VOGONS


Reply 40 of 91, by BushLin

User metadata
Rank Member
Rank
Member
The Serpent Rider wrote:

So main question is what fastest videocards are usable with SS7 machine

The one with least driver overdraw. Best choise should be 3dfx, early GeForce and S3.
Techically you can put in there extremely fast card like GeForce 6600GT, but it will perform overall worse due to quite sophisticated drivers.

This was really early in the thread, I think he told you everything you need to know at that point. You're so CPU limited on the platform that even GF4/FX are probably giving you worse performance than something older. GF6 series doesn't offer decent Windows 98 drivers and for performance on XP... Jesus, use something else.

Screw period correct; I wanted a faster system back then. I choose no dropped frames, super fast loading, fully compatible and quiet operation.

Reply 41 of 91, by ruthan

User metadata
Rank Oldbie
Rank
Oldbie

For those who are interested, in this thread you can find test of Geforce 3 Ti 200 AGP with Core 2 Duo VIA880 system, its nice to know there are limits of this card in DOS, know that there are better dos cards, but there are zillion test in windows from other people..
Old+Modern videocards pure DOS benchmarking- which one is fastest?, need your numbers + analysis, 320x200 to 1600x1200!

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 42 of 91, by ruthan

User metadata
Rank Oldbie
Rank
Oldbie

I revisited Phil Best video cards for SS7 for video series.. I saw it multiple times, but i usually watch it on second monitor and doing something else primarily, but yersterday i really aim on it and tried to harvest some data from it.

Its tremendous piece of work, probably best one on the topic, but its 3 years old and when i look on it from results perspective there are few major flaws, not enough data from our case. I dont do game testing, but i do some sort of performance testing and analysis for living so i know something about it:

  1. All cards where tested, within very limited test suited - with 3DMarks - i dont care about these results at all there are missleading for real Win98 games era for which these are actually used, they have too much bias to new features, which are not used in games too often.
    - 2 OpenGL games on same engine (Quake I and Quake II).
    - 1 Direct 3D game (Incoming), which has problem with modern Nvidia cards (G3 a G4), so they arent tested because some glitch, i wonder if this glitch is really unfixable.
    - DOS - Doom - all cards are fast enough 90 + FPS, some do 140, but game is 35 FPS capped, Quake there is not info about resolution, but i bet that it is 320x200 all cards here have some performance, because it is in reality CPU test, any card is not limiter for 320x200, interested results would be 640x480.
    There are these dos bemchmarks - 3Dbenches etc.. but its very silimar, its not good test, do what we want to discover.
  2. Fastest tested cards - Geforce 4 TI 4200 (slowest from Tis G4) and Geforce 3 200 Ti (slowest from Tis G3), are fastest cards on paper and fastest cards in test, even they are using more modern drivers that rest of Nvidia cards. Phil excluded them from non most of Nvidia testing comparision, because his point of view - there are too modern, not period correct etc.. and they have Incoming problem. So i dont really see reason why would be more fast cards with a bit more modern drivers like Geforce 5/6 faster, so far faster card, better result and there is not such big jump in drivers version.
  3. Other flaw of Dos test, is there are not Fastvid and MTRR results, it can enhance results with some card a lot.. There is also new discovery of VID_NOPAGEFLIP 1 command in Quake1, which can improve Quake performance in high resolution a lot.
  4. There are interesting results with Radeon 9200, but its like with Geforce 3/4 in later comparsion is used Radeon 7200 SDR instead of it.. and there much faster Radeon 9xxx with more computing units.. so i believe that results would be much faster.
  5. S3,Matrox, SIS, Kyro - have some major flaw with Direct3D or OpenGL performance, so they are lost of time from max performance point of use.
  6. 3Dfx - there si still some gain for high resolutions with Voodoo 5 5500, Voodoo 4 4500 is same as Voodoo 3 3500. All these cards are still significatly slower that Geforce 3 /4 , but because they are excluded from lots most of general comparison.. I would say that test is designed to show 3Dfx as winners 😀 Yeah Glide is nice, but you can have Nvidia as primary card and Vooodoo as secondary.
  7. Geforce 3/4 are destroing results of cards in antialiasing 2x, 4x test.. even when without it other card have similar performance, when you turn it on, Nvidia cards are much faster.
  8. Even when some cards have in last tested resolution, big performance overhead 90 FPS etc.. there are not higher resolutions tested (i dunno can OpenGL Quake do 1600x1200, 1024x768 for Q2 is not sure for sure, same with Incoming, i already wrote same about Dos test), so performance of better cards looks similar - because is limited by CPU, but when would test higher resolution you would see real benefic of faster cards.. and im quite sure that performance would be reasonable around ~60 FPS.
  9. Update: there are also quite a lot of AGP bios setting, unless they are here just to show that there is AGP support, it have influence on performance - as i understand Phils test he started with Nvidia cards and after that i has to switch MB for 3dfx testing, but 3dfx card ignore AGP features, maybe, maybe Nvidia AGP settings (AGP turbo disabled etc..) is not optimal for other cards.

