VOGONS


First post, by ruthan

User metadata
Rank Oldbie
Rank
Oldbie

This post is WIP - Work in progress it would be updated, with new info.

If you are not in mood to read ( Someone was born to read, someone to lead:) :
1), here is results link - html view
2) If you wan to benchmark, download Phils DosBench, PathBench (download link is in 1st post) from vvbee , here is VideoBios size utility from FalcoSoft and for Quake2 would need own Quake2 retail data + Q2DOS or this Q2DOS + Q2 single demo testing package . Fastvid (its lite package no vspeed.exe etc..) and MTRR speed enhancers are already in Phils pack, to change resolutions in Quake games use ingame options - its more safe that videomodes numbers because there are cards specific and some cards as S3 support some strange resolutions so, generic 640x480,1024x768 numbers actually trigger lesser resolutions, so automatic scrips are not working.
Magic lines for Quake consoles (opens with tilda key - ~):

Quake1: Timedemo demo1 <presss ENTER> If you have blinking screen problem (some Nvidia cards problem), tryvid_nopageflip 1 or vid_wait 0 - console command - all characters has to be lowercase, after command execution you have change resolution to take settings effect [urlhttps://www.vogons.org/viewtopic.php?p=709331#p709331] Be aware that vid_nopageflip 1 is saved in game configuration even between game execution, so if want to do "standard" testing you need to set it back to 0 by vid_nopageflip 0 and change resolution
to make it working
Detail explanation how it works[/url] thx FalcoSoft.

Quake2:
open console
Timedemo 1 <presss ENTER>
map demo1.dm2 <presss ENTER>// For Full version of games
map Q2Demo1.dm2 <press ENTER> // For Q2 Single player demo, demo has different end, but fps is by my testing ~ same

Results are in console.
I prefer full line of results, if you own Quake2, otherwise all other used tools are free, so there is not excuse - crash is valid result too, if results is slideshow (<3 FPS), you can quit benchmark and give me forecast of speed (Q2 reports results for part of benchmark, Q1 you can stop benchmark by run benchmark command from console again and you will find results from last incomplete run in console (benchmark performance is pretty consistent so its not big deal you benchmark only part of it), Doom running on everything). For 3dfx cards use software mode, DOS Glide results cant be posted as additional info. !DONT use Windows 9x DOS mode, only pure DOS! . You can post Virtual machine / DosBox results too with info about physical hw + DOS version or used virtualization tool (i know Vmware and Virtual DOS sounds is broken, i dunno about QEMU, ESX and other more obscure engines..). If you machine overclocked, pleas report stock and overclocked clocks (i dont mean results just something like Core 2 Duo E6660 2.4Ghz @3.5 GHz). You can use any version of DOS - FreeDOS, DR-DOS, older MS-DOS only pleas report it, FASTVID regarding of its documentation is working only in Realmode (=himem.sys / himemX.exe only no EMM386/JEMM/QEMM etc..) and Q2DOS has up to 4x faster results in Realmode than in EMS mode (NO EMS mode has very similar results).
3) If you have problem to run pure DOS on your machine look here.

Hell ou,
Phil has some nice DOS benchmarks for some PCI and AGP cards, but i wonder about performance on more modern PCIE/ Radeon AGP 3850 / Geforce 520 PCI etc.. cards on system as X58,intel X6X, i865, newest VIA chipsets etc.. but are also pure DOS compatible. I prefer real game test against synthetic benchmarks. Did someone these test?

Discussion / target - which things would be also interesting to discover:
1) which more modern Videocards are glitchless in DOS and which arent (i added any incompatibility note which i met to my Sound cards compatibility testing i can migrate them to special sheet later, i tested lots of games on lots of videocards in process) For now you can usethis matrix for older cards.
2) How more modern videocards which are starting in underclocked speed, until proper speed is initialized by drivers would compared to older cards. . => Some of them performance same or better than older cards.
3) What is exactly 2D performance of more modern card, i saw some Win98 results and because of removing 2D acceleration features some modern graphics card performance was lame.. => Some of them performance same or better than older cards.
4) Discover if DOS videocards speed up utilities are working for modern cards => look at my results, there are very results with and without speed up utility
5) Could Vesa enhancers fix broken default vesa glithces and its possible to make them working on more modern card? Its Univbe thing, i still dont fully understand it, i even used it properly. Even back in era when i used DOS i had cards, which just worked.
6) What is best videocard for Quake 1-1600x1200? => Well 1600x1200 is not supported, Quake ends with 1280x1024.. what is 5:4, so im rather testing 1024.
8) I remember that at some period was possible change videocards frequencies through Bios flashing, wouldnt be possible to change that too slow clocks for default boot for some modern card, to improve their pure DOS performance?
9) If exist some Build engine benchmark i will add it to results..
10) Which is fastest Dos videocard? => If only take to consideration games in our benchmarking suite it would some highend of Radeon HD from 4xxx to 7xxx line, because performance in not based only on raw power, i cant tell for sure, what card would be the best without results, but my guess with that would be HD 7990 - its 375W TDP monster vacuum cleaner.. but i probably some card with 35-50W TDP would have 80-90% of its performance.

Analysis:
- I found some results, here is -Vogons thread.
- Phils DOS bench results DB - https://docs.google.com/spreadsheets/d/1lvF9n … wpNU/edit#gid=0 and Vogons thread
Structure - there is only problem that its a bit over engineered, CPU info is far on right and cpu info far on left and if was some videocard or cpu enhancer used is on righ collumn.. Simply with Phils sheet you have to do much scrolling as from left to right to discover what was how tested, i wanted to have all important performance parameters on the left and all collums on 1 monitor screen and hide syntetic benchmarks, sorted it by Quake performance test. Merge stock and overclocking results. Results is here: // But because its copy, i will be quickly outdated, i did it for myself in 15 minutes, to create this analysis, maybe it could help other people..
https://docs.google.com/spreadsheets/d/1PPPOD … 343A/edit#gid=0
Data - much more severe problem, is that Quake in 320x200 results in hunders of FPS, is not good to tell you about performance in high resultions as 640x480 or 1024x768.. and for such machines is pure nonsense to use 320x200, what is difference between 200,300,500 FPS.. and what for is such results if in 640x480 or in 1024x768 you have less than 30 FPS?

