It looks like the Voodoo1 driver I selected comes with the option to disable Vsync for D3D and Glide. For the Voodoo2 driver, I would need to install a driver add-on to do this. I assume the Voodoo2's driver defaults to vsync enabled?

For frame rates less than 60 fps, do you generally disable vsync when benchmarking, or leave it enabled?

feipoa wrote:

For frame rates less than 60 fps, do you generally disable vsync when benchmarking, or leave it enabled?

I would leave it disabled, because it would give a more accurate indication of how fast the system is.
Namely, if your framerate is less than 60 fps, the next-best framerate with vsync enabled is 30 fps.

Alright then. Since I'm almost finished with the comparison with Vsync enabled, I'll include results for both enabled and disabled cases. So far, the Voodoo2 is coming out ahead by a greater margin than I thought it would.

I did some tests a while ago on a 486 AMD (5x86@160mhz cpu, 32mb memory i think) and my favorite 3dfx games: screamer, screamer 2 (glide) and screamer rally (glide). The first game was unplayable at 640x480 as far as i'm concerned (no 3d/glide support) and the other 2 were pretty playable but only just... on some parts the fps was a little too low.

I tested with voodoo1, voodoo2 and voodoo3 pci and the framerate counter were my own eyes 😮) It did feel like the voodoo2 was the ''smoothest'' of them all in the toughest parts of the games so i would go for that even tough it was very very close between the 3 cards. I used the 12mb version of the card.

Also tested with a Cyrix 5x86 100GP at 120mhz (3x40mhz), i felt like the AMD was a slight bit faster. I thought the Cyrix at 160mhz or the AMD at 200mhz would be a fine running machine with these games but neither would go that high.

Just silly testing ofcourse, these games run best on a mmx or p2 imho. Nomatter what 3dfx card, it needs a bit of cpu power to run well. But still amazing to see how much power the 3dfx cards added to the 486, as in software mode none of these games did even come close to beeing playable (and even looked far worse ofcourse).

I have added the screamer games to my list of games to hunt down.

Yes, the Voodoo2 does have the cleaner output signal compared to the Voodoo1, however, it wasn't nearly as bad as I thought it would be. The pass through on the Voodoo2 also looks a bit clearer.

I have my results for vsync enabled, and half for vsync disabled. Do some games not work well with Vsync disabled? I was noticing that Forsaken didn't seem to detect my Voodoo1 display adapter when Vsync was disabled.

As an average over all the games tested, the Voodoo2 had a 13% improvement over the Voodoo1 when Vsync was enabled.

Enclosed are my results. I am not sure why Vogons is resizing the first and last image, but not the middle one. Click for full size images. The system specs are as follows.

• Biostar MB-8533UUD motherboard
  IBM 5x86c at 133 MHz (2x66 MHz FSB)
  64 MB FPM RAM (single stick)
  256 KB cache in write-through mode
  Matrox G200 16 MB (2D PCI graphics card)
  Audician32plus (ISA sound card)
  Promise UltraDMA 100 (PCI hard drive controller card)
  3dfx glide patches used whenever possible
  Voodoo1 and Voodoo2 were run at 640x486x16

The last chart is probably the most telling.

The percent increase in frame rate when using a Voodoo2 over a Voodoo1 on a 486 ranged for 0% to 36%. The most iconic game, GLQuake, showed an increase of 26.1% when Vsync was enabled and 18.8% when Vsync was disabled. In general, the values with and without Vsync were about the same, with a few exceptions. For those exceptions, the benchmark results generally inceased when Vsync was disabled.

As an average over all the games tested, the Voodoo2 had a 13% improvement over the Voodoo1 when Vsync was enabled, or a 11% improvement over the Voodoo1 when Vsync was disabled.

The Voodoo2 had slightly better signal quality in 3D and with 2D pass through. The Voodoo1 can sometimes appear too light, but this should be adjustable with the gamma controls contained in the driver. Aside from the generalised increase in frame rate, the other benefit of using the Voodoo2 is that running the card at 800x600 often came without any performance penalty. With a few rare exceptions, the Voodoo1 is limited to 640x480.

One benefit of the Voodoo1 is a smaller PCB. The Voodoo2 in my case just barely touches the back of the hard drive. Another benefit of the Voodoo1 is perhaps a slight increase in game compatability, although it does seem like there are work arounds for the Voodoo2 in most cases.

Of the games tested here, I personally consider them all playable, with the exception of Incoming, Unreal, and Turok 2. Turok 1 kept crashing after about a minute of game play, on both cards. Hexen II GL is marginally playable. There are many other early 3D games which play just fine on this system, however they are not included because they either do not have or I did not bother to look for a frame rate feature.