I checked also results within ultimate 6x86 thread are with Matrox G200, its slimmer there is some scalling with Quake / Quake 2, but Phils results and some cards have much faster results, so cpu has some influence but results with faster cards could be 3x faster.. So not much interesting data here.

For these results i cant really say that faster cards and but idea and that some more faster cards will not have bigger performance.. Its far from closed case and all important aspects discovered state..

Last edited by ruthan on 2019-06-20, 17:25. Edited 3 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 43 of 91, by BinaryDemon

User metadata
Rank Oldbie
Rank
Oldbie

I think it's important to look at all options, because period correct hardware is becoming increasingly expensive. Sure It's nice to recommend a Voodoo3 to everyone but not everyone has the budget for that. If I can get a <$20 Radeon 9250/Geforce 5200 and achieve basically the same performance in WIn98, then that's important to know.

Check out DOSBox Distro:

https://sites.google.com/site/dosboxdistro/ [*]

a lightweight Linux distro (tinycore) which boots off a usb flash drive and goes straight to DOSBox.

Make your dos retrogaming experience portable!

Reply 44 of 91, by cyclone3d

User metadata
Rank l33t++
Rank
l33t++
BinaryDemon wrote:

I think it's important to look at all options, because period correct hardware is becoming increasingly expensive. Sure It's nice to recommend a Voodoo3 to everyone but not everyone has the budget for that. If I can get a <$20 Radeon 9250/Geforce 5200 and achieve basically the same performance in WIn98, then that's important to know.

Well.... since I will be working on a project to clean out my un-needed / un-wanted hardware I will be also working on testing absolutely everything before putting it up for sale. The CPU I will be using for SS7 platform will be a K6-III+ 450 that I will overclock as much as possible.

Motherboard will be an ASUS P5A-B or P5A as those will work fine at the 120Mhz fsb.

I also have a few VIA based boards I could test with.. but from my experience form back in the day when I worked at a computer shop when this stuff was new is that the ASUS P5A / P5A-B was a decent amount faster than the VIA solutions.

I'll definitely be using an SSD for the drive so HDD lag won't be a factor. Probably also use a Promise PCI SATA controller for even faster transfer speeds but not sure if it will make a difference at all.

Yamaha modified setupds and drivers
Yamaha XG repository
YMF7x4 Guide
Aopen AW744L II SB-LINK

Reply 45 of 91, by ruthan

User metadata
Rank Oldbie
Rank
Oldbie
BinaryDemon wrote:

Well.... since I will be working on a project to clean out my un-needed / un-wanted hardware I will be also working on testing absolutely everything before putting it up for sale. The CPU I will be using for SS7 platform will be a K6-III+ 450 that I will overclock as much as possible..

Motherboard will be an ASUS P5A-B or P5A as those will work fine at the 120Mhz fsb.

