I was finally able to run the GeForce FX 5900XT benchmarks as promised. Sorry for the delay, but... life happened and I had to return to my hometown.
I initially started doing the benchmarks on a Pentium Dual Core E5800 build, but I now switched to an Athlon 64 build that I have in this other location.
Unfortunately, no time to create graphs, so... yeah, death by text coming right up! 😀

This Winfast A350XT board that I have here is a truly remarkable overclocker: I was able to push it to GeForce FX 5950 Ultra speeds and it can go even faster than that!
So, further down, you will also see benchmarks with the overclocked card. Please note that the GeForce FX 5950 Ultra has 256 MB of RAM, and that extra memory might actually matter for the 4x AA and 8x AA tests, so in these extreme circumstances the results might not reflect the performance of a true GeForce FX5950 Ultra.

Some notes:
- As you can see, in GLQuake the card is bottlenecking the CPU even at 640 x 480, so I've decided to also test MUCH lower resolutions just for fun (at least in some games) . 😀
- For the last section of the GPU OC tests, I've only included the highest resolution/color depth + AA/AF tests (since lower resolutions and color depths were starting to get very boring - nothing spectacular to see).
- As I mentioned in a previous thread, the "utbench" Unreal Tournament benchmark is very flawed and does not reflect the actual game performance (which is much better than that), but I included it anyway.
- if you'd like me to specifically test something else, let me know and I'll do my best (I'll be near this PC until Saturday)

1#######################################################################################################################
2#######################################################################################################################
3MB: Asus A8V Deluxe
4CPU: Athlon 64 "Venice" 3800+ 2.4 GHz OC @ 2.7 GHz (FSB225 x 12)
5RAM: 1 GB DDR Single Channel
6VGA: Leadtek Winfast A350XT GeForce FX 5900XT 128 MB (default clocks: 390 MHz Core / 700 MHz memory)
7SOUND: Yamaha YMF724F-V
8DRIVERS:
9   - VIA HyperionPro V5.12A (only chipset & AGP drivers)
10   - nVIDIA Detonator 45.23 (forced installation)
11   - Yamaha 7x4 VXD sound drivers v4.07.1040
12DirectX 8.1
13
14NOTE: All tests performed with sound, unless otherwise specified
15#######################################################################################################################
16#######################################################################################################################
17
183DMark99: 23717 3D marks / 57495 CPU marks
193DMark2000: 17558 3D marks / 958 CPU marks
203DMark2001SE: 15820
21
22GLQuake  320 x  240 x 16 (window): 2021.2 FPS (demo1 / Default settings)
23GLQuake  400 x  300 x 16 (window): 1644.6 FPS (demo1 / Default settings)
24GLQuake  512 x  384 x 16 (window): 1116.8 FPS (demo1 / Default settings)
25GLQuake  640 x  480 x 16:  884.6 FPS (demo1 / Default settings)
26GLQuake 1024 x  768 x 16:  447.1 FPS (demo1 / Default settings)
27GLQuake 1024 x  768 x 32:  436.8 FPS (demo1 / Default settings)
28GLQuake 1280 x 1024 x 16:  291.3 FPS (demo1 / Default settings)
29GLQuake 1280 x 1024 x 32:  284.9 FPS (demo1 / Default settings)
30GLQuake 1280 x 1024 x 32:  105.8 FPS (demo1 / Default settings / 4x AA + 8x AF)
31GLQuake 1280 x 1024 x 32:   51.2 FPS (demo1 / Default settings / 8x AA + 8x AF)
32
33Quake 2  320 x 240 x 16: 763.8 FPS (demo1.dm2 / Default settings)
34Quake 2  400 x 300 x 16: 760.5 FPS (demo1.dm2 / Default settings)
35Quake 2  512 x 384 x 16: 747.3 FPS (demo1.dm2 / Default settings)
36Quake 2  640 x 480 x 16: 749.9 FPS (demo1.dm2 / Default settings)
37Quake 2 1024 x 768 x 16: 479.5 FPS (demo1.dm2 / Default settings)
38Quake 2 1024 x 768 x 32: 448.0 FPS (demo1.dm2 / Default settings)
39Quake 2 1280 x 960 x 16: 334.1 FPS (demo1.dm2 / Default settings)
40Quake 2 1280 x 960 x 32: 311.3 FPS (demo1.dm2 / Default settings)
41Quake 2 1280 x 960 x 32: 138.5 FPS (demo1.dm2 / Default settings / 4x AA + 8x AF)
42Quake 2 1280 x 960 x 32:  49.3 FPS (demo1.dm2 / Default settings / 8x AA + 8x AF)
43
44
45Quake 3  640 x  480 x 16: 331.9 FPS (demo four / Default settings)
46Quake 3 1024 x  768 x 16: 326.4 FPS (demo four / Default settings)
47Quake 3 1024 x  768 x 32: 322.8 FPS (demo four / Default settings)
48Quake 3 1280 x 1024 x 16: 301.9 FPS (demo four / Default settings)
49Quake 3 1280 x 1024 x 32: 292.4 FPS (demo four / Default settings)
50Quake 3 1280 x 1024 x 32: 255.1 FPS (demo four / Full details + Trilinear filtering)
51Quake 3 1280 x 1024 x 32: 120.9 FPS (demo four / Full details + Trilinear filtering + 4x AA + 8x AF)
52Quake 3 1280 x 1024 x 32:  41.6 FPS (demo four / Full details + Trilinear filtering + 8x AA + 8x AF)
53
54
55MDK2  640 x  480 x 16: 422.89 FPS (Default settings / EAX sound)
56MDK2 1024 x  768 x 16: 416.46 FPS (Default settings / EAX sound)
57MDK2 1024 x  768 x 32: 402.93 FPS (Default settings / EAX sound)
58MDK2 1280 x 1024 x 16: 340.46 FPS (Default settings / EAX sound)
59MDK2 1280 x 1024 x 32: 304.79 FPS (Default settings / EAX sound)
60MDK2 1280 x 1024 x 32: 288.27 FPS (Full details + Trilinear filtering / EAX sound)