Some results Phils sheet data Quake 320x200 - its not ideal source of info for high resolution prediction, but lets use them:
- im ignoring Doom results, because Quake number seems to be a bit more conclusive.. a make IMHO more sense to reflect system performance
- fastest cards, ordered from fastest to a bit slower (of course that there is CPU influence of results, but even best CPU cant push slow cards, ro simply i selected fastest achieved results):
- Radeon 48xx (1170 FPS),54xx (940), 5670 (722), 5770 (880),5750 (1090) - 7xxx (700) PCI-e 475-1130 FPS, here is seems that these cards have some performance head room and result is CPU depended
- Radeon X800 XT tested, so i expect that X850 would be same or even better - 930 FPS
- here i add my new result Vooodoo 3 2000 - without FastVID on Core E6600 - 550 FPS
- here i add my new result Matrox Mystique 4 MB - 511 FPS with Core E6300 (slowest Core 2 Duo)
- Radeon 9550-9800 AGP/PCI-e - 400 -480 FPS // = X600 because is simply almost renamed X9800..
- Radeon X800 XL - 440 FPS
- Mobility Radeon X700 - 440 FPS
- Geforce 7900 GTX - 430 FPS
- Radeon X800 Pro - 420 FPS
- Radeon 8500 - 400 FPS
- Radeon X1950 - 400 FPS
- Geforce 3 TI / Geforce 5900XT / Geforce MX440/ Riva TN2 M64 (yeah slimer..), Geforce 9800 GTX+ - 350-400 FPS // Nvidia time
- Intel HD in CPU (Pentium G620) / Intel Core i7-3770 - 325 FPS

Conclusions from Phils project results (2018, check online sheet for up to date results):
- fastest cards - AMD/ATI cards are clear winners
- from results is can say that Phils Quake results - 320x200 is document called ultimate videocard benchmark is i reality CPU benchmark, because i lowres results it really scale with CPU with same work at unless you are using slower old videocards (1 MB video card or something like that even with 4 MB Matrox or Voodoo3 - they are capped around 500 FPS, slower older cards 100-300 FPS), you can see that results for 640x480, 1024x768 etc.. have very similar results regardless of CPU, so these would really Videocard benchmark..
- its quite interesting that best DOS performers arent Win98 compatible, so i want to multiboot machine, you have to do some compromise, or you need Gigabyte / HP OEM boards with primary PCI-E selection.. If want cream de la cream - best multiboot wet dream machine (i will now ignore videocards DOS compatibility some fastest cards can have some DOS games compatibility problem like srolling etc..) - configuration will look like this:
DOS - Radeon 4890 / Win98 - Radeon X850 / XP - Geforce first Titan / Win7- Win10- Geforce 2080 Ti, of course if you care about noise and power consumation is better something like this:
DOS - Radeon 5450 ( there are card with passive cooling) / Win98 - Radeon X800 (there passivelly cooled cards) / XP - 750 TI (there are also in KalmX passively cooler) / Geforce 2070 / 2080.. other solution would be water cooling ring with 1slot watercooling blocks.
- fastest cards are PCIE, other AGP and PCI cards cca 3x Slower (Update: lower i achieved better results with faster CPU, not is not so big deal 350 vs 500 FPS) that best PCie card:
best PCI cards are for now MGA 1064SG (340FPS). A bit Slower is S3 VirgeDX (300 FPS), S3Trio64V+ (250), Vooodoo 5 5000 PCI
- what is big problem with Phils results for Quake you cant really translate 320x200 performance to higher resolutions, because with Geforce 220 - i have 300+ FPS in 320x200, but only 12 FPS in 640x480 and 4 FPS in 1024x768. Natural scaling if anything would work as expected, lets say linearly with number of pixels would be something like that - 320x200-300 FPS; 640x480 - 300/4.8= ~62;1024x768=300/12=25 FPS.
- its seems that performance of modern Nvidia cards in DOS sucks in comparision with ATI for low res- Geforce 640GT - 320 FPS, Geforce 660 - 320,, i know that even 120 FPS is enough for high end LCD/CRT, but dont for just numbers comparision.
- lots of very old cards can be really pushed a lot in small resolution by fast CPU, it should that there engine are very effective or they dont do much
- fastest AGP card in test was Radeon 9600SE - 482 FPS, maybe because i didnt find any X8xx AGP in test, R 9600 SE i bit slower that even fastest PCI cards, but could be problem with used CPU power, because it was tested with P4 3.6 HT only.

My results, in other words, i keep Doom 320x200 + Quake 320x200 for results "mapping"+ added 640x480/1024x768 + Quake 2 DOS port - 640x480 / 1024x768 /1600x1200:
My measurements with few reused numbers from Phils video, before i discover his sheet:
https://docs.zoho.com/sheet/published.do?rid= … c62b9&mode=html

Conclusions from my sheets project:
- first of all, graphics enhancers utilities are working with most of cards, at least for some combination of resolution and game - most often with Quake1 (for which is there even additional VID_NOPAGEFLIP 1 speed hack to bypass v-synch shenanigans), less often with Quake 2, much less often for Doom and Doom boost is relatively small, in comparision with them, for Quake they could give you 20! times more performance in some causes, its brutal, almost unheard as i bit shame that such thing in not used by Microsoft and inbuild in MS-DOS nativelly.
- Doom performance on lots of modern, quick cards simply sucks and enhancers are not cure, what is very strange in comparision with old PCI 1-2MB cards, which all are 60 FPS+ for Doom.
- if you dont have fast modern machine, there is performance advantage between real dos machine and Dosbox
- what i dont really understand is that some cards are faster in 1024x768 than with 640x480 in Quake on lots of cards, only conclusion is that Quake or its inbuld vesa drivers are bugged, i dont believe that there is something in chips architectures what cause this - like dynamic overclocking or i thing that almost all modern card are in DOS using its booting minimal idle clock, which are cant be triggered by some DOS code and for better clocks are needed OS drivers. => Update: VID_NOPAGEFLIP 1 is probably cure, more testing needed.. - there could be V-Sync FPS cap with some cards and resolutions, details are here by FalcoSoft
- FalcoSoft about DOS refresh rate and FPS cap: Your monitor's specification does not matter here. Under DOS all SVGA/VESA modes are limited to 60 Hz unless the VESA BIOS supports VESA3 refresh rate setting functions. Unfortunately NVidia disabled these refresh rate settings more than a decade ago (last working versions were Geforce 4 series cards AFAIK and other vendors never supported it). But none of the known games supported VESA 3 refresh rate settings by default. You had to set 60+ Hz refresh rates manually with a special utility - UniRefresh.