It would be lost of work, because you have lots of crazy strange pieces of HW.. but are also often promising that you will test it and you usually dont, as far as i remember:)

Phil overcloked k6-3+ 400 to 550, so i thing that k6-3+ would be safe to do same, its would be nice to some results comparision, pleas check how AGP speed scale with FSB setting, with some configurations could faster FSB mean slower PCI- AGP at least for SC370 MB, i dunno if for SS7, i have to check manuals. With 120 FSB x 5.5 multiplier with good cooling you can get maybe, maybe to 660 MHz.
Other interesting thing is it uses ALI chipsets all Phils test where done with VIA MPV3 chipset.

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 46 of 91, by ruthan

User metadata
Rank Oldbie
Rank
Oldbie

Hmm, this Asus MB (Asus p5ab-107) has cpu and mb temperature sensor and fan fan alarm, its first one, when i see it.. but there is note - with optimal Hardware monitor only.. i dunno what it means some special card?
Update: It seems to be optimal on board HW - with uses some ISA legacy addresses.

FSB - AGP clock scale linearly with FSB so for 120 Mhz, its 80 MHz.. for some FSB is underclocked but only slightly to 60 MHz. And if understand correct there is 6x multiplier for AMD k6-X so maybe, maybe 6*120=720 MHz..

Other interesting info is: That memory sticks with more than 18 chips are not supported.
It also has - SPDIF for digital audio, its some C-Media chip so probably not too much fun in Dos, but it claims that it has Dos compatibility..

So far, this Asus manual is much better and more detailed and others from same period which i saw, thumbs up!

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 47 of 91, by cyclone3d

User metadata
Rank l33t++
Rank
l33t++

Yeah, 120Mhz means 80Mhz AGP and 40Mhz PCI. But that is what I used back in the day when I had a K6-2 550 running at 660Mhz.

I really hope to get to 720Mhz. I do have much better cooling solutions than back in the day so heat won't be a problem. I'll probably mod a heatpipe cooler to use on it. Also have a peltier, but I know that will just end up icing up the socket and I don't fancy frying a board.

If neither of the K6-3+ CPUs I have will do 720, I'll try with my K6-2+ 550.

None of my boards have the onboard audio option.

There are actually also some undocumented fsb settings. I used to have all the different settings written down somewhere but that was years ago. Pretty sure the clockgen supports either 133 or 140Mhz fsb.. but of course the board will not run stable at either. I'll have to look up the clockgen data sheet to double check. Back in the day, everything was discovered by trial-and-error. Now it is way easier because all the data sheets are easily. accessible.

I currently have 3x P5A-B and 1x P5A as well as a broken P5A... worked when I got it.. then I went to swap out the CPU for a faster one and it wouldn't POST again. The broken is the newer revision that has trouble with the + CPUs anyway so not a huge loss.

Yamaha modified setupds and drivers
Yamaha XG repository
YMF7x4 Guide
Aopen AW744L II SB-LINK

Reply 48 of 91, by havli

User metadata
Rank Oldbie
Rank
Oldbie

720 MHz is impossible.... well unless you want to use liquid nitrogen or similar ways of cooling. With some luck you get little over 600, but not more than that.

For most S7 boards external cache is the limiting factor when overclocking. So you can try to disable it in order to run at faster FSB. There is some performance loss of course but I think rather small with on-die L2 equipped CPUs. For example my Gigabyte GA-5AX can do 125 MHz with external cache and 135 MHz without. AOpen AX59 Pro runs at 124 MHz without cache, but only 112 with cache.

