I would recommend a contemporary graphics card. Later graphics cards have a high driver overhead because they were born into systems with significantly more RAM and CPU power. There are then such nice effects as that these graphics cards perform rather poorly in comparison with slow CPUs and then scale very far with CPU performance and perform quite well again with fast CPU.

The best option for your 233MMX are the Voodoos as matze79 has already written (V2, V3, Banshee). Alternatively still for PCI: G200/G450, Riva128/TNT1, Permedia 2, Voodoo1 whereby the Permedia 2 and the Voodoo1 mark the lower limit performance-wise and the others also demand their compromises and are also not cheap. For fast, compatible DOS, V3 and Banshee would be first choice. I would look for these variants in your place and accept the surcharge for it.

I tried NFS Porsche on a 233MMX + Voodoo2 system within the past year. Iirc, it could be made sorta playable but it wasn't a very good experience. The CPU + bus speeds just weren't there. As soon as anything remotely taxing occurred (like other cars coming onscreen) it was chugsville.

NFS 2:SE on the other hand - that is a different, and (on this class of machine) much more pleasant, story!

TNT2 M64
Generally easy to find and very cheap. It has plenty of power for a Pentium MMX. About as fast as the first TNT or a little faster.
The best 3D accelerator for late 486 and Pentium 1/MMX on a budget. Though the PCI variants might cost more.

A savage4 would be very acceptable for a mmx. Passable 3d for up to 98, and excellent dos compatibility.

Shreddoc wrote on 2021-06-20, 07:35:

I tried NFS Porsche on a 233MMX + Voodoo2 system within the past year. Iirc, it could be made sorta playable but it wasn't a very good experience. The CPU + bus speeds just weren't there. As soon as anything remotely taxing occurred (like other cars coming onscreen) it was chugsville.

NFS 2:SE on the other hand - that is a different, and (on this class of machine) much more pleasant, story!

You can play NFS Porsche nicely on MMX but one needs to turn audio quality to low, it affects framerates hugely. In a sense between nicely playable and unplayable.

Paadam wrote on 2021-08-15, 18:40:

You can play NFS Porsche nicely on MMX but one needs to turn audio quality to low, it affects framerates hugely. In a sense between nicely playable and unplayable.

That reminds me of when I tried playing Croc 2 on my MMX 233 with a Voodoo 1 card. Officially the GPU meets the minimum requirements, but the CPU doesn't. It says it needs at least a 266mhz Pentium II. At first no matter what graphics settings I used, the game ran really badly at a constant 10 or lower fps in the hub world. However, turning off the ambient audio made the game run significantly better. Wonder why the ambient audio brings the game to such a crawl. Now with ambient audio disabled and lowest draw distance but highest detail settings at 640x480, the game mostly runs at a solid 30fps but depending on the level will drop to around 12 fps for a few seconds in certain spots. It doesn't matter how low you turn the detail or resolution down, the performance is exactly the same. Guess it's a CPU bottleneck of some kind. Imo, it's playable but pretty jarring when those drops occur.

I’ll just echo most of us here: 3dfx Voodoo/Voodoo2.

Really the only 3D-accelerators worth it on pre-Pentium 2 systems. They’re about as fast (on systems this old), but the V2 will allow you to select 800x600 whereas the V1 can only do 640x480. Even the Voodoo 2 is already slightly wasted here. The Voodoo3 will not be worse and will not necessitate having a separate graphics card for 2D, BUT you’ll be hard pressed to get better performance due to the CPU bottleneck.

I speak from experience. Back in the day, a friend of mine had a Celeron 400 MHz with a TNT2 in it. Impressed by the performance I went out and grabbed a TNT2 M64 PCI for my Pentium MMX and found to my disappointment that it was actually a little slower than my old Voodoo had been on the exact same system.

You can use a newer video card for sure, but many graphics cards starting already with the TNT/TNT2/V3/G400 perform considerably better if you have at least a P2.

leonardo wrote on 2021-08-20, 19:09:

but the V2 will allow you to select 800x600 whereas the V1 can only do 640x480.

