anyone could try to reproduce.

twitch.tv/oldskooljay - playing the obscure, forgotten & weird - most Tuesdays & Thursdays @ 6:30 PM PDT. Bonus streams elsewhen!

Reply 3 of 9, by boxpressed

User metadata
Rank Oldbie
Rank
Oldbie

Yes, Fastvid or equivalent is essential. Here's a thread where I learn about its equivalent for Athlon XP CPUs.

Fastvid + P3 & BX > Athlon XP & KT133A&

Reply 4 of 9, by infiniteclouds

User metadata
Rank Oldbie
Rank
Oldbie

Do try FastVid - I'm really curious for an accurate comparison to S939. With MTRRLFBE (Athlon 64 4000+) I get 40 FPS in Quake Timedemo1 @ 1280x1024. Unfortunately MTRRLFBE does not solve the abysmally poor performance of build engine games. NOFLB seems to smooth things out but the ingame FPS counter still claims 10 FPS. There's no way it's accurate because it is much smoother once NOLFB is running -- the problem then becomes terrible tearing.

Still questing how to get build games to take advantage of the Athlon 64's power.

Reply 5 of 9, by xjas

User metadata
Rank l33t
Rank
l33t

^^ Results with Fastvid updated.

infiniteclouds wrote:

I'm really curious for an accurate comparison to S939. With MTRRLFBE (Athlon 64 4000+) I get 40 FPS in Quake Timedemo1 @ 1280x1024. Unfortunately MTRRLFBE does not solve the abysmally poor performance of build engine games.

I got 48 FPS in Quake on the Radeon so that seems pretty close to the mark (discounting the Voodoo Banshee result which seems unusually low.) Saw 44/48 in Hexen 2 on the two cards.

Fastvid didn't seem to make much difference in Duke3D, maybe 3-5 FPS at best, so the bottleneck seemingly lies elsewhere.

I'm also curious to see what happens on fast AMDs (754/939/+) with the other games so please post up if anyone gives it a try!

Last edited by xjas on 2017-09-13, 02:31. Edited 1 time in total.

twitch.tv/oldskooljay - playing the obscure, forgotten & weird - most Tuesdays & Thursdays @ 6:30 PM PDT. Bonus streams elsewhen!

Reply 6 of 9, by Standard Def Steve

User metadata
Rank Oldbie
Rank
Oldbie

Another thing to keep in mind is that VIA chipsets are quite slow in DOS.

I have two P3 machines:

440BX
Celeron 1.4GHz, 100FSB, 256K L2'
512MB PC100 CL2
Voodoo3 AGP

Via Apollo Pro266T
PIII-S 1.58GHz, 150FSB, 512K L2
2GB DDR-300, CL2
6800GT AGP

There are some DOS tests where the BX/Celery/SDRAM system actually manages to outperform the VIA/PIII/DDR system. Yet, in Windows, the VIA system absolutely demolishes the BX system.

94 MHz NEC VR4300 | SGI Reality CoPro | 8MB RDRAM | Each game gets its own SSD - nooice!

Reply 7 of 9, by nforce4max

User metadata
Rank l33t
Rank
l33t

I stay away from doing dos gaming on pentium 4 and newer systems though as you clearly proven that for some games it makes a big difference but personally I stay away from modern systems for this use for the usual reasons plus I like period correct systems.

If all else just shot gun it and have multiple systems so that one isn't forced to make it work with just one or two systems.

On a far away planet reading your posts in the year 10,191.

Reply 8 of 9, by infiniteclouds

User metadata
Rank Oldbie
Rank
Oldbie
xjas wrote:

I'm also curious to see what happens on fast AMDs (754/939/+) with the other games so please post up if anyone gives it a try!

I don't have Hexen or Descent 2 on hand... but build engine games (Blood and Duke3D) run poorly. 800x600 is between 30-40FP on the rooftop... and more like 40-50FPS on the streets.. running NOLFB improves performance and FPS seems locked at 48 -- never dips. This is only at 800x600 though... so there is something that hasn't been figured out yet about getting Athlon 64s to run well with Build.

Also, can someone tell me how to remove NOLFB from memory? It's nice to get that boost in performance for this game but it is annoying having to restart to get rid of it -- it puts severe limitations on Quake's available resolutions when it's in memory.

Reply 9 of 9, by dr_st

User metadata
Rank l33t
Rank
l33t

xjas,
Great job and big thanks for taking the time to run these benchmarks.

I happily stand corrected on my previous assumptions that DOS games cannot max out a fast P4. I would say, based on your results, that 800x600 is roughly the cutoff resolution. On 800x600, a fast P-III should be enough for all games (assuming FASTVID is applied). Beyond that - a P4 starts being useful.

https://cloakedthargoid.wordpress.com/ - Random content on hardware, software, games and toys