HW museum.cz - my collection of PC hardware

Reply 49 of 91, by ruthan

User metadata
Rank Oldbie
Rank
Oldbie

Im not big in overclocking (there are guys as Agent007 - but i dunno if he is in SS7), i just mentioned it as max, but i saw some thread where someone claim that got 700 MHz working (there are claims like this, but not proof: the k6-3+ is a good overclocking i upgraded my old system. i got a k6-3+ to 750mhz pretty stable i might post some benchmarks. k6-3+ 750 vs tbird 750). Im sure that somewhere would be site with overclocking records on air ordered by CPU..
There are some threads, someone claims that ALI chipset could reach 120 FSB and its better than MVP3 for overclocking:
Re: Overclocking AMD K6-2+/500

So overclocking guide:
http://cpu-central.com/articles.asp?article_i … 79&decor_int=28

Otherwise Asus p5a-b regarding of manual p5ab-107 has on board sound card and FSB 120 and 6x multiplier options, i dont read p5a manuals.

Last edited by ruthan on 2019-06-18, 20:52. 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 50 of 91, by cyclone3d

User metadata
Rank l33t++
Rank
l33t++
havli wrote:

720 MHz is impossible.... well unless you want to use liquid nitrogen or similar ways of cooling. With some luck you get little over 600, but not more than that.

For most S7 boards external cache is the limiting factor when overclocking. So you can try to disable it in order to run at faster FSB. There is some performance loss of course but I think rather small with on-die L2 equipped CPUs. For example my Gigabyte GA-5AX can do 125 MHz with external cache and 135 MHz without. AOpen AX59 Pro runs at 124 MHz without cache, but only 112 with cache.

I dunno about that. Like I said before, I was running a K6-2 550 at 660Mhz back in the day when everybody back then said it was impossible. And that was on a sucky air cooler.

The K6-2+ and K6-3+ run at a lower stock voltage and seem to be able to overclock a lot easier. I put one of those in an old laptop with a really tiny cooler and was able to run it at 550Mhz at stock voltage.

I''m going to see how high I can get. Maybe I won't hit 720Mhz.. but trying to get there will be fun.

Not only are the coolers infinitely better now, but the power supplies back then were also really, really sucky as well. I'll be using a Seasonic 350w ATX or higher PSU for the testing.

Yamaha modified setupds and drivers
Yamaha XG repository
YMF7x4 Guide
Aopen AW744L II SB-LINK

Reply 51 of 91, by ruthan

User metadata
Rank Oldbie
Rank
Oldbie

I checked once more Phils K6-III+ Geforce 5500 PCI video:
https://www.youtube.com/watch?v=key3tXC0axI

There are some cards, but not comparision with other cards with previous test, but.. when we compare it with SS7 videocards test - https://www.youtube.com/watch?v=md5Wl7t-VfI I dunno fi these test are in 32 or 16 bit, he has fancy graphs, but would say that he needs ad more legend /setup info, arent often even in description.. I also missed driver version info.

GLQuake 1024x768 - 93 FPS => Geforce 3/4 - can do 123 FPS in 16bit; and exactly the same in 32bit.. i not sure, it is not mistake on Phils side, but ok.. G5 PCI is 25% slower..
Quake 2 1024x768 - 70 FPS => 16 bit - G3 - 75 FPS /v G4 -68 FPS; 32 bit same exactly same FPS.. again i would expect a bit worse results than in 16bit - G5 PCI is 5% slower..
Incoming 1024x768 - 58 FPS => there is really interesting that G3/G4 had glitches (excluded from results).. G5 test is ok and valid.. 16bit - G3/G4 no results but even TNT 2 Pro can do 90 FPS; 32 bit - G3/G4 again no result, but TNT2 - 62 FPS and nothing faster tested; .. If results would be 16bit, they sucks, but if they would be 32 bit an no glitches, its very good.

I lazy to check V3 results, i previous post and video and i would say that Geforce 5500 even with in PCI version pretty limited by PCI, i know, i know that V3 doesnt use AGP feature either.. has same or very similar results as Vooodoo 3, so if someone care more about money and can bypass Glide somehow.. and probably some compatibility problems, but want to use 32 bit and some AA, Geforce 5 is good option.

