I find these bottlenecking comments funny. Don't want the mighty FX to be bottlenecked by a measly Athlon64? Then try playing at a nice resolution like 1600x1200, or at least 1280x1024, and throw in some eye candy like AF and/or AA. And then you'll realize that the Athlon64 is perfectly capable of driving a Radeon x800xt, nevermind the mighty FX.

But mah 640x480...

The Serpent Rider wrote on 2022-06-18, 17:16:

But mah 640x480...

Not sure if this comment was directed at me, since I never claimed that an Athlon64 would be a bottleneck for a GeForce4 Ti4200. I merely pointed out that such a system could draw out more performance from that card compared to something slower like an AthlonXP 1700+ or a Tualatin 1.4 GHz.

If anyone's interested, here's a comparison. Same graphics card, same drivers, same OS.

So apparently the 500 mhz makes some difference with the gf4 but I can't explain those results:
X2 1000 noaa noaf: 32,36
X2 1000 4aa 8af: 11,73
X2 1600 noaa noaf: 36,03
X2 1500 4aa 8af: 26,44

With 3dmark 2k:
1000 mhz: 12379
1500 mhz: 15072

How is this possible? I'm thinking something is wrong with X2, the difference is too wide...

Joseph_Joestar wrote on 2022-06-18, 17:36:

Not sure if this comment was directed at me, since I never claimed that an Athlon64 would be a bottleneck for a GeForce4 Ti4200. I merely pointed out that such a system could draw out more performance from that card compared to something slower like an AthlonXP 1700+ or a Tualatin 1.4 GHz.

If anyone's interested, here's a comparison. Same graphics card, same drivers, same OS.

I'm pretty sure it wasn't directed at you. 😀
Very nice comparison! And, oh, oh, your Athlon 64 seems to be a major bottleneck at those hundreds of FPS 😁

Jokes aside, people don't realize, but there will always be a bottleneck. You can try as best as possible to mix and match parts, but in the end one of them will constitute the bottleneck.

Personally, I prefer to be CPU bound, not GPU bound. As long as the CPU can sustain over 75 FPS (my monitor's refresh rate), so that I get consistent frametimes/a smooth experience in the games I'm playing (with vsync enabled), it's perfectly fine for me.
But when it comes to video cards, it can't hurt to have more power (without going too overboard).

Nemo1985 wrote on 2022-06-18, 17:51:

So apparently the 500 mhz makes some difference with the gf4 but I can't explain those results:
[...]
How is this possible? I'm thinking something is wrong with X2, the difference is too high...

Are you on the same nVIDIA driver as before, or have you downgraded in the meantime? Make sure that you're not doing multiple changes at the same time. 😀 Especially looking at the 4aa 8af results, those can't possibly be related to CPU frequency changes (the CPU already demonstrated that it can sustain 32 FPS even at 1 GHz).
Anyway, have no experience with that game, but judging by its release date... it's deep into WinXP territory, so probably best to play it on a more powerful WinXP machine.

You are definitely right about X2... I'm installing the 45.23 right now... I tested a "more period correct game", finally the results are right, MDK2, at 1000 mhz without aa and af I get 195 fps, while with aa and af 74,07. If I ramp up the frequency to 1500 mhz:
255 fps and 74,27.
Now I clean the system and try with the 45.23

Nemo1985 wrote on 2022-06-18, 18:08:

You are definitely right about X2... I'm installing the 45.23 right now... I tested a "more period correct game", finally the results are right, MDK2, at 1000 mhz without aa and af I get 195 fps, while with aa and af 74,07. If I ramp up the frequency to 1500 mhz:
255 fps and 74,27.
Now I clean the system and try with the 45.23

Yeah, the MDK2 results are pretty good (can probably be better with older drivers).
As for X2, that one seems to be heavily GPU bottlenecked (and possibly also suffering from poor Win98 optimization).