- i dont have explanation for X58 Xeon56xx results its much faster cpu and ram than Core 2 Duo systems, 320x200 results is fine, but 640x480 or 1024x768 are very slow regardless of card for PCI. Only problem which i can see is that my machine have poor DOS HDD performance 3 MB/s for integrated Sata controller in IDE, i didnt care because its DOS and its enough for it, but i dunno how is Quake loading texture, is there are loaded before benchmark or there are constantly download and unloading from Storage, because that would means that benchmark is very storage speed depend. I have to try some Ramdisk and compare results (with whole Dosbench on Ramdisk results are still the same, only PC speaker get constant beep when i run benchmark from Ramdrive). I dont have a clue, instructions should be backwards compatible, maybe some problem with some HP O.E.M board implemetation.. but im confused. X56xx card resonable performance only with Matrox Mystique 4 MB and even that, even much faster CPU and memory (DDR3 1333 vs DDR2 667 ) is much slower that slowest Core 2 Duo - 30 vs 57 FPS in 1024x768.
- its seems that number of Stream processors on AMD cards really doesnt matter for DOS, important probably other 2D processing part of chip and its clocks and memory clocks (or lets say its default boot clock which are used for DOS)
- with very fast single thread CPU like 8700K at stock clock, i have with Vmware in every benchmark 100+ FPS, so i looks like winner, that we can throw away old machine, unfortunately Vmware and Virtualbox DOS sound cards drivers suck a lot and FX and FM are not working both together, Vmware pushing other things, so unless someone write own new driver this is now way, way is maybe with other MBs.. pass through some PCI sound card, but i far i know passthrough has only Vmware ESX and its hassle and i dunno it properly working for sound cards.. Even with 8700K is not Dosbox fast enough for max. resolution Dos Q1/Q2 gaming.
- we now have Radeon HD results for 1280x1024 and these cards are really fast also for this results at least in Quake 1, we already knew about 320x200, there are lots of cards with brutal 320x200 Quake 1 performance, but there are quickly lossing performance with resolutions, Radeons HD dont..
- other interesting thing is that Q2 is in general much faster in same resolution than Q1, logic would expect otherwise because Q2 is fully 3D and there are more effect, but Q2 engine could be simply better or Q2DOS has some good speed hacks.

Submit your results and help community:
- you at start of this post how to..
- It would be nice if Phil will add columns for higher resolution to his sheet, but don fear of duplicity, i promise that i would migrate all your results from here to his sheet..

Additional info
- list of DOS games with inbuild FPS counter and how to enable it
- best Windows 98 AGP card thread-- Rayers utility for disable on-chip scaling on nVidia VGA in DOS - its about DVI picture sharpness
- Windows performance of PCI cards up to newest - Youtube video.
- S3 Dos picture too bright problem thread - its VBios ROM problem

Additional tools

  • VSPEED.EXE that reports the video - throughput for bit blit operations from DRAM to VRAM for both the banked VGA and linear frame buffer mechanisms. There is more info about FastVId and Vspeed.
  • Screamer 2 and its demo + Screamer rally at the start of their setup, testing testing videocard speed - there is some result in MB/s is some videocard transfer rate, rate is better.
  • DosTest packages from Roy Longbottom (icnluded in my DriversAndTools Package directory DOSTESTS) - Videodos and dosbus benchmarks, which can measure, memory bus and videocard performace in 640/800/1024
  • On some card is possible to discover boot / DOS clocks on Windows, more details are here from FalcoSoft, linked NiBiTor is Nvidia only and support up to Nvidia PCI-E 590 GTX from Geforce 5 or something like that. I also can change boot (DOS) clocks, for cards which have multiple power / speed states.
  • Here some Doom only results measured by someone outside of Vogons.

Vbios size:
- For most and possible available pure Dos conventional memory is good to have smaller videobios as possible, it a bit problem on modern videocards, especially if UMBPCI is not working with you MB (and its only for realmode, so it will not fix EMS mode) i created thread about that:
here , you can report it too, i added column for it in our sheet too.

List of DOS highres games where performance could matter:
link

Videocards Compatibility (here im mainly gather some material for compatibility research and no i dont to do all by myself, i head that exists something like community..):
- compatibility matrix with old cards (up to Voodoo 5 and similar) - https://gona.mactar.hu/DOS_TESTS/
- you can find lots of info about videocards incompatibilities as side product of my modern chipsets + sound card testing, all its in notes columns (maybe sometime will someone transform it in to some unified text block)
- Radeon HD AGP glitches thx to LSS10999 for info - Vogons link
- NVIDIA Maxwell2/Pascal Vesa bug - info from FalcoSoft, there is also fix for Geforce 960 (maybe more) WinXP DOS VESA/VGA modes- NVIDIA Kepler/Maxwell/Pascal VESA Bios Bug (workaround found) . I simulated it with G970 and G1030. Same of similar bug has even Gefore 6xx / 7xx line. Rayer and FalcoSoft are working on fix.
- free Univbe release - Vogons drivers library - http://www.vogonsdrivers.com/getfile.php?fileid=346
- its not DOS thing, but consider DOS/Win9x combo card (because who has enough slots, especially with two slots cards), there is Win9x DOS mode full gub with Geforce 4/5/6/7 cards, details
- Tip by AvanlonH I always test the DOS compatibility of all my AGP and PCI-E graphics cards by running Rambo 3 (1988) in the same setup. Running 'Rambo.exe' or 'Rambo.exe r' (game setup option) will crash the system using both these cards. Plugging in an old 1994 PCI Diamond Stealth 64 2MB (S3 Vision964) and it works. This game will not work with most modern cards. Failing cards: FX5200, 4200ti, NVS 280, NVS 300, AMD HD4770.