The Voodoo Graphics can do 800x600 without z-buffer. It is not very practical though and barely any game supported this.

Banshee is also good choice for MMX.

leonardo wrote on 2021-08-20, 19:09:

They’re about as fast (on systems this old), but the V2 will allow you to select 800x600 whereas the V1 can only do 640x480.

No... they are... not. Even a Voodoo 2 can bottleneck an overclocked Pentium MMX 233 in GLQuake at 640 x 480 (yes, at 640 x 480!)
A Voodoo 1 is slow as hell, severely bottlenecking a late Pentium MMX CPU in most games where it matters.

For V1 you always have to check if the game supports 512x384 or 640x400 because this is sometimes a lot faster then 640x480.
I mostly use V1 only to get the dosglide-support and for the very early win95-games.

Well, I had got an ELSA Gloria graphics card in a Pentium 1 system at some point.
The original Gloria for PCI bus had a Permedia 2 (?) accelerator for OpengGL (+HEIDI, Powerlib) and a ViRGE 325 for VGA (for OS and BIOS support) and DirectX/S3D..

http://www.vgamuseum.info/index.php/cpu/item/ … 0-elsa-gloria-l

It was a nice card, I think. Unfortunately, OpenGL wasn't so good supported at the time.
It was all about DirectX, rather. Anyway, good old Trio64's/Virge's 2D core was fine for DDraw, WinG etc.:

So the software-renderer of DirectX 6.x (?) had something to work with.
And with an MMX capable CPU, the software rasterizer was even happier. 😁

PS: I also gad got a FireGL card at some point.
Not sure for which PC generation, though.

Jo22 wrote on 2021-08-21, 19:09:

Unfortunately, OpenGL wasn't so good supported at the time.

That's due to OpenGL's workstation roots and only very expensive cards supported this properly at the time. The first use of OpenGL for consumer appeared with GLQuake and consumer hardware didn't support the API. This is where MiniGL implementations started to appear so GLQuake could be playable on consumer hardware with 3Dfx being the first. Eventually they became full OpenGL implementations much later on.

Here's an excerpt from the GLQuake 0.95 readme file about the state of OpenGL at the time:

Theoretically, glquake will run on any compliant OpenGL that supports the texture objects extensions, but unless it is very pow […]
Show full quote

Theoretically, glquake will run on any compliant OpenGL that supports the
texture objects extensions, but unless it is very powerfull hardware that
accelerates everything needed, the game play will not be acceptable. If it
has to go through any software emulation paths, the performance will likely
by well under one frame per second.

At this time (march '97), the only standard opengl hardware that can play
glquake reasonably is an intergraph realizm, which is a VERY expensive card.
3dlabs has been improving their performance significantly, but with the
available drivers it still isn't good enough to play. Some of the current
3dlabs drivers for glint and permedia baords can also crash NT when exiting
from a full screen run, so I don't recommend running glquake on 3dlabs
hardware.

3dfx has provided an opengl32.dll that implements everything glquake needs,
but it is not a full opengl implementation. Other opengl applications are
very unlikely to work with it, so consider it basically a "glquake driver".
See the encluded 3dfx.txt for specific instalation notes. 3dfx can only run
full screen, but you must still have your desktop set to a 16 bit color mode
for glquake to start.

Pajeroking wrote on 2021-06-16, 10:34:

Hi, I have a Compaq Deskpro 4000, Pentium MMX 233, 64 Mb ram, S3 2mb, Win 98. The PC has no AGP slot! The S3 satisfies all my Do […]
Show full quote