Also, since you seem to enjoy enabling eye candy features like AA / AF, I highly recommend a GeForce FX 5900(XT) or an ATI Radeon 9800 PRO. The GeForce 4 Ti's performance quickly goes downhill once you enable AA/AF.

bloodem wrote on 2022-06-18, 18:24:

Yeah, the MDK2 results are pretty good (can probably be better with older drivers). As for X2, that one seems to be heavily GPU […]
Show full quote

Nemo1985 wrote on 2022-06-18, 18:08:

You are definitely right about X2... I'm installing the 45.23 right now... I tested a "more period correct game", finally the results are right, MDK2, at 1000 mhz without aa and af I get 195 fps, while with aa and af 74,07. If I ramp up the frequency to 1500 mhz:
255 fps and 74,27.
Now I clean the system and try with the 45.23

Yeah, the MDK2 results are pretty good (can probably be better with older drivers).
As for X2, that one seems to be heavily GPU bottlenecked (and possibly also suffering from poor Win98 optimization).

Also, since you seem to enjoy enabling eye candy features like AA / AF, I highly recommend a GeForce FX 5900(XT) or an ATI Radeon 9800 PRO.

I'm using such aa and af settings for testing, back in time when I had the gf4 I loved the quincunx aa. Also I think that the spirit of this machine is to run older games at 32bps and some filters since 300 fps on mdk2 are nice but pretty useless. Now I test the driver you adviced, then I will switch to the Fx5900xt. Do you have any other game to suggest? Maybe something older to check the compatibility? I was also thinking to test the dos performance.

Nemo1985 wrote on 2022-06-18, 18:38:

I'm using such aa and af settings for testing, back in time when I had the gf4 I loved the quincunx aa.

Oh, OK.

Nemo1985 wrote on 2022-06-18, 18:38:

Do you have any other game to suggest? Maybe something older to check the compatibility? I was also thinking to test the dos performance.

Well, my go-to "quick Win98 performance benchmarks" are... the usual suspects:
- GLQuake
- Quake 2
- Quake 3
- Unreal
- Unreal Tournament
- MDK2
- Expendable

As for compatibility... it depends what games you like. I usually test quite a few. Some of them are:
- GLQuake
- Quake 2
- Quake 3
- MDK2
- Unreal
- Unreal Tournament
- Expendable
- Need for Speed 3 Hot Pursuit
- Need for Speed High Stakes
- Need for Speed Porsche
- Need for Speed Hot Pursuit 2
- Nocturne
- GTA 3
- GTA Vice City
- Final Fantasy 7
- Prince of Persia 3D
- Return to Castle Wolfenstein
- Re-Volt
- Tomb Raider 3
- Tomb Raider The Last Revelation
- Tomb Raide Chronicles

As for compatibility... it depends what games you like. I usually test quite a few. Some of them are:

Don't forget about Midtown Madness, which is infamous "Achilles' heel" of Nvidia cards.

The Serpent Rider wrote on 2022-06-18, 18:49:

Don't forget about Midtown Madness, which is infamous "Achilles' heel" of Nvidia cards.

Interesting, was not aware that it was known to have issues with nVIDIA cards.
Then again, I've last played it (for about 15 minutes) 22 years ago... and let's just say I wasn't a fan, so I don't even own it (blasphemy, I know) 😀

So i'm ready to share some results.
Mainly with dos performance and compatibility, those are obtained booting directly on ms-dos, so no kind of lfb and wc is used. I tried to use MTRRLFBE but I had some stability issues mainly with fx card at 640x480, unreadable text and stuff like that.
PC Player crashed complaining to be out of memory for the textures. Fastdoom crashes with a black screen.
Using the dos prompt I had no issues except for PC Player 640 which gave the same error, while at 320 it worked but with some visual glitches.
Results for doom are the ticks, so lower values are better

Before I switch video card in windows. I'd like to know if someone likes to have some game tested, I used so far those games without and with AA\AF: X2, Mdk2, Unreal gold, other the test suite 3dmark00 and 01, initally I wanted to use the 99 too but as always with nvidia despite I have disabled the vsync it is on with that benchmark, the only way I found to disable it is using powerstrip, but I would prefer to do not install it this time (I disabled the vsync both in drivers and with rivatuner). I used the 45.23 and 56.64.