Thread about how to measure framerate in DOS games would FPS counter
- its out of scope of my benchmarking, its more work, but it could help to analyze performance in games without inbuild benchmarking tools..
- VGA capture thread 28 pages of info..
- how to capture resolution bellow 640x480 with Aver cards..

Most interesting yet untested cards - test them if you have them..
- Kyro II, Riva 128/TNT/TNT2/TNT2 ultra, Geforce 1/3, Vooodoo 4/5, Radeon X850, Radeon HD 2xxx,Radeon HD 5xxx,Radeon HD 6xxx, Radeon HD7xxx, more modern Matroxes like G200/G400, Intel MB Videocards, Geforce 750 / 750 Ti (especially Q1 1280x1024 if it blinking as Gerforce 9xx/1xxx).
- also would be nice test some already tested cards with slower / faster CPUs, to discover how their performance scale.

My similar threads:
DOS PCI soundcards with modern chipsets. + branched configs and drivers and tools package
DOS Network cards and networking - some modern Gigabit PCI/PCI-E working fine in DOS
PCI Video cards benchmark: Electric Boogaloo Edition - Similar create after this thread

Disclaimer:
As you probably find out, im not good with letters+grammar in every language, i never had brain for that, i was always better with numbers and analytics, sorry. If something in text is not clear, just ask, i will fix it.

Last edited by ruthan on 2021-01-21, 17:13. Edited 112 times in total.

Im old goal oriented goatman, i care about facts and freedom, not about egos+prejudices. Hoarding=sickness. If you want respect, gain it by your behavior. I hate stupid SW limits, SW=virtual world, everything should be possible if you have enough raw HW.

Reply 1 of 71, by ruthan

User metadata
Rank Oldbie
Rank
Oldbie

I merged first 2 posts, did some clean up.. lots of new results include Q2DOS to in sheet, results are still the very wild.

Im old goal oriented goatman, i care about facts and freedom, not about egos+prejudices. Hoarding=sickness. If you want respect, gain it by your behavior. I hate stupid SW limits, SW=virtual world, everything should be possible if you have enough raw HW.

Reply 2 of 71, by bakemono

User metadata
Rank Oldbie
Rank
Oldbie

I tried Quake 1 on a GeForce GT 240 and results were a bit weird. 1280x1024 is much faster than 1024x768 or 640x480. CPU is Athlon II X2 3.2GHx with AMD 770 chipset running at PCIe 1.1 speed (PCIe 2.0 and HT3.0 are supported but cause the chipset to overheat)

320x200 - 352fps
640x480 - 29.5fps
1024x768 - 15.0fps
1280x1024 - 55.5fps

with FASTVID:
640x480 - 29.8fps
1024x768 - 29.8fps
1280x1024 - 82.1fps

Reply 3 of 71, by ruthan

User metadata
Rank Oldbie
Rank
Oldbie

Hi Bakemono, thanks for results i will add them, or sent me PM if you want to have wire access to sheet.

1280x1024 its nice catch, i will add column for that, i never tested it, because i have Q2 with 1600x1200 test, but with such results make sense to add it.
I will also add CPU benchmark column - PathBench, to be sure how CPU perform is dos.. I dont really know if best for games, but its good start and benchmark (if is interger or floating point bench, Update: I contacted author its good for floats) is very user friendly. If someone has something better tell me.
Because im still benchmarking on few CPus is not big deal to add numbers, well i there is virtual machines and Dosbox hassle but its still few numbers until other people start post more results. => Top of the first post updated.

I would need also memory speed info.

Your results arent more weird that other results, which i measured and with other people measured with 320x200 in Phils sheet.. and these are also first results with AMD cpu and chipset. I have only something with Geforce 220.

You can try MTRR.. from Rayer instead FASTVID, im always using better results.

Im old goal oriented goatman, i care about facts and freedom, not about egos+prejudices. Hoarding=sickness. If you want respect, gain it by your behavior. I hate stupid SW limits, SW=virtual world, everything should be possible if you have enough raw HW.

Reply 4 of 71, by RayeR

User metadata
Rank Oldbie
Rank
Oldbie

I just add quickly some notes for clarification...

*There's no any 2D acceleration under DOS or used by DOS games. It doesn't matter how many billions of transistors your GPU have, in VESA mode every GPU becomes a dummy framebuffer device that does only one thing - it receive memory writes from host CPU to MMIO via PCIE/PCI bus store it in VRAM and send it to your display.

*So this benchmark is more about PCIE bus and memory througput of used chipset and speed of used CPU than about VGA card itself. When someone else will benchmark the same VGA card in a diffenet MB/CPU he can get significantly differnt result than you.

*it doesn't matter if you get 400 or 500FPS as your LCD display only 60FPS usually.

*What really matters is if MTRR is properly set to write combining mode - this makes a difference in 1-2 magnitude order. I know that in some special cases MTRRLFBE may fail to set MTRR properly and the speedup effect doesn't appear. It should be job of the BIOS to initialize MTRR properly but most of BIOSes just ignore it or set it wrong way that complicates enabling it via MTRRLFBE. I have experimental version that tries to adjust bad MTRR entries but not enoug clean way as I would want, no time...

*Also it matters how many videomodes VBIOS supports and how many bugs it have. I made a VESATEST utility to check it.

*I have no explanation why in some cases higer resolution modes makes better FPS than lower. In my case FPS decrease monotone way as expected.

*If you want to play Quake in 1600x1200 mode use QDOS instead with native SB live and HDA support. And for Doom try DOSDOOM or ZDOOM

*If XHDD/UIDE/UDMA driver doesn't help to your poor HDD speed big ramdisk may help. You already need to limit XMS for Yamaha driver so this unused memory may be used by RAMDISK.

