For me a voodoo 2000 or banshee seems like the best idea, maybe a savage4? Rage 128 and tnt seem like a good pair with ss7.

I'm not one for jumping ahead 5+ years for a GPU, fx5200 aside. But if you have some, I'd like to see some NV6 and NV7's and ati hd2/3 cards. in such a system, But not many 3.3v supported cards that new. or a ss7 board with 1.5v support/

After watching your benchmarks, I found a GF3 Ti200 for a little over $5. Goal is to be CPU-limited for max graphics.

For my K6-2+ 600MHz system I chose a Quadro 2 Pro combined with Voodoo2 SLI.
While the Quadro2 Pro is basically always CPU limited it has the advantage of DVI output.
Benchmarks showed that going from a single Voodoo2 to SLI does not increase fps noteworthy, it just allows to use 1024 resolution.
So a Voodoo3 is most likely also always CPU limited.

elianda wrote:

For my K6-2+ 600MHz system I chose a Quadro 2 Pro combined with Voodoo2 SLI. While the Quadro2 Pro is basically always CPU limit […]
Show full quote

For my K6-2+ 600MHz system I chose a Quadro 2 Pro combined with Voodoo2 SLI.
While the Quadro2 Pro is basically always CPU limited it has the advantage of DVI output.
Benchmarks showed that going from a single Voodoo2 to SLI does not increase fps noteworthy, it just allows to use 1024 resolution.
So a Voodoo3 is most likely also always CPU limited.

True, but it seems that faster, more powerful architectures squeeze more limitation out of these CPUs. Although I do like my Voodoo3, I kind of wish it were a PCI version. I only want Win98 for certain old Glide games anyway, otherwise the Voodoo3 has awesome 2d quality.

Updated the first post with 3dfx results!

PhilsComputerLab wrote:

Updated the first post with 3dfx results!

Nicely done!

Overall I think you drew the right conclusion.
Read your update and watching the video right now 😁

Only note is the lack of a Voodoo 2 (doesn't have to be SLI afaic) and I'm wondering how Voodoo 2 SLI + TNT2 would fare? The only downside I could see is the lack of good image quality because of the lack of Voodoo 3, though I never really saw the image quality of V2 as a problem, personally I think it has its own charm 😀

It does seem cumbersome though to create a mix of the 2, I'd probably just build 2 rigs, one with a V3 and one with TNT2, but people who have less space will possibly prefer the single system solution.

edit: Nvm, you're right about V2 being similar to cards you already tested 😊

Woohoo! Go Voodoo 3!

That was the card in my Pentium III machine when I was a kid. I never even had to think about video performance as it always worked perfectly for everything, which led me into the false conclusion for years that only the CPU mattered 🤣

Yea V2 SLI is pretty much exactly like the V3 somewhere between 2000 and 3000.

keenmaster486, that's exactly why they are so easy to recommend. They just work, compatibility is excellent and you can just enjoy the games.

Hello Vogons, I've been lurking for a while, but this is my first official post! I enjoy Phil's YouTube channel, and decided to join the discussion.

Endorsing the Voodoo 3, over the Voodoo 4, makes sense because this SS7 era of gaming lacked the raw power to play games with 32 bit color. At least they lacked the raw power to play them at an ideal frame rate. Perhaps Phil's coverage of ATi will prove otherwise!

I'm curious though, is there a list that shows which games took advantage of 32-bit color back then? I wonder what such a system would be missing out on? I know Phil covers a couple of them in his video, but how big of a deal was 32 bit back then? I remember playing a lot of games in 16-bit, simply because I couldn't tell the difference. Sadly, this is no longer the case. Sometimes I think ignorance really is bliss!

well........32-bit color isn't really cpu discriminative, and isn't more expensive on the old PCI or AGP bus. It's okay to enjoy 32-bit rendering on SS7 with a more powerful/capable card. Some are faster in 32-bit color than 16-bit (PowerVR particularly)

The whole deal with 32-bit then was "die grainy dithering die!!!!". 3dfx users usually doesn't notice a huge difference since their 16-bit dither is filtered to hell and back with blur to show for it. The lack of a green tinge to the colors helps too.

Usually it's PSX/N64 ports and Glide-centric games that don't take advantage of 32-bit color since they'd often have paletted textures or 16-bit textures at best, though 32-bit can be useful at swizzling paletted data to true color for no image quality loss on newer palette-less cards.

Nice thread!

Phil, some of your numbers seem somewhat low on GLQuake and Quake 2 for example. On the other hand, Incoming is faster than my results 😖!
Perhaps you'd like to check and compare, I created a thread with detailed configurations a while back. Hope it helps!
Pentium II/III VS K6-III+&

SS7 seems to be very sensitive to the parts and software you use. I had someone else on YT share his results and they were lower than mine.

What stands out is the use of MiniGL. In Quake 2, is that the option 3dfx OpenGL? Or by replacing files manually?

I used the default OpenGL option, same with GLQuake. I think all the cards I used support full OpenGL ICD.

The VIA results used AmigaMerlin, I think these are the best drivers available? Could also be a factor, not sure.