I'd like to know if someone likes to have some game tested

AquaMark
Unreal Tournament 2003 FlyBy (Antalus+Asbestos)
Dungeon Siege
Return to Castle Wolfenstein (checkpoint.dm_67)

The Serpent Rider wrote on 2022-06-19, 10:10:

AquaMark Unreal Tournament 2003 FlyBy (Antalus+Asbestos) Dungeon Siege Return to Castle Wolfenstein (checkpoint.dm_67) […]
Show full quote

I'd like to know if someone likes to have some game tested

AquaMark
Unreal Tournament 2003 FlyBy (Antalus+Asbestos)
Dungeon Siege
Return to Castle Wolfenstein (checkpoint.dm_67)

I do not have ut2003 and rtcw but I tested aquamark, I also have Dungeon Siege but after the benchmark is over, I can't find the results, the program justs shut down and apparently there isn't the fps.log file in the directory.

Edit: ok found it!

Nemo1985 wrote on 2022-06-18, 18:38:

back in time when I had the gf4 I loved the quincunx aa

Quincunx AA is great in terms of performance, but it blurs the image a bit too much for my tastes.

It also blurs the UI elements (like the HUD in Quake 2) when forced through the drivers.

Joseph_Joestar wrote on 2022-06-19, 14:54:

Nemo1985 wrote on 2022-06-18, 18:38:

back in time when I had the gf4 I loved the quincunx aa

Quincunx AA is great in terms of performance, but it blurs the image a bit too much for my tastes.

It also blurs the UI elements (like the HUD in Quake 2) when forced through the drivers.

I remember that that was the main quirk, but back in time the suggestion was to increase the AF to relieve the issue, was it wrong?

Nemo1985 wrote on 2022-06-19, 15:46:

I remember that that was the main quirk, but back in time the suggestion was to increase the AF to relieve the issue, was it wrong?

Cranking up AF definitively doesn't remove the blur on UI elements caused by quincunx AA, like the aforementioned Quake 2 HUD.

I don't think it de-blurs the 3D objects either.

Quincunx is hybrid method, which also works on transparent surfaces, unlike classical multisampling AA, but at cost of picture sharpness. Generally, it's not big issue on high resolution.

So the last afternoon I'm done testing the Gf4 both with 45.23 and 56.64 the latest do a real mess with aa and af with the gf4, despite setting them on the control panel, I had hard time to get them working, I had to create profiles for the games and despite of that I had to restart windows to make unreal stoped using aa\af despite being disabled.
That being said, when I switched to the 5900xt I started to have problems. I had to manually install the 45.23 claiming it was a 5900 since the xt is not supported. After the reboot I had a black screen, then after a reset windows worked but only at 16 colors.
I started from from a fresh copy of windows, no joy. I also tried the 53.04 and the latest nvidia drivers for windows 98 for series 5 the 81.98.
The behaviour is always the same, before installing the video drivers everything seems to work just fine (except that during the windows installation the phase of detecting hardware was very slow), after that windows is unable to boot and when it does (at 16 colors) it's slow as hell or hang. If I moved the mouse I could hear the speaker make the same noise of when you press a key too much or too many keys.
Once windows also throwed a general protection error.
I also tried to disable every agp optimizations in the bios, use the agp 4x instead of 8x, but I had the same issues.

Any suggestion is welcome.

bloodem wrote on 2022-06-18, 15:02:

Yeah, that was a strange issue, it might be related to a specific BIOS version that Phil is using, or maybe the "non-Deluxe" A8V board has some other quirks.
I've never had any driver issues with my Asus A8V Deluxe build on which I have an Athlon 64 4000+ and an nVIDIA GeForce FX 5900XT (with a forced installation of driver version 45.23).

bloodem, do you have the chance to test your board with a fx 5900?

How does 45.23 compare to "cheat" drivers from nvidia LINK ?
(I know futuremark confirmed 44.03 and 43.51 versions as bogus)

PS. 4400/4600 Ti wins vs. FX 5700, when second TMU is needed (GF4 Ti are dual TMU per pipe, FX 5700 is single TMU per pipe design).