Gigabyte GA-P67-DS3-B3, Core i7-2600K @4,5GHz, 8GB DDR3, 128GB SSD, GTX970(GF7900GT), SB Audigy + YMF724F + DreamBlaster combo + LPC2ISA

Reply 5 of 71, by ruthan

User metadata
Rank Oldbie
Rank
Oldbie

Thanks, few reaction based on results.

RayeR wrote:

*There's no any 2D acceleration under DOS or used by DOS games. It doesn't matter how many billions of transistors your GPU have, in VESA mode every GPU becomes a dummy framebuffer device that does only one thing - it receive memory writes from host CPU to MMIO via PCIE/PCI bus store it in VRAM and send it to your display.

*So this benchmark is more about PCIE bus and memory througput of used chipset and speed of used CPU than about VGA card itself. When someone else will benchmark the same VGA card in a diffenet MB/CPU he can get significantly differnt result than you.[/quote]
We know that, but apparently it matters how such dummy framebuffer device is implemented in videocard, or how videocard and its VRAM are clocked for boot mode until are drivers initialized or there is other factor (s) which decide how quick is CPU-MEM-BUS-FrameBuffer transfers. Because numbers on same machine / CPU vary a lot for different cards on same bus - PCI (look at Q1 - 320x200-E6600 results - ATI Mach 64 - 154/ Virge twice as much and Voodoo 3 PCI triple ) / AGP / PCI-E is the same.. and Nvidia PCI-E cards are much slower in general than ATI / AMD on same bus. It also doesnt really explain why are modern PCI-E cards, such slower than PCI / AGP ones for Doom.

RayeR wrote:

*it doesn't matter if you get 400 or 500FPS as your LCD display only 60FPS usually.

Yeah, but there are 200 Hz CRTs and 240 Hz LCD TN "gaming" panels and some utilities to set DOS refresh rates, to in theory you can get at least 240 Hz. But yeah these numbers are good mainly only to show raw theoretical performance.

RayeR wrote:

*What really matters is if MTRR is properly set to write combining mode - this makes a difference in 1-2 magnitude order. I know that in some special cases MTRRLFBE may fail to set MTRR properly and the speedup effect doesn't appear. It should be job of the BIOS to initialize MTRR properly but most of BIOSes just ignore it or set it wrong way that complicates enabling it via MTRRLFBE. I have experimental version that tries to adjust bad MTRR entries but not enoug clean way as I would want, no time...

Yeah, in some specific case and higher resolutions are MTRR or FASTVID up to 10x faster.. than results without it.

RayeR wrote:

*Also it matters how many videomodes VBIOS supports and how many bugs it have. I made a VESATEST utility to check it.

What is meant by videomode, only resolution or something else?

RayeR wrote:

*If you want to play Quake in 1600x1200 mode use QDOS instead with native SB live and HDA support. And for Doom try DOSDOOM or ZDOOM

Yeah but Q2DOS is already in test suite and its DOS to Win port, Q1DOS would probably have similar results, target of benchmarking is not boost DOOM or Quake, but measure real DOS performance to have idea, how other highres dos games would perform. Unfortunately there not too much DOS games with ability to measure FPS.

*If XHDD/UIDE/UDMA driver doesn't help to your poor HDD, speed big ramdisk may help. You already need to limit XMS for Yamaha driver so this unused memory may be used by RAMDISK.[/quote]
I had some HDD related hiccups only with Q2DOS (on all machines, even on them very is DOS raw HDD speed ok 30 MB+) it allocates at start i thing ~170 MB of RAM, it loads assets into ram during level loading, but probably not all.. so some are still loaded during benchmark, but when you run 2nd or 3rd run its not longer here and from 3rd run are results is the same, difference between 1st and 3rd run could be quite big 20-30% for high FPS values and 640x480, with higher resolutions there are much less hiccups nad much less differences between multiple run. I tried ramdisks for Quake 1 and Doom there is not difference, for Q2 i would have to use some other ramdisk too because its quite big - 300 MB.
Sound drivers - Q1 is testing with -nosound parameters, Doom is the same, im testing Q2 without -sndpci for HD sound cards and loaded sound drivers for legacy SB support, but, even when is enabled there is very small difference, so its not sources of any strange numbers.

Im old goal oriented goatman, i care about facts and freedom, not about egos+prejudices. Hoarding=sickness. If you want respect, gain it by your behavior. I hate stupid SW limits, SW=virtual world, everything should be possible if you have enough raw HW.

Reply 6 of 71, by LSS10999

User metadata
Rank Oldbie
Rank
Oldbie
ruthan wrote:

What is meant by videomode, only resolution or something else?

It can be either available resolutions, or the ability to correctly render images at that resolution.

Some video BIOS have flaws that prevented certain resolutions from displaying correctly, like the Radeon HD series AGP. From what I tested, those cards have missing or flawed 8-bit color implementations that resulted in graphical glitches in some (not all) games, such as the images appears messed up with garbled lines, incorrect palette (some colors rendered into different ones).

Also, the connection type between video card and monitor may also affect compatibilities, depending how the game were programmed. Some video modes/instructions might not be compatible with a non-VGA connection (such as setting mode 13h with high bit set to skip clearing of video memory (93h)). Connecting through an adapter will not magically make incompatible video modes/instructions functional, only native ports (the connection without any adapter) count.

Reply 7 of 71, by ruthan

User metadata
Rank Oldbie
Rank
Oldbie
LSS10999 wrote:

It can be either available resolutions, or the ability to correctly render images at that resolution.

