Bushwack, I was pondering your GLQuake results more.

I am suspicious of that V1000 getting 23 fps at 640x480. I just ran Hexen II in both GL and VH2 modes on my Screamin' 3D and GL was what I would call unplayable. It seemed to range from 5-12 fps or so. I was using the final MiniGL. VH2 is much faster and is quite playable at 640x480 even with full anti-aliasing on (r_antialias 7). I suppose that it's possible that Hexen II is considerably more demanding than Quake, even though they use the same engine.....

Also, did you run a timedemo or a timerefresh? Timedemos are much more useful measures of performance because there is more texture loading and more practical in-game situations.

Anyway, I've now posted a gallery of vHexen II images on V1000. Full anti-aliasing enabled (r_antialias 7). There are a few software mode captures in there too.

My Screamin' 3D card seems to be dieing. It locks up occasionally, mostly in Windows. Sometimes I get screen artifacts when it freezes. I've tried two different mobos and removed all cards other than a NIC (tried two of those too). Tried some BIOS settings and several CPUs too. Fortunately, I can get it to behave in 3D games though for screenies.

Everything, including both the game and drivers are at default settings. I am using drivers specific to that card, not reference drivers. Also all drivers are 1997 or earlier. And since all drivers are at default, v-sinc is enabled during the timedemo, which was time demo1. You can disable v-sinc but then things would not be at default settings, and who runs games with v-sinc off anyway? Just easier to leave v-sinc on by default for doing general benchmarks during that generation of cards because frame rates averaged around 30, well below or monitor refresh rates.

Guess I should have stated that.

If you your getting under 20fps with a V1000 then something is off. What CPU are you running with it.

I told ya earlier that I'm using a Celeron 1200. 😀 No CPU bottlenecks here.

According to the 1998 Gold MiniGL readme, triple buffering is enabled by default. This should negate issues with vsync and framerates. However, triple buffering may be a problem with 4MB cards because an extra screen buffer consumes more video RAM. I have no idea if their older OpenGL drivers have triple buffering support or whether it's on by default if so. This will definitely affect benchmark results because double buffering w/ vsync will lock the framerate near fractions of the refresh rate.

For example, if the card can't manage 60 fps, it will be reduced to ~30 fps. If it can't do 30 fps, it will be dropped to around 20fps. If not 20 fps, it will lock to around 15 fps. And so on.

I have to try out GLQuake and vQuake again later. You should try out vHexen II and Hexen II GL and see how it goes for you.

Running a celeron 1200 with those cards your sure gonna get some wonky results. 😉 Your Voodoo card wouldn't run any faster I'm pretty sur,e but I never tried a Verite in a newer system. Yes you'd have to turn v-sync off then for sure.

Here's my V1000 collection:

Nice collection!

bushwack wrote:

Running a celeron 1200 with those cards your sure gonna get some wonky results. 😉

It's an old Abit BF6 440BX mobo running a Slot T w/ a Tualeron 1200. I'm trying to eliminate CPU bottlenecks completely. It won't cause anything wonky. Heh. Mainly I just don't want to deal with a sluggish system and I see little reason to. I need to have access to XP for a few reasons too so I have it dual booting with 98SE.

Your Voodoo card wouldn't run any faster I'm pretty sur,e but I never tried a Verite in a newer system.

A Voodoo card is undeniably faster than a V1000. It has several times the fillrate partly because V1000 can't combine very many visual features without huge performance losses. V1000 also has performance problems with z-buffering that Voodoo doesn't have. Consider that a Voodoo 1 can run Unreal more than adequately. V1000 is nowhere near that level.

I think its certain that your P200MMX is very limiting for a Voodoo1. Back in 1997, I ran a P2-266 with a Voodoo1. Eventually I had a P2-450 with it.

If you read the link to the Google Groups Usenet post in my first post (Info section), you can find out about some of V1000's quirks and limitations. And why vQuake won't run much faster on a V2200 unless you feed it a lot of CPU power.

swaaye wrote:

It's an old Abit BF6 440BX mobo running a Slot T w/ a Tualeron 1200. I'm trying to eliminate CPU bottlenecks completely. It won't cause anything wonky. Heh.

A Voodoo card is absolutely faster than a V1000. It has roughly 2x the fillrate and it doesn't have anywhere near the same issues with visual features reducing performance, or with z-buffered pixel performance.

You may be having a PCI bus mastering issue.

The original Voodoo is not CPU bound as far as I know. I used to get ~30 fps in quake in my old Pentium 75 and when I upgraded my system to a P166mmx @250 I still got the same numbers. I remember back in the day when people upgrading to PIIs was wondering why their Voodoos still ran the same.

bushwack wrote:

You may be having a PCI bus mastering issue.