Nice chart, but:

IBM 5x86c at 133 MHz (2x66 MHz FSB)

Was the PCI bus running at 66 MHz as well? Because then the results are completely useless, as they would only show that the Voodoo2 is faster when overclocked. Also, almost no one runs his AMD X5 or Cyrix 5x86 with 66 MHz FSB, because finding a CPU and a mainboard that run stable at that speed (or support it at all) is almost impossible. Such a test should be done with everything running within specs, otherwise the results are of no use.

Besides, it would be nice to have information on how to reproduce these results. Which drivers and patches were used exactly? How did you measure the framerate esp. in games like Descent that don't come with a time demo? And how do you display the fps in Tomb Raider? Asked that already but did not get an answer so far.

The Voodoo1 and Voodoo2 were tested on the exact same system with identical settings. We are looking at a delta between the two Voodoo cards, not looking at absolute speed. The PCI bus is running at 33 MHz. Cache timings are at slowest and DRAM has 1 wait state. A system with an IBM 5x86c running at 2x66 is only marginally faster than the Cyrix 5x86 running at 4x33. I've compared this in the past and have both such systems setup.

feipoa wrote:

The Voodoo1 and Voodoo2 were tested on the exact same system with identical settings. We are looking at a delta between the two Voodoo cards, not looking at absolute speed.

The point was how much a Voodoo2 would improve frame rate on a 486. You could have very well used a Pentium 90 and the results would be similar.

At 66 MHz FSB, RAM speed is way, way higher than when running the FSB at 33/40 MHz, which is what most people do with these CPUs. A 486 does not use an FSB of 66 MHz. While these results might be legit for your configuration (which is ok since you asked the question), they don't apply to 486-class machines in general. Comparing hardware in an overclocked system makes no sense if you want to measure the difference when used in a system that is not overclocked.

You are free to perform your own tests and reach your own conclusions, so don't shit on others' results because you feel like they don't interest you. Every piece of data is important in the bigger picture.

Thank you for these feipoa, we finally have some numbers for 486 systems and Voodoo1 vs Voodoo2. 😀

derSammler wrote:

feipoa wrote:

The Voodoo1 and Voodoo2 were tested on the exact same system with identical settings. We are looking at a delta between the two Voodoo cards, not looking at absolute speed.

The point was how much a Voodoo2 would improve frame rate on a 486. You could have very well used a Pentium 90 and the results would be similar.

At 66 MHz FSB, RAM speed is way, way higher than when running the FSB at 33/40 MHz, which is what most people do with these CPUs. A 486 does not use an FSB of 66 MHz. While these results might be legit for your configuration (which is ok since you asked the question), they don't apply to 486-class machines in general. Comparing hardware in an overclocked system makes no sense if you want to measure the difference when used in a system that is not overclocked.

The memory speed with a 66 MHz FSB on a socket 3 isn't all that much higher than with a 40 MHz FSB because wait states must be added. According to cachechk, a Cyrix 5x86-120 has a DRAM read speed of 55 MB/s, whereas the IBM 5x86c-133/2x has a read speed of 70 MB/s. L2 read speeds of a Cyrix 5x86-120 were recorded at 96 MB/s, whereas the IBM 5x86c-133/2s has a cache read speed of 102 MB/s.

In the past, I have compared the results of an Am5x86-160 in GLQuake vs that of the IBM 5x86c-133 and found the results to be not all that dissimilar. I only chose the IBM 5x86c-133/2x for this recent run, not to make you mad, but because the system has the Voodoo2 already in place and games were setup.

Based on your choice of wording, I am getting the impression that your intent is to discredit this work and to express rage. Would you like me to run the same set of tests using an Am5x86 running at 160 MHz? And if I were to do this, what would be your comment if the percent difference between the Voodoo1 and Voodoo2 are nearly identical to what I have already presented?

Hold on now! WHERE ARE THE SCREAMER RESULTS?? 🤣

Good job and thanks, very appreciated! Pretty good fps in some games i must say, especially for a 486 system.

It would be interesting to compare Voodoo 1 and Voodoo 2 at the same clock speeds in this situations.

could You show on 1 or 2 examples the difference to a system on standard bus speeds? just so we could see how much this really changes things..
also, where' s the screamer, pretty please ^^

thanks for the input, that was quite a lot of work 😀

i do assume that none of these programs automatically takes advantage of the new functionality in voodoo2, right? so this actually shows that voodoo2 does a voodoo1's job faster?
where there ever techdemos released that would leverage for example the new triangle setup and allow comparison with a v1?

meljor wrote:

Hold on now! WHERE ARE THE SCREAMER RESULTS?? 🤣

Good job and thanks, very appreciated! Pretty good fps in some games i must say, especially for a 486 system.

I was reading on Wiki that Screamer is a DOS-based software mode game. Is there a 3dfx patch for it and does the game have a means to run either a timedemo or instantaneous frame rates?

Do Screamer 2 or Screamer Rally provide a means to run in 3dfx glide mode? And is there a means to measure frame rate via a counter or timedemo?