Glitches - except of missing 1600x1200 resolution on lots of cards in Q2 (i think that this VESA mode is not simply listed in its video bios), all cards with this problem has 0 result, i already stated in first post, i never had time and mood to fiddle with Univbe, i had problems only with S3 Virge: it needs driver even for 640x480 and there is listed 1600x1200 in Q2DOS even when card not enough memory for it and there is are also broken colors in menu and for lets say 1,2 seconds of game after gameplay is started.
I also had problem with Geforce 1030 which i tested only to show if still possible use it in pure DOS, with some advanced video text modes used, but File Wizzard file manager, games were fine. Last problem which i have is with Matrox Mystique 4 MB + X58 - i already created special thread for it, graphics is fine, but there are some unwanted beeps even during boot and sometimes after resolution change constant beep.
Any of this problem not affecting benchmarking directly... but of course real video cards selection, has to be compromise between compatibility and performance. For compatibility is there for now with matrix (at least for other cards):
https://gona.mactar.hu/DOS_TESTS/
From my mainly sound cards testing i think that i can add next 10 problematic games,which have some problem with some videocard, but to now i aim to performance and after that, it would make a sense test few best performing cards compatibility, rather than test lots of cards with lots of games, its much less work than to do it vice versa.

LSS10999 wrote:

Also, the connection type between video card and monitor may also affect compatibilities.

Do you think that i could affect the performance too? Old cards which im testing have only VGA(D-SUB), so i have to test them with VGA monitor, never without VGA, im testing with DVI, some cards have the both.

Last edited by ruthan on 2018-10-11, 03:16. Edited 1 time in total.

Im old goal oriented goatman, i care about facts and freedom, not about egos+prejudices. Hoarding=sickness. If you want respect, gain it by your behavior. I hate stupid SW limits, SW=virtual world, everything should be possible if you have enough raw HW.

Reply 8 of 71, by LSS10999

User metadata
Rank Oldbie
Rank
Oldbie
ruthan wrote:

Do you think that i could affect the performance too? Old cards which im testing have only VGA, so i have to test them with VGA monitor, never without VGA, im testing with DVI, some cards have the both.

I'm not sure about performance, but some video commands are not usable when you're not using native VGA (that is, using a native D-Sub port on your video card, not through an adapter). I only know that one particular case of such for now, but the exact list of unsupported commands under non-VGA connection is yet to be known (maybe Analog hole prevention is somewhat involved).

Reply 9 of 71, by Falcosoft

User metadata
Rank Oldbie
Rank
Oldbie

I also had problem with Geforce 1030 which i tested only to show if still possible use it in pure DOS, with some advanced video text modes used, but File Wizzard file manager, games were fine

It seems all NVIDIA Maxwell2/Pascal Video cards have a serious VESA bug that can cause jerky movements/missing frames and flickering depending how a program uses double/triple buffering. Detailed info here:
NVIDIA Kepler/Maxwell/Pascal VESA Bios Bug (workaround found)
Maybe it's most easy to test this problem with VBETEST from UNIVBE's package and PCPBENCH.EXE from Phil's DOS benchmark package (in high resolution modes + WC enabled in MTRRs).
But I have also written a special test just to identify this problem (downloadable from above link).

Website, Facebook, Youtube
Falcosoft Soundfont Midi Player + Munt VSTi + BassMidi VSTi
VST Midi Driver Midi Mapper

Reply 10 of 71, by ruthan

User metadata
Rank Oldbie
Rank
Oldbie

I added Quake 1280x1024 results for 2/3 of tested configurations, for some 1280 is faster than 1024, i add the rest later, i have some HW on different location.

Now it would be really nice to have some Radeon 2xxx - 7xxx results, but i dont have such cards.

Otherwise i find some tools, which could help to investigate, some clues why are videocard performance results so strange:
- Screamer 1/2 and its demo at the start of their setup, testing testing videocard speed - there is some result in MB/s is some videocard transfer rate, rate is better.
- i find DosTest packages from Roy Longbottom (added to my DriversAndTools Package directory DOSTESTS) between other tests, are there Videodos and dosbus benchmarks, which can measure, memory bus and videocard performace in 640/800/1024. Even in Dosbox is 1024 results bigger than 640. You can find there also results from other old machines for comparison, interesting are fast 486 results with comparision with PII/III there is not big performance boost, how could be expected.

VideoDos2.png
Filename
VideoDos2.png
File size
844.5 KiB
Views
6581 views
File license
Fair use/fair dealing exception
VideoDos.png
Filename
VideoDos.png
File size
43.58 KiB
Views
6581 views
File license
Fair use/fair dealing exception
Videodos.jpg
Filename
Videodos.jpg
File size
74.1 KiB
Views
6581 views
File license
Fair use/fair dealing exception
ScremInfo.png
Filename
ScremInfo.png
File size
218.02 KiB
Views
6581 views
File license
Fair use/fair dealing exception
Bus.png
Filename
Bus.png
File size
1.49 MiB
Views
6581 views
File license
Fair use/fair dealing exception

I care more about results (which cards are better), than reasons why number are what they are (im not saying that im not interested at all), but if someone is interested in, these measurement to could help to investigate reasons behind numbers.

Last edited by ruthan on 2018-10-17, 20:46. Edited 1 time in total.

Im old goal oriented goatman, i care about facts and freedom, not about egos+prejudices. Hoarding=sickness. If you want respect, gain it by your behavior. I hate stupid SW limits, SW=virtual world, everything should be possible if you have enough raw HW.

Reply 11 of 71, by Falcosoft

User metadata
Rank Oldbie
Rank
Oldbie

A possible explanation for the weird 1280x1024 Quake 1 results:
As I have written above NVIDIA Maxwell2/Pascal cards have flawed VESA BIOS that prevents hardware based double/triple buffering. If you run Quake 1 on these cards the majority of VESA 2.0 modes (640x480/800x600/1024x768) produce horrible flickering/missing frames. The exception is 1280x1024 that works flawlessly on these buggy cards. So my theory is that Quake 1 in 1280x1024 video mode uses only 1 video page and uses software buffering unlike in other VESA 2.0 modes where it uses more than 1 video pages and hardware based buffering. This can explain the performance difference in case of other cards also. In case of software buffering the CPU+system memory speed is more important than the hardware page flipping speed of the video hardware (which can be restricted also by V-Sync).

Last edited by Falcosoft on 2018-10-17, 16:33. Edited 1 time in total.