Hi,
I have a Compaq Deskpro 4000, Pentium MMX 233, 64 Mb ram, S3 2mb, Win 98.
The PC has no AGP slot!
The S3 satisfies all my Dos gaming and low spec Windows games (Aoe 1, KAM, NFS 1 etc.).
However, I have some games that most probably won't run very smooth.
I would like to try the following games:
- NFS Porsche;
- Turok;
- Pharaoh;
- Populous the beginning;
- Roller Coaster Tycoon 1;
- Carmageddon 2;
- Dune 2000;
- Half Life ( not mandatory).
- Grand Prix 2. ~10 fps on Svga on S3 🙁
~~~~~~~~~
The most easiest way is to buy a Voodoo 1/2. But of course that is very rare and expensive. I did some research and acceptable alternatives are : Riva 128, Matrox G200 maybe even a S3 Savage 4.
The question is not which card is the best, but which one is available and has an acceptable price. That means I won't wait for a Vodoo, whichever card appears first and has an acceptable price, i will buy it.
Will any of these cards run the listed games on an acceptable fps ( 20-30) at an acceptable resolution ( Vga, not necessary Svga)?
Is that a good plan? Or can I go with a newer card non period corect that has PCI but is cheaper/more available?

Most of those games are unbearable on an MMX 233.

Not really. The 90s wasn't as 60fps spoiled as many say, and most of the games on that list are very well optimized anyway (especially Populous 3, Dune 2000 and RCT1!).

The only terrible PMMX233 scenario I can imagine for these is if it's a board with >64MB caching issues. Maybe Porsche and Half-Life will chug, but a good Voodoo2/Banshee/3 will ease this pain a bit. The processor isn't the problem for most here. It's that implied 1996 S3 card

At the time I had a Riva128, which was good enough for Dark Forces 2, Quake 2, I think an OEM copy of Motoracer. Although I'm pretty sure by the time Blood 2 came out, I had upgraded to a RivaTNT which came out in 1998.

It's a fun card for the authentic non-voodoo 1997 experience, which a Pentium 233 MMX definitely is the end of the line for. But things were moving so fast back then, it was very quickly eclipsed. If for no other reason than it's cramped 4 MB of VRAM.

Having done some significant gaming on a P233 system recently, you basically run out of CPU juice for nearly everything midway through 1999. You begin seeing P2 creep into the recommended system requirements a lot. And some games that came out in 1998, like Unreal, are going to be sub 30 fps almost no matter what you do.

I'm actually going to be rebuilding my P233 system with a recapped motherboard soon and a PCI Riva128. So maybe I can get back to this thread with how far I can stretch it when I do.

Namrok wrote on 2021-08-21, 23:13:

Having done some significant gaming on a P233 system recently, you basically run out of CPU juice for nearly everything midway through 1999. You begin seeing P2 creep into the recommended system requirements a lot. And some games that came out in 1998, like Unreal, are going to be sub 30 fps almost no matter what you do.

That's been my approximate experience too. Peak performance is 1997 releases and earlier. Later is on a case-by-case basis; some fine, some not.

More or less the common dynamic in those times of fast CPU development, wasn't it. The 90's, early 00's. 2-3 years of cutting edge gaming at best, with a given CPU, and then you're mostly consigned to the back catalogs.

Not so different now. The time interval has just (let's say) doubled-ish. Semi-relevant excerpt, [from] :

Industry experts have not reached a consensus on exactly when Moore's law will cease to apply. Microprocessor architects report that semiconductor advancement has slowed industry-wide since around 2010, below the pace predicted by Moore's law. However, as of 2018, leading semiconductor manufacturers have developed IC fabrication processes in mass production which are claimed to keep pace with Moore's law.

Shreddoc wrote on 2021-08-22, 01:33:

That's been my approximate experience too. Peak performance is 1997 releases and earlier. Later is on a case-by-case basis; some […]
Show full quote

Namrok wrote on 2021-08-21, 23:13:

Having done some significant gaming on a P233 system recently, you basically run out of CPU juice for nearly everything midway through 1999. You begin seeing P2 creep into the recommended system requirements a lot. And some games that came out in 1998, like Unreal, are going to be sub 30 fps almost no matter what you do.

That's been my approximate experience too. Peak performance is 1997 releases and earlier. Later is on a case-by-case basis; some fine, some not.

More or less the common dynamic in those times of fast CPU development, wasn't it. The 90's, early 00's. 2-3 years of cutting edge gaming at best, with a given CPU, and then you're mostly consigned to the back catalogs.