But i still thing that there could be maybe even much better cards like Radeon 9500-9850 or Geforce 6 AGP, for example i have X800 cards in my system are working fine with most played Win98 games with fast CPU can do 90 FPS in Quake III in 1600x1200 and its only one generation from Radeon 9500-9850, that radeon which tested Phil - 9200 is actually some Radeon 8xxx (R2xx) group chip that real Radeon 9xxx series (R3xx with 256bit bus, DirectX 9 card, SM 2, new AA technology, multisampling).. is lowend previous generation with name of new generation - what is quite typical for ATI and Nvidia and this era..

Last edited by ruthan on 2019-06-20, 17:29. Edited 4 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 52 of 91, by The Serpent Rider

User metadata
Rank l33t++
Rank
l33t++

ATI drivers are too complicated to work sufficiently fast on Pentium MMX/AMD K6-2.

I must be some kind of standard: the anonymous gangbanger of the 21st century.

Reply 53 of 91, by BinaryDemon

User metadata
Rank Oldbie
Rank
Oldbie
The Serpent Rider wrote:

ATI drivers are too complicated to work sufficiently fast on Pentium MMX/AMD K6-2.

Has anyone tried installing the drivers without installing the entire Catalyst suite? I remember that generally this resulted in worse performance, but maybe for this specific SS7 scenario it could reduce overhead?

Check out DOSBox Distro:

https://sites.google.com/site/dosboxdistro/ [*]

a lightweight Linux distro (tinycore) which boots off a usb flash drive and goes straight to DOSBox.

Make your dos retrogaming experience portable!

Reply 54 of 91, by ruthan

User metadata
Rank Oldbie
Rank
Oldbie
The Serpent Rider wrote:

ATI drivers are too complicated to work sufficiently fast on Pentium MMX/AMD K6-2.

You can be right, but give us some numbers as proof..

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 55 of 91, by ruthan

User metadata
Rank Oldbie
Rank
Oldbie

3 more video from Phil to analyze i knew that they exists, but they need deeper look for out case..
Geforce 4 4600 Ti review
https://www.youtube.com/watch?v=hBsFpK13wFc&t=303s
Radeon 9700 Pro review:
https://www.youtube.com/watch?v=zejKjmpUtWc
Nvidia Win98 speed evolution:
https://www.youtube.com/watch?v=HRhm4aGNI3o

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 56 of 91, by ruthan

User metadata
Rank Oldbie
Rank
Oldbie

Geforce 4 4600 TI
If dont overlook and overheard something there is info which CPU was used, so this video can us tell only, that with some probably fast not SS7 cpu, it could be quite fast in Win98. I checked even comments..

Radeon 9700 PRO AGP
Once agains benchmark without test setup info.. big flaw. From Phils first slide its +76% faster (i even used Excel for it) than Geforce 4 TI 4600 in same machine, but its XP test and quite sure its with faster than SS7 CPU.

Nvidia drivers version influence in Windows 98:
Tested with 1 GHz PIII, so its hard to say with it measn for SS7, but..
Expendable(Direct3D-640x480x16)
- TNT2 - 3% decrease up to driver 61.76, so its virtually nothing, because some bugfixes degraded performance.
- Geforce 256(=1) /G2 - there is huge gradual performance decrease - from 95 to 52 FPS in 71.XX driver.
- Geforce 4 - there not results for 7x and 8x driver, but between 28x32 and 61.76 si decrease few percent, and
its 30% difference between G4 and older cards with 61.76 driver.
- Geforce 5800 / 5950 there is decrease i would say 15% with 71.xx driver, but in 81.98 its back and otherwise difference is < 10%, so no big deal.
Expadable(Direct3D-1024x768x32)
- TN2 is the same
- Geforce 1 /2 - its very similar again huge decrease.
- Geforce 4 - now there is big difference from 77 fps to 50 Fps.
- Geforce 5/5 - except bad 71.xx thre is i would say 5% difference.