61MDK2 1280 x 1024 x 32: 111.80 FPS (Full details + Trilinear filtering + 4x AA + 8x AF / EAX sound)
62MDK2 1280 x 1024 x 32:  39.98 FPS (Full details + Trilinear filtering + 8x AA + 8x AF / EAX sound)
63
64
65Expendable  640 x  480 x 16: AVG 224.34 FPS / HIGH 309 FPS / LOW 157 FPS (NO SOUND)
66Expendable  640 x  480 x 16: AVG 211.92 FPS / HIGH 280 FPS / LOW 148 FPS (EAX SOUND)
67Expendable 1024 x  768 x 16: AVG 209.01 FPS / HIGH 281 FPS / LOW 148 FPS (EAX SOUND)
68Expendable 1024 x  768 x 32: AVG 209.15 FPS / HIGH 283 FPS / LOW 150 FPS (EAX SOUND)
69Expendable 1280 x 1024 x 16: AVG 201.27 FPS / HIGH 282 FPS / LOW 145 FPS (EAX SOUND)
70Expendable 1280 x 1024 x 32: AVG 202.23 FPS / HIGH 270 FPS / LOW 146 FPS (EAX SOUND)
71Expendable 1280 x 1024 x 32: AVG 147.65 FPS / HIGH 216 FPS / LOW 108 FPS (EAX SOUND / 4x AA + 8x AF)
72Expendable 1280 x 1024 x 32: AVG  55.43 FPS / HIGH  68 FPS / LOW  37 FPS (EAX SOUND / 8x AA + 8x AF)
73
74
75
76Unreal Tournament  640 x  480 x 16: AVG 90.98 FPS / HIGH 146.61 FPS / LOW 53.64 FPS (Full details / utbench)
77Unreal Tournament 1024 x  768 x 16: AVG 88.86 FPS / HIGH 143.89 FPS / LOW 49.05 FPS (Full details / utbench)
78Unreal Tournament 1024 x  768 x 32: AVG 88.94 FPS / HIGH 142.54 FPS / LOW 51.49 FPS (Full details / utbench)
79Unreal Tournament 1280 x 1024 x 16: AVG 87.22 FPS / HIGH 143.25 FPS / LOW 51.35 FPS (Full details / utbench)
80Unreal Tournament 1280 x 1024 x 32: AVG 87.40 FPS / HIGH 140.48 FPS / LOW 51.71 FPS (Full details / utbench)
81Unreal Tournament 1280 x 1024 x 32: AVG 57.03 FPS / HIGH  95.86 FPS / LOW 27.25 FPS (Full details / utbench / 4x AA + 8x AF)
82Unreal Tournament 1280 x 1024 x 32: AVG 33.96 FPS / HIGH  59.06 FPS / LOW 17.51 FPS (Full details / utbench / 8x AA + 8x AF)
83
84
85Unreal  640 x  480 x 16: AVG 248.98 FPS / HIGH 502.14 FPS / LOW 118.42 FPS (Full details)
86Unreal 1024 x  768 x 16: AVG 237.79 FPS / HIGH 528.99 FPS / LOW 116.56 FPS (Full details)
87Unreal 1024 x  768 x 32: AVG 221.62 FPS / HIGH 479.70 FPS / LOW 105.54 FPS (Full details)
88Unreal 1280 x 1024 x 16: AVG 219.16 FPS / HIGH 437.56 FPS / LOW 105.33 FPS (Full details)
89Unreal 1280 x 1024 x 32: AVG 202.13 FPS / HIGH 351.92 FPS / LOW  97.80 FPS (Full details)
90Unreal 1280 x 1024 x 32: AVG 121.93 FPS / HIGH 196.12 FPS / LOW  66.62 FPS (Full details / 4x AA + 8x AF)
91Unreal 1280 x 1024 x 32: AVG  46.38 FPS / HIGH  67.97 FPS / LOW  28.49 FPS (Full details / 8x AA + 8x AF)

