VOGONS


First post, by Frunzl

User metadata
Rank Newbie
Rank
Newbie

Hi,

ever since I learned about the 16x AA and AF capabilities of the 5x, 6x and 7x Geforce series' Quadro cards, I've been wanting to build a PCIe system with the best possible image quality for (later) Win98 games.

I figure, I am not alone and I want to share a recent discovery.

I benchmarked a number of 7th gen cards on my X58 system (with a Xeon X5687 @ 4.75 GHz), namely the 7600 GT, 7900 GT, 7900 GTX, 7950 GX2, (two) Quadro FX 4500 and a Quadro FX 5500. None of them ran 3DMark2000 (freezes) and I attributed it to the sketchy, modified 82.69 drivers, you have to use with this generation under Win98 (seems to be "common knowledge" in many forum posts that 3DM2K simply doesnt work with these). There were some additional incompatibilities, for example, the Quadro FX 5500 refuses to start Unreal at all. No Idea, why.

Anyway, here's my discovery: some 7th gen cards can run 3DMark2000 after all!

I just got both my Quadro FX 4500 to complete it flawlessly by flashing older VBIOS versions. The cards are two different revisions. The VBIOS versions, which do manage to run 3DM2000 are listed below:

- Rev. A01 with double-sided VRAM and cooler on the back: 05.70.02.15.02

- Rev. A00 with single sided VRAM: 05.70.02.41.01

Both can be downloaded from techpowerup.

Maybe this can be of use to someone dealing with compatibility issues of the 7th gen cards under Win98.

If anyone is interested, I will post the benchmark results for all the different PCIe cards (5th, 6th, 7th gen), including max AA and AF. I am also in the process of trying different VBIOS versions for the Quadro FX 5500.

Reply 1 of 5, by DoZator

User metadata
Rank Member
Rank
Member

Interesting information. Thank you.

Although, usually, as far as I know, everything works fine, at least with the latest normal version of the ForceWare 77.72 driver, which is actually fully compatible with video adapters based on the G70 GPU (Except that for 512MB versions of cards, an additional corrected NVCORE.VxD is required), by adding data about the card to the installation INF. In one of the topics earlier, I gave the results of testing in some versions of 3DMARK, it seems, here:
Re: PCIe devices on Windows 98 SE

Maybe I was lucky with vBIOS, although I already used several different cards (Based on the G70 and G71) and had no similar problems. Now I have a Quadro FX 4500 and I should probably double-check this phenomenon again...

Reply 2 of 5, by Frunzl

User metadata
Rank Newbie
Rank
Newbie
DoZator wrote on 2023-09-30, 17:29:
Interesting information. Thank you. […]
Show full quote

Interesting information. Thank you.

Although, usually, as far as I know, everything works fine, at least with the latest normal version of the ForceWare 77.72 driver, which is actually fully compatible with video adapters based on the G70 GPU (Except that for 512MB versions of cards, an additional corrected NVCORE.VxD is required), by adding data about the card to the installation INF. In one of the topics earlier, I gave the results of testing in some versions of 3DMARK, it seems, here:
Re: PCIe devices on Windows 98 SE

Maybe I was lucky with vBIOS, although I already used several different cards (Based on the G70 and G71) and had no similar problems. Now I have a Quadro FX 4500 and I should probably double-check this phenomenon again...

Thank you for sharing this! I was completely unaware that version 77.72 (which is also my go to version for the earlier cards btw) runs on the Quadro 4500. I modified the inf and it works indeed! I have yet to flash the previously non-working BIOSes and try if 3DMark works with those as well. Would be interesting to know your BIOS version.

However, you write that you successfully used it with G71 cards? I tried it with most of my cards, and it worked exclusively on the 4500, but not the Quadro FX 5500 or Geforce 7900s or the 7600 GT (G73). Which one did you test?