Not so different now. The time interval has just (let's say) doubled-ish. Semi-relevant excerpt, [from] :

Industry experts have not reached a consensus on exactly when Moore's law will cease to apply. Microprocessor architects report that semiconductor advancement has slowed industry-wide since around 2010, below the pace predicted by Moore's law. However, as of 2018, leading semiconductor manufacturers have developed IC fabrication processes in mass production which are claimed to keep pace with Moore's law.

Back then rapid obsolescence was active due to many breakthroughs. A computer that was considered high-end at the time quickly become low-end about one or two years.

As for Moore's law many times people has stated during the course of time that it was dying and they were wrong. The only thing here is that it's slowing down due to the increasingly costs for each new process. Moore's law does not only applies to CPUs, it also applies to chipsets, SoC, GPU, etc.

Currently we are seeing "chiplets" and die stacking as a workaround to the cost issue of the recent manufacturing process.

Gmlb256 wrote on 2021-08-22, 02:31:

Back then rapid obsolescence was active due to many breakthroughs. A computer that was considered high-end at the time quickly b […]
Show full quote

Shreddoc wrote on 2021-08-22, 01:33:

That's been my approximate experience too. Peak performance is 1997 releases and earlier. Later is on a case-by-case basis; some […]
Show full quote

Namrok wrote on 2021-08-21, 23:13:

Having done some significant gaming on a P233 system recently, you basically run out of CPU juice for nearly everything midway through 1999. You begin seeing P2 creep into the recommended system requirements a lot. And some games that came out in 1998, like Unreal, are going to be sub 30 fps almost no matter what you do.

That's been my approximate experience too. Peak performance is 1997 releases and earlier. Later is on a case-by-case basis; some fine, some not.

More or less the common dynamic in those times of fast CPU development, wasn't it. The 90's, early 00's. 2-3 years of cutting edge gaming at best, with a given CPU, and then you're mostly consigned to the back catalogs.

Not so different now. The time interval has just (let's say) doubled-ish. Semi-relevant excerpt, [from] :

Industry experts have not reached a consensus on exactly when Moore's law will cease to apply. Microprocessor architects report that semiconductor advancement has slowed industry-wide since around 2010, below the pace predicted by Moore's law. However, as of 2018, leading semiconductor manufacturers have developed IC fabrication processes in mass production which are claimed to keep pace with Moore's law.

Back then rapid obsolescence was active due to many breakthroughs. A computer that was considered high-end at the time quickly become low-end about one or two years.

As for Moore's law many times people has stated during the course of time that it was dying and they were wrong. The only thing here is that it's slowing down due to the increasingly costs for each new process. Moore's law does not only applies to CPUs, it also applies to chipsets, SoC, GPU, etc.

Currently we are seeing "chiplets" and die stacking as a workaround to the cost issue of the recent manufacturing process.

Moore's Law is only strictly possible when the time component is allowed to be elastic. That is basically what Moore did originally: his first version of "the law" was based around a component doubling every ONE year, but later doubled it to two years when forced by practical considerations. That two years is now cited on Wikipedia, etc. However an increasing timeframe is more in keeping with his actual methods. And in that sense, his Law is indeed not broken by the recent slowdowns. On the other hand, for a person who sees the law as strict, unchanging specifics, then ohyesit's-definitely-possibly-broken... 🤣

Not that it greatly matters!, since Moore's Law was only ever a changing set of simple extrapolations at certain points in time, never intended to take on the Nostradamus-like quasi-religious status of lasting prophecy that it now has.

Jo22 wrote on 2021-08-21, 19:09:

Well, I had got an ELSA Gloria graphics card in a Pentium 1 system at some point.
The original Gloria for PCI bus had a Permedia 2 (?) accelerator for OpengGL (+HEIDI, Powerlib) and a ViRGE 325 for VGA (for OS and BIOS support) and DirectX/S3D..

http://www.vgamuseum.info/index.php/cpu/item/ … 0-elsa-gloria-l

That is of course not a Permedia 2, she wouldn't have any need for the extra S3 chip.