So yea, good to know that there is some more performance to be had 😀 Especially if these are with sound, that's very impressive.

My approach to benching is more conservative. No tweaking, BIOS defaults, default drivers and off you go. That leaves people with some headroom for tweaking and beating the results 🤣

But maybe I could do a video "Tweaking SS7". So yea, please let me know what MiniGL you used, seems to work really well 😁 3DNow! is also impressive...

EDIT: Bloody lot of work you put into that project. So many benchmarks. I stripped it down to just a few benchmarks because otherwise it just becomes too hard...

leileilol wrote:

well........32-bit color isn't really cpu discriminative, and isn't more expensive on the old PCI or AGP bus. It's okay to enjoy 32-bit rendering on SS7 with a more powerful/capable card. Some are faster in 32-bit color than 16-bit (PowerVR particularly)

The whole deal with 32-bit then was "die grainy dithering die!!!!". 3dfx users usually doesn't notice a huge difference since their 16-bit dither is filtered to hell and back with blur to show for it. The lack of a green tinge to the colors helps too.

Usually it's PSX/N64 ports and Glide-centric games that don't take advantage of 32-bit color since they'd often have paletted textures or 16-bit textures at best, though 32-bit can be useful at swizzling paletted data to true color for no image quality loss on newer palette-less cards.

I didn't know that PowerVR was faster with 32-bit, how interesting! I had long forgotten about PowerVR as a PC solution. While I remember the Kyro cards, I'll always remember PowerVR for their contribution to the console market.

I like the idea of finding the right graphics card to match the SS7. I think a more powerful graphics card would do well with 32-bit colors. However, I'm concerned that when it get's too powerful, we start to become CPU limited. It's a brute force approach. I like the more nuanced approach of trying to find a GPU that starts to top out, around the same time as the CPU. In that scenario, we might find that 32-bit color gaming on the SS7, is not ideal, and that a more powerful platform required.

Hu1kamania wrote:

I like the more nuanced approach of trying to find a GPU that starts to top out, around the same time as the CPU. In that scenario, we might find that 32-bit color gaming on the SS7, is not ideal, and that a more powerful platform required.

That's exactly what this project is about 😀

I'll try to add the charts to this thread, not everyone wants to watch the video. Some charts will be a bit packed though with too many results...

PhilsComputerLab wrote:

I'll try to add the charts to this thread, not everyone wants to watch the video. Some charts will be a bit packed though with too many results...

I don't think there's such a thing as too many results around these parts. 😉

clueless1 wrote:

PhilsComputerLab wrote:

I'll try to add the charts to this thread, not everyone wants to watch the video. Some charts will be a bit packed though with too many results...

I don't think there's such a thing as too many results around these parts. 😉

We shall see 🤣

I can add a results table at the bottom, should make it easier...

Thank you for your kind words Phil!

Yes, all of the tests include sound. Someone suggested I disable it, but I felt like that was a little silly. I wanted to have some benchmarks so that people that are thinking about building a Super Socket 7 system get a good idea of performance and keep a reference at their side so they can tweak their systems.

I used MiniGL 1.47, which according to falconfly.de is the best version to use with the K6-2 (apparently they started doing some Athlon optimizations later on with 1.48 and 1.49). I had to replace the existing miniGL in Quake 2's folder of course and then I ran 3Dnow! 3DfxGL and 3DfxGL for the non-3Dnow! tests. The 3DNow! OpenGL and Default OpenGL options produced the same results, they both used the MiniGL since I put it in the game's folder.

OpenGL ICD is a lot slower than the MiniGL, unfortunately. Also, are your tests using 3DNow! ? The AmigaMerlin drivers may be holding the cards back, since they were created for much faster CPUs, so they may have tweaked certain settings. Swaaye also brought the ALi AGP Utility to my attention, which was a very handy tool in figuring out why I was getting shit performance sometimes. It usually was all about Write Combining, I had to fool around with that a lot, until I had a setting that worked optimally. That tweaking Super Socket 7 video sounds awesome!

Well it seems you already tweaked SS7 to the max 😀

Yea, none of my tests use any 3DNow! patches, MiniGL or anything performance boosting.

Crikey that MiniGL is fast! With the AmigaMerlin drivers in GLQuake Demo 1 640x480x16 the score improved from around 117 to 176!

And Quake II from around 66 to 82! The 3Dnow! 3DfxGL however gave me a grey screen.

Now with the sound, the Quake games are not very demanding, Incoming however is, especially if you run it default, it will cost quite some CPU power. That is going to be the reason of the discrepancy.

EDIT: Half Life comes with 1.45 MiniGL, I wonder how much of a boost 1.47 gives, because I think that game is really heavy on the CPU.

clueless1 wrote:

PhilsComputerLab wrote:

I'll try to add the charts to this thread, not everyone wants to watch the video. Some charts will be a bit packed though with too many results...

I don't think there's such a thing as too many results around these parts. 😉

I'd say there aren't too many parts for all the results we want 😜

Done!

First post updated with some graphs 😀