Reply 3 of 5, by DoZator

User metadata
Rank Member
Rank
Member

G70 only. Everything that is newer requires an 8X version of the ForceWare driver (Relatively stable and most productive only in OpenGL). For some particularly problematic cases, it was also necessary to roll back and nvOpenGL.DLL up to an earlier version. In some cases, this also applies to the 77.72 driver, when rolling back this file to version 71.84 helps for stable operation of some OpenGL applications.And for 8X versions of the driver, earlier versions of this file are also suitable, fixing some problems, but only from the corresponding 8X versions (There were several of them). However, by rolling back nvOpenGL.DLL, the OpenGL version is also being lowered (Some programs may need, for example, version 2.0 or higher). So, for example, the original nvOpenGL.DLL as of version 77.72, it provides support for OpenGL 2.0, whereas nvOpenGL.DLL from version 71.84 only OpenGL 1.5.3. Version 82.69 supports OpenGL 2.0.1

If you look more specifically at the support of OpenGL extensions by different drivers, the picture looks something like this (According to AIDA64):

Detonator 40.52 for Win9x - OpenGL 1.4.0; extensions - 41; OpenGL 1.4 - 100%; OpenGL 3.0 - 0%
Detonator 45.23 for Win9x - OpenGL 1.4.0; extensions - 42; OpenGL 1.5 - 33%; OpenGL 3.0 - 0%
ForceWare 56.56 for Win9x - OpenGL 1.5.0; extensions - 110; OpenGL 2.0 - 30%; OpenGL 3.0 - 4%
ForceWare 66.42 for Win9x - OpenGL 1.5.2; extensions - 125; OpenGL 2.0 - 80%; OpenGL 3.0 - 4%
ForceWare 77.72 for Win9x - OpenGL 2.0.0; extensions - 129; OpenGL 2.0 - 100%; OpenGL 3.0 - 23%
ForceWare 82.69 for Win9x - OpenGL 2.0.1; extensions - 133; OpenGL 2.1 - 66%; OpenGL 3.0 - 28%

In practice, this is reflected in the performance of OpenGL games, as well as support for OpenGL applications. These points should also be taken into account when solving problems with the driver. However, versions above 77.72, although better in OpenGL, but D3D \ DirectDraw in them is almost completely broken and even where it still works somehow, the performance is lower than the baseboard (And many games generally crash with an error). That's why I ditched the G71 in favor of the G70, because it's still possible to put the relatively hassle-free ForceWare 77.72 on it.

Reply 4 of 5, by Frunzl

User metadata
Rank Newbie
Rank
Newbie

The poor D3D compatibility of the later drivers was precisely why I was hesitant to use a 7x card. On the other hand, the 6x cards can struggle performance-wise with 16xAA at higher resolutions, even on a fast system, so I was really stuck.

This solves a big problem for me, thanks! 😀 Also, the FX 4500 is clearly the best card for this.

Since you are way more knowledgable than me regarding these drivers, there is another thing I noticed, while benchmarking. All the Quadro cards (not only the FX 4500, but also the FX 3450 and the FX 1300, tested with different driver versions) seem to be unable to perform 16xAA/AF at 1600x1200 resolution and in reality only do 8x. This seems to be limited to newer games. I noticed it with 3DM2001 and Serious Sam (both D3D and OpenGL), but not 99max. Have you ever seen this as well? Can you think of a possible cause?

Reply 5 of 5, by DoZator

User metadata
Rank Member
Rank
Member

Most 9x apps and games were developed primarily for low resolutions (the most common at that time). Large ones can be delivered, but as a rule, performance suffers from this. Even 1024x768 or 1280x720 is already a small draw on FPS. Also, old 9x games, which it is advisable to run under Win9x, have rather simple textures and graphics, which, at high resolutions, look less natural, because all the graphic features of that time, inconsistencies of seams, for example, become more pronounced. And no, I didn't try 1600x1200, because my LCD supported 1920x1080 as much as possible initially, but later I flashed one of its inputs to 1600x900 to get rid of scaling problems. Which I'm currently using. And on the CRT, I also set this resolution in 100hz mode. Larger 100hz was no longer pulled, with the exception, perhaps, of the same 1920x1080. However, on a 21-inch CRT monitor, at this resolution, everything looks quite small. That's why I've never checked anything like this before.

You can try setting anti-aliasing in RivaTuner. And also rename the name of the EXE program\game (Perhaps the driver detects the application by name and applies the appropriate "optimization").

Here is an example of similar optimizations:

9x_7772_OGL.1696414103.png

But they can also be protected deeper, at the VxD \ DLL level of the driver itself. Moreover, changing these values in the registry does not lead to anything, because the driver restores these values if they are changed. Therefore, only renaming the executable file can be of any use.