The Serpent Rider wrote:

It would be interesting to compare Voodoo 1 and Voodoo 2 at the same clock speeds in this situations.

It would be! That would tell of us the increase in performance of the Voodoo2 is mostly a result of the higher clock rate. I'll take a look at this the next time I have the system out of the closet.

mrau wrote:

could You show on 1 or 2 examples the difference to a system on standard bus speeds? just so we could see how much this really changes things..

Sure, I don't mind expanding on this work. Would you like me to test an Am5x86-160? I do not predict significant deviance from the percent difference already presented. The Am5x86-160 can run with the fastest L2 cache timings, while the IBM 5x86c-133 must run with the slowest. The Am5x86-160 can also run with the fastest DRAM timings, while the IBM 5x86c-133 needs a wait stated added. You'll have to wait some time as I have to make up a test bed, clone the HDD, and find all the same hardware components. I currently have another testbed in place, which will need to conclude first.

mrau wrote:

so this actually shows that voodoo2 does a voodoo1's job faster?

That seems to be the prevailing hypothesis at the moment, at least according to The Serpent Rider's request.

mrau wrote:

where there ever techdemos released that would leverage for example the new triangle setup and allow comparison with a v1?

I'm not familiar enough to provide sufficient comment. Maybe someone else has some idea?
-
-
I have some results here of a Cyrix 5x86-133/4x vs an Am5x86-160 using a Voodoo3. GLQuake with Cyrix was 27.6 fps, whereas it was 27.5 fps on the AMD. This was done on a SiS496-based chipset motherboard. IBM 5x86c-133 and Cyrix 5x86-133 results are nearly identical, recorded results may even be in that thread somewhere. Re: Modern graphics on a 486

Would it be fair to assume that a Voodoo3 is identical to the voodoo2 in these tests?

dirkmirk wrote:

Would it be fair to assume that a Voodoo3 is identical to the voodoo2 in these tests?

Umm, it might be marginally faster due to higher memory bandwidth.

dirkmirk wrote:

Would it be fair to assume that a Voodoo3 is identical to the voodoo2 in these tests?

I suspect it is close on such an under powered system. Is there interest in a Voodoo2 vs. Voodoo3 comparison on a socket 3 board? I would have to use a board with a SiS496 chipset for this.

feipoa wrote:

I was reading on Wiki that Screamer is a DOS-based software mode game. Is there a 3dfx patch for it and does the game have a means to run either a timedemo or instantaneous frame rates?

Do Screamer 2 or Screamer Rally provide a means to run in 3dfx glide mode? And is there a means to measure frame rate via a counter or timedemo?

there is 3dfx mode/patch, i'm not sure which; as for fps counter i have no clue myself; i remember youtube videos showing tests, but i cant recall any details;

feipoa wrote:

The Serpent Rider wrote:

It would be interesting to compare Voodoo 1 and Voodoo 2 at the same clock speeds in this situations.

It would be! That would tell of us the increase in performance of the Voodoo2 is mostly a result of the higher clock rate. I'll take a look at this the next time I have the system out of the closet.

and what are we going to do if it turn out voodoo2 is slower clock for clock? :>

feipoa wrote:

mrau wrote:

could You show on 1 or 2 examples the difference to a system on standard bus speeds? just so we could see how much this really changes things..

Sure, I don't mind expanding on this work. Would you like me to test an Am5x86-160? I do not predict significant deviance from the percent difference already presented. The Am5x86-160 can run with the fastest L2 cache timings, while the IBM 5x86c-133 must run with the slowest. The Am5x86-160 can also run with the fastest DRAM timings, while the IBM 5x86c-133 needs a wait stated added. You'll have to wait some time as I have to make up a test bed, clone the HDD, and find all the same hardware components. I currently have another testbed in place, which will need to conclude first.

imho this does necessarily need to be tested on the fastest hardware - the excessive power of the cpu might drive expectations of people with slower cpus..

feipoa wrote:

mrau wrote:

so this actually shows that voodoo2 does a voodoo1's job faster?

That seems to be the prevailing hypothesis at the moment, at least according to The Serpent Rider's request.

so now the question arises - what calculations can be moved to the gpu with a voodoo2 and hoow does one do that; i have never seen any docs on this

feipoa wrote:

I have some results here of a Cyrix 5x86-133/4x vs an Am5x86-160 using a Voodoo3. GLQuake with Cyrix was 27.6 fps, whereas it was 27.5 fps on the AMD. This was done on a SiS496-based chipset motherboard. IBM 5x86c-133 and Cyrix 5x86-133 results are nearly identical, recorded results may even be in that thread somewhere. Re: Modern graphics on a 486

that's quite playable! was that in 640x480?

Screamer (1) is a dos game without a 3dfx patch unfortunately, screamer 2 and screamer rally both have 3dfx patches ( for voodoo2 see this: Voodoo 2 DOS Glide compatibility matrix )

I don't know of a way to measure the fps in dos glide games..