Expendable(Direct3D-1600x1200x32) // Very similar to 1024x768.
Drakan 640x480x16 / 1024x768x32- very similar to Expendable 1024x768
Drakan 1600x1200 - simlar to Expendable 640x480
Unreal Tournament - 640x480x16 - there are same gaps for all cards except bat GF4 performance with 56.64.. its
gradual, but from oldest, fastest to newest its 15%.
Unreal Tournament - 1024x768- TN2 is slightly gradually faster, other otherwise its similar to 640x480 but over difference is even smaller - 10%.
Quake 2 640x480x16 (OpenGL) - there not big difference -/+ 10, for Geforce 4 and FX even less.
Quake 2 1024x768x32 (OpenGL) - +/- 15% - best driver is 30.82,, for Geforce 4 and FX even less.
Quake 2 1600x1200x32 (OpenGL) - +/- 15% - best driver is 30.82,, for Geforce 4 and FX minimal gap.
Quake 3 640x480x16 (OpenGL) - except TNT2 where is 10% last its in few percent +/- here we can see for which game where drivers probably optimized, or great IDsoft code, i dunno, , except early 6.31 and older drivers.
Quake 3 1024x768x32 +/- few percents, except early 6.31 and older drivers.
Quake 3 1600x1200x32 - +/- few percents, except early 8.05 and older drivers.
MDK2 (Direct3D) 640x480x16 - there is 15 decrease from 45.23 driver to newest, there is small gain from 56.64 for G4/ FX.
MDK2 1024x768x32- except 6.31 drivers and earlier, there is max 10% difference, less with G4/G5.
MDK2 1600x1200x32 - - except 8.05 drivers and earlier, there is max 10% difference, less with G4/G5.
Serious Sam - 640x480 - decrease for TNT2, other cards are ok.
Serious Sam - 1024x768 - decrease for TNT2, other cards are ok in +-5%
Serious Sam - 1600x1200 - +- 15%, except 81.92 there where FX 5800 an Geforce 4 are 25% slower, G 5900 is only few
percent slower.

Overall Nvidia cards performance:
- This video shows that newer Geforce cards sucks even with much faster 1 GHz PIII with Win98, so its not primary problem of slow SS7 cpu.
- This also first and so far best of show, how bad performance of G4 and G5 cards are in Windows 98, regardless of drivers, because their performance is not degrade too much with drivers as for other cards.
- Even these much faster cards, are often slower or at same level as much older cards.
- There's some Direct3D games where are Geforce 5 cards in 1/3 slower than older cards, in OpenGL games are usually slightly slower, but rarely faster, some Direct 3D games is diffence smaller 15%, not 30%.

Between Geforce 5 and 6 is quite big difference in raw power, so that still could be fast, but after this analysis i believe in it a bit less, but its about measurement not believe.. Nothing is changing for Radeon 9700, as shows XP video it has lots of horse power, now we need to see, how bad are Windows 98 drivers and how much is limited by CPU.

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 57 of 91, by _UV_

User metadata
Rank Newbie
Rank
Newbie
ruthan wrote:
I added still unanswered questions to 1st post. Radeon R3xx - 9500-9800 XT Windows 98 SS7 performance. Nvidia G6/7 Win 98 perfor […]
Show full quote

I added still unanswered questions to 1st post.
Radeon R3xx - 9500-9800 XT Windows 98 SS7 performance.
Nvidia G6/7 Win 98 performance.
Windows 2000 / XP later Nvidia / ATI cards performance, would it be better than in Win98? Would here faster card make sense?
ALI and VIA chipsets performance difference? At least regarding 1 of Phils video there is quite big Windows 98 performance difference between 2 SS7 MBs with different chipsets.. and DOS performance difference is significant too.

Radeon R3xx excellent performance, especially if you match it with proper drivers working with SS7 era games
Nvidia G6/7 same, but searching for drivers fully compatible with old games between hard and impossible, R3xx a bit better

In both cases this cards isn't for SS7 or even later S370 Tualatin setups.