Website, Facebook, Youtube
Falcosoft Soundfont Midi Player + Munt VSTi + BassMidi VSTi
VST Midi Driver Midi Mapper

Reply 12 of 71, by ruthan

User metadata
Rank Oldbie
Rank
Oldbie

I think that there would be multiple factors in play, not only one.
- when i compare Quake 1 vs. Q2DOS, Q2DOS performance scale fine across the cards with resolution for every card is 640x480 faster than 1024x768 and its faster than 1600x1200 - as expected, but with Quake 1, there is also problem of that 1024x768 is faster than 640x480 with lots of cards (mainly not NV) even old S3/ATI if are not video cards enhancers used (when are used, there much less exceptions and there are mainly on X58 platform, there has to be other problem, because all results are much slower than on Core 2 Duo - details are bellow )- i dont know details about video pages implementations, maybe other cards using more video pages for 640x480 or comparison with 1024x768?
- there is still not explanation for poor 320x200 Doom performance on more modern cards in comparison to PCI ones, up to Geforce 6 on Nvidia side its ok, but after there is performance drop to 1/3 to 1/2 in comparison with older cards, same on more modern ATI cards.
- my guess is that X58 has some bus problem, because dos raw CPU power is there, memory sub system also seems fine, but games videocards performance really sucks.. it could be problem with my HP O.E.M MB only.. we need more results - Agent007 has X58 too and lots of cards, even Geforce 1080 TI, i already wrote him.

Im old goal oriented goatman, i care about facts and freedom, not about egos+prejudices. Hoarding=sickness. If you want respect, gain it by your behavior. I hate stupid SW limits, SW=virtual world, everything should be possible if you have enough raw HW.

Reply 13 of 71, by ruthan

User metadata
Rank Oldbie
Rank
Oldbie

i added missing results from my 2nd lair and added few new rows for Dosbox on fast 8700K and Dos 8700K + Geforce 970 test.

I can confirm that screen blinking bug on Geforce 970 and 1030, in 640x480 in Quake 1, 320x200 and 1280x1024 are fine.. Also all Q2DOS modes -640,1024 and 1600 are fine too, there is some software work around.

Im old goal oriented goatman, i care about facts and freedom, not about egos+prejudices. Hoarding=sickness. If you want respect, gain it by your behavior. I hate stupid SW limits, SW=virtual world, everything should be possible if you have enough raw HW.

Reply 14 of 71, by AvalonH

User metadata
Rank Member
Rank
Member

Running an E7600@3.16ghz (stock speed) on G41-ES2L, 4GB.
Using MS-DOS7.1, only loading himem.sys and umbpci, no other tsr's, no emm386.

Radeon HD-4770 PCI-E video card.
Quake 1
1280x1024 with fastvid LFB+ - 154fps
vspeed.exe with fastvid
Banked - 330.43MB/s
LFB - 426.77MB/s

Nvidia Quadro NVS 300 PCI-E video card
Quake 1
1280x1024 with fastvid LFB+ - 105fps
vspeed.exe with fastvid
Banked - 40.92MB/S
LFB - 424.92MB/s

I always test the DOS compatibility of all my AGP and PCI-E graphics cards by running Rambo 3 in the same setup. Running 'Rambo.exe' or 'Rambo.exe r' (game setup option) will crash the system using both these cards.
Plugging in an old 1994 PCI Diamond Stealth 64 2MB (S3 Vision964) and it works. This game will not work with most modern cards.

Reply 15 of 71, by ruthan

User metadata
Rank Oldbie
Rank
Oldbie

AvalonH,
thanks for reporting, i already added results to sheet.

Its not competition but for Quake 1 1280x1024 you are 2 fastest results, 2x/3x faster that my best card. Radeon HD are beast for DOS, even for higher resolutions than 320x200 which already showed Phils benchmarks sheet, its good to know, because i nice results for 320x200 with other cards, but with resolution performance degrades really quickly.
Quadro NVS 300 PCI-E is big surprise, great passively cooled card with 17 W TDP, that what i like.

- It would be really nice, you can measure other resolutions, games and results without enhancers.
- Its funny that people usually reporting ram size, which really doesnt matters in DOS, if you have more than 64 MB (for Q2DOS benchmark), but dont report ram clocks
- thanks for compatibility tip, i added it to first post, same as vspeed.exe (i created no tools selection)
- BTW are using with this chipset some DOS sound card, if yes which one?

Im old goal oriented goatman, i care about facts and freedom, not about egos+prejudices. Hoarding=sickness. If you want respect, gain it by your behavior. I hate stupid SW limits, SW=virtual world, everything should be possible if you have enough raw HW.

Reply 16 of 71, by AvalonH

User metadata
Rank Member
Rank
Member
ruthan wrote:
AvalonH, thanks for reporting, i already added results to sheet. […]
Show full quote

AvalonH,
thanks for reporting, i already added results to sheet.

Its not competition but for Quake 1 1280x1024 you are 2 fastest results, 2x/3x faster that my best card. Radeon HD are beast for DOS, even for higher resolutions than 320x200 which already showed Phils benchmarks sheet, its good to know, because i nice results for 320x200 with other cards, but with resolution performance degrades really quickly.
Quadro NVS 300 PCI-E is big surprise, great passively cooled card with 17 W TDP, that what i like.

- It would be really nice, you can measure other resolutions, games and results without enhancers.
- Its funny that people usually reporting ram size, which really doesnt matters in DOS, if you have more than 64 MB (for Q2DOS benchmark), but dont report ram clocks
- thanks for compatibility tip, i added it to first post, same as vspeed.exe (i created no tools selection)
- BTW are using with this chipset some DOS sound card, if yes which one?