Very unlikely with a 440BX.

bushwack wrote:

The original Voodoo is not CPU bound as far as I know. I used to get ~30 fps in quake in my old Pentium 75 and when I upgraded my system to a P166mmx @250 I still got the same numbers. I remember back in the day when people upgrading to PIIs was wondering why their Voodoos still ran the same.

Of course a Voodoo1 would be CPU bound with a P75. It was CPU bound in various games even with a P2-300.

Are you trying to say that you think a V1000 is a Voodoo1's equal? That's simply not the case....

swaaye wrote:

Of course a Voodoo1 would be CPU bound with a P75. It was CPU bound in various games even with a P2-300. P75!?

It's not, all I can say is try it. The Voodoo1 was an amazing card that turned any cheap Pentium computer into an arcade machine.

It would depend on the game. Try that P75+Voodoo1 in Unreal sometime. 😵 😵 😵

swaaye wrote:

It would depend on the game. Try that P75+Voodoo1 in Unreal sometime. 😀

Yeah Unreal really needs a PII.

Thing about that setup Tom was running is that the 3dfx cards were overclocked about 20% with vsync off and a couple other teaks that would make the card run faster but looked like trash when you actually went to play. I remember doing the same thing years ago.

Yeah I noticed that he overclocked them and so I just got rid of the link.

Anyway, this is a Verite thread. Go make me some screens or something!!!! 😎 😎

I'm jealous of your V1000L Screamin' 3D btw. My V1000E board seems to be dieing cuz it locks up randomly doing GUI acceleration. Even tried different systems. It works ok in 3D games, strangely enough.

So what's the difference between the V1000E and the V1000L boards?

bushwack wrote:

So what's the difference between the V1000E and the V1000L boards?

V1000L is a newer 3.3v chip design. V1000E is a 5v chip. L runs cooler and it may run at a higher clock speed.

-Added some rather nice looking screenshots of Grand Prix Legends.

And I ran some benches on the V1000, V2200 and Voodoo1. 😁

This is with my Tualeron 1200 MHz setup on the 440BX mobo. CPU should not be a bottleneck as a result. I used the Gold 1998 MiniGL for the Verite cards. vQuake was run in plain DOS 7 (Ctrl-F5 at boot) and GLQuake is within Win98SE. All default settings. The "AA7" vQuake result with with "r_antialias 7" set in the game for full AA. With vQuake, it appears that the V2200 is hitting vsync as it is enabled by default there.

1vQuake
2              V1000  V2200                  
3320x200        55.4   59.6
4320x200 AA7    48.5   52.7
5640x480        23.1   31.0
6
7GLQuake (timedemo demo2)
8              V1000  V2200  Voodoo1 (timedemo demo2)
9320x200        27.4   74.7     71.4
10640x480         6.4   24.9     25.2

Also, I ran Rendition's DMATEST with the V1000 installed and received a 51MB/s DMA result.

vHexen II looks very nice....
this thread is VERY interesting, thanks for all contributing to it!

keropi wrote:

vHexen II looks very nice....
this thread is VERY interesting, thanks for all contributing to it!

Wow, Grand Prix Legends looks even nicer! 😁 The screenies throughout this entire thread have been excellent.

swaaye wrote:

Try that P75+Voodoo1 in Unreal sometime. 😵 😵 😵

6fps with am5x86+voodoo2

interestingly, 6fps with just am5x86 in software as well
it just sucks at the floating point lighting calcs

After messing with the quake engine a bit I'm almost interested to make a retro machine 3d benchmark as hexen2 and quake have *unreasonable* texture sizes and such that aren't even the power of 2 which causes a lot of uglyness and even slowdown. It's mostly differing the results in 'may the best round-down win' 🙁

swaaye wrote:

Bushwack, I was pondering your GLQuake results more.

I am suspicious of that V1000 getting 23 fps at 640x480.

It just now occurred to me that the chart I put up says GLQuake when in fact the rendition cards are running vQuake. Oops. The Rendition GL port didn't come out till 98' did you say swaaye?. I put a note above the chart.

Also the Apocalypse 3D (PowerVR chip) and the Viper 330 (nVidia Riva) are using their own GL mini port drivers, not the same one as the regular GL port of the Voodoo cards.

bushwack wrote:

It just now occurred to me that the chart I put up says GLQuake when in fact the rendition cards are running vQuake. Oops. The Rendition GL port didn't come out till 98' did you say swaaye?. I put a note above the chart.

That would explain things. The V1000 does pretty well in vQuake. It's not hugely behind the V2200. In GLQuake however, the V1000 really dies off. The readme for the MiniGL says that the V1000 just isn't well suited for OpenGL.

I'm not sure when the first MiniGL came out. The one I used was the "Gold" release from 1998. There may have been betas in 1997.

leileilol wrote:

6fps with am5x86+voodoo2

interestingly, 6fps with just am5x86 in software as well
it just sucks at the floating point lighting calcs

I'm fairly amazed that it ran at all. 😀