1. AGP signalling, some cards not compatible with AGP 1.0 (1x/2x) or have serious problems with power delivered by mobo over bus.
2. Drivers too new for a target period (i assume 96-2002). Not all games are like Q2. For example pretty popular title - MotoRacer hates CPUs faster than 233MHz and accelerator cards higher than Voodoo 3 or TNT2, it's either too fast to play or can't use 3D acceleration.
3. This cards and drivers require Dx9 or atleast Dx8 to work, and have pretty significant overhead for early CPUs. That is a territory of P4/Athlon XP/A64.

What is better for pure Dos and Win98

in the name of the topic means you targeting even lower 1994-2000. So you don't need anything stronger than a GF2 or Radeon 8500 paired with 1GHz Coppermine/Athlon. For "pure" DOS 200MHz Pentium or K6 with TNT and Voodoo2 already overkill.

As for Windows 2000 or XP please don't fool yourself, select a proper period for which you want to build your PC.

ALI and VIA chipsets heavily depends on mobo brand and how they treat customers. IMO AOpen and Chaitech do the best they can, others either not use all potential or made this boards much worse than for Intel chipsets.

Reply 58 of 91, by ph4nt0m

User metadata
Rank Member
Rank
Member

Radeon 9700 and 9800 are AGP 1.0 compatible. They also have an aux power connector, floppy style for 9700 and HDD style for 9800. They don't draw much power from the AGP slot. It's mostly from +5V and +12V of the aux connector.

This review in Russian tells the story. Here are the stats in burn mode:

9800 Pro, 380MHz core / 680MHz memory:
3.3V AGP: 9.23W
5V AGP: 2.39W
12V AGP: 1.76W
5V aux: 15.35W
12V aux: 1.78W

Overall 13.38W from the AGP slot and 17.13W from the aux conn.

9800XT, 412MHz core / 730MHz memory:
3.3V AGP: 4.42W
5V AGP: 2.50W
12V AGP: 2.01W
5V aux: 11.04W
12V aux: 18.25W

Overall 8.93W from the AGP slot and 29.29W from the aux conn.

My Active Sales on CPU-World

Reply 59 of 91, by ruthan

User metadata
Rank Oldbie
Rank
Oldbie

_UV_: Thanks for reply lots of urbans legends here.. some myth-busting is needed 😀

1) Besides Radeons, there at least some Geforce 6 cards with AGP 3.3 and power connector.

2) I guest that not really Win98 would have problem, but maybe very old Win95 games from start of 3D era would have problems, but they were always problematic.. even with some more, modern 3Dfx cards and i guest that there would be some solutions and patches. I dont play MotoRacer, but it need to be discussed on game basis, usually someone is able to find some solution as for problematic Dos games.. i dont want to fiddle with it fastest AGP cards are not for you.. As i wrote before, i believe in secondary PCI cards as Voodoo 2/3 for these games. Its good quite good solution, i thing that people trying there best to find research why dont to do this.. its better be more positive.

3) I dont thing that something like card requiring DX9 exists, DirectX is backward compatible.. You can still play DX6/7 games even on Windows 10, yes are issues but its not about that cards not compatible with older DX versions at all. Yes these cards are particularly optimized for these, but have more raw power - so situation is quite balanced.. even drivers and card architecture have some history, co its usually something entirely new.

4) Period correctness - i was using XP (OS is from 2001) with K6-2 350 + Voodoo 3 (1999 HW), it wasnt something entirely stupid, when you had enough RAM. Windows 2000 i thing would be much HW resource friendly.. and even lots of developers used them by day. Lots of people preferred Windows NT agains Windows 95/98.. I pure theory Windows 2000 could be sweet spot and there is quite lots of hack how to make even some XP games running.

I think that people are overzealous with that period correctness. I personally have one retro fast machine with X58, where even Windows 98 are running for lots of games fine, im even able regarding my tests to make 85% Dos games from 160 running with sound and i just need 1 more machine with ISA sound card for rest.. i dont want to have zillion of machines per years, i rather safe some space for old Mac / Amigas and Consoles. I started with 286 and monochrome monitor with PC, but i dont need such experience anymore i like fast machines.

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.