The Quadro NVS300 were selling for $4 on ebay, so I tried it, very cheap but you need a DMS to VGA/DVI cable.
The system actually has 8GB DDR2 800mhz. Mem command only shows 4GB of this in MS-Dos 7.1. I can access memory above 4GB but I have to load EMS64.exe (requires a 64bit CPU in Dos). Good for creating even bigger ramdisks.
https://board.flatassembler.net/topic.php?p=190123

I boot to DOS from a 128GB USB stick using fat32, no usb drivers, straight from BIOS boot menu. Then make a 2GB ramdisk, and copy DOS system files and any games, utils across.
I should point out the benchmark results I posted above with Quake 1 were the same if I ran it from the ramdisk or straight from the USB drive, it didn't make a difference.

For audio I use a PCI Audigy 2 ZS, after loading sbeinit.exe it requires AUDIGY12.EXE and then RayeR's serren.exe to get it working. Only the FM Adlib and GM/MT32 works, with the 8MB waveset loaded. I can't get the digital wav part of the SB emulation working on the G41 motherboard. The General Midi works very well and MT32 emulation is surprisingly good in games than ONLY use the default instrument set like Exterminator.

Did you try Rambo 3 in DOS on any of your cards, it fails on all modern cards I use. The game itself is just standard VGA compatibility, which I expected these cards to pass. It does not use any VBE Vesa modes.
Cards that I have tested that fail are: FX5200, 4200ti, NVS 280, NVS 300, AMD HD4770.
I will post more results later, I have a few with vspeed and Quake 1. I also tried UNIVBE 6.53. It seems to slow down newer cards, but on an old PCI 'Diamond stealth 64' 2MB it boosted Quake 1 massively in 640x480 from 20fps to 40fps.

Reply 17 of 71, by ruthan

User metadata
Rank Oldbie
Rank
Oldbie

You can have lots of ram in DOS, but important is how many of ram games are using, Q2DOS can allocate huge heap, but from my testing for benchmark, if you are using realmode (himem only) there is not performance improvement from 64 MB of RAM above, results are very similar with EMM386 NOEMS, with EMM386 EMS - game is not working at all, or it could be up to 4x slower.

I tried Quake 1 timedemo on ramdisk and at least on my systems not any performance improvements here.

About sound card - i was curios if there is not any new compatible DOS card to report, with your system if you want digital sound - you would need Yamaha 724/744 or Aureal Vortex V1/V2 cards, they are working even with ICH10 sought bridge , you have ICH7.

I didnt try Rambo 3, but cards is your list are quite modern from DOS perspective, so its not big surprise, all Nvidias cards even RIva TNT/128 have quite a lot compatibility problems, i think that it is because Nvidia started to make cards in Windows era, there never was primary DOS cards and i think they never fixed it.

Univbe - i something which i never fully used / understood, i tried it on few cards and usually i got unsupported chipset fail. If it can boost performance its good to know.

Im old goal oriented goatman, i care about facts and freedom, not about egos+prejudices. Hoarding=sickness. If you want respect, gain it by your behavior. I hate stupid SW limits, SW=virtual world, everything should be possible if you have enough raw HW.

Reply 18 of 71, by ruthan

User metadata
Rank Oldbie
Rank
Oldbie

I tested Q2DOS with Quake 2 Single player demo, its working fine, only name of demo files differs its now:
Q2Demo1.dm2 !not demo1.dm2, so commands are:
timedemo 1
map Q2demo1.dm2

Performance at least in Dosbox is same as with full version, so there is now not excuse, expect valid laziness, not to test every metric in my sheet.
I added pathb for CPU raw power testing to package too. For Q2DOS info check this link: http://dk.toastednet.org/Q2DOS/ for benchmarking make sense to keep benchmarking with one version of Q2DOS, not upgrade it in future.

Here is link for Q2DOS demo package:
https://www.dropbox.com/s/yd091nkze3vc371/Q2DoSDem.zip?dl=0 // Click on Download button - Direct Download in right up corner, i hate this DP *.zip browse feature..

Im old goal oriented goatman, i care about facts and freedom, not about egos+prejudices. Hoarding=sickness. If you want respect, gain it by your behavior. I hate stupid SW limits, SW=virtual world, everything should be possible if you have enough raw HW.

Reply 19 of 71, by ruthan

User metadata
Rank Oldbie
Rank
Oldbie

Three more cards tested i results are as every time unexpected:

  1. TNT2 M64 - its faster than all of tested Geforce cards from G2 to G970/G1030, Doom 200+ FPS. Only disappoinment is Q1 - 1280x1024 - there is Nvidia blinking screen problem (update with Q2 and same resolution its fine, so its some problematic implementation) and its too slow and its not fast enough for Q2DOS 1600x1200 - 30 FPS, otherwise everywhere 60+ FPS and even results without enhancers are quite good.
    If someone have proper TNT 2 / ultra, it would be very interesting to test it, i think that i could make Q2 1600x1200 60 FPS..
  2. Radeon X800 PCIE- i have multiple of these cards, but they not worked with my VIA board, i suspect that there are half dead (they are passively cooled so memory chips often died), or its something wrong with Gigabyte bios. Today i tested new one its working fine.
    Doom 200 FPS+ Q1 every where expect 1280x1024 60 FPS+, even there is quite good 50 FPS and Q2 60 FPS+ FPS, except 1600x1200 where it is 42 FPS, what is my best physical machine result yet - so i almost get my seal of quality mark:) for 60 FPS in every test. I wonder if faster CPU (i tested with Cure 2 Duo slimer - E6300 1800 MHz ) would push results to 60 FPS (Update: yes it can with Xeon56xx i got 60 FPS) or not and there are faster X800 variants - X800 XL, , X850 PRO, X850 XT/PE. Big advantage is these cards are also one of best for Win98 (my personal favorite) and have AGP variant too.. and i tested quite a lot DOS game with it and compatibility was better than with Geforce cards.. So it definitely one of the best options.
  3. Radeon HD 3450 PCI-E - even when this card is most crippled Radeon HD 3xxx, im disappointed, there better other cards, only good results are in Q2DOS with enhancers, without enhancers is very bad, Doom 40 FPS.

Im old goal oriented goatman, i care about facts and freedom, not about egos+prejudices. Hoarding=sickness. If you want respect, gain it by your behavior. I hate stupid SW limits, SW=virtual world, everything should be possible if you have enough raw HW.