1#######################################################################################################################
2#######################################################################################################################
3Leadtek Winfast A350XT GeForce FX 5900XT 128 MB OC @ 5950 Ultra speeds (475 MHz Core / 950 MHz memory)
4#######################################################################################################################
5#######################################################################################################################
6
73DMark99: 24226 3D marks / 57228 CPU marks
83DMark2000: 19660 3D marks / 969 CPU marks
93DMark2001SE: 16536
10
11
12GLQuake  320 x  240 x 16 (window): 2133.1 FPS (demo1 / Default settings)
13GLQuake  400 x  300 x 16 (window): 1925.7 FPS (demo1 / Default settings)
14GLQuake  512 x  384 x 16 (window): 1458.8 FPS (demo1 / Default settings)
15GLQuake  640 x  480 x 16: 1021.2 FPS (demo1 / Default settings)
16GLQuake 1024 x  768 x 16:  526.7 FPS (demo1 / Default settings)
17GLQuake 1024 x  768 x 32:  519.4 FPS (demo1 / Default settings)
18GLQuake 1280 x 1024 x 16:  347.9 FPS (demo1 / Default settings)
19GLQuake 1280 x 1024 x 32:  343.5 FPS (demo1 / Default settings)
20GLQuake 1280 x 1024 x 32:  139.8 FPS (demo1 / Default settings / 4x AA + 8x AF)
21GLQuake 1280 x 1024 x 32:   62.6 FPS (demo1 / Default settings / 8x AA + 8x AF)
22
23
24
25Quake 2  320 x 240 x 16: 763.9 FPS (demo1.dm2 / Default settings)
26Quake 2  400 x 300 x 16: 762.2 FPS (demo1.dm2 / Default settings)
27Quake 2  512 x 384 x 16: 759.1 FPS (demo1.dm2 / Default settings)
28Quake 2  640 x 480 x 16: 749.7 FPS (demo1.dm2 / Default settings)
29Quake 2 1024 x 768 x 16: 569.9 FPS (demo1.dm2 / Default settings)
30Quake 2 1024 x 768 x 32: 547.3 FPS (demo1.dm2 / Default settings)
31Quake 2 1280 x 960 x 16: 404.8 FPS (demo1.dm2 / Default settings)
32Quake 2 1280 x 960 x 32: 389.7 FPS (demo1.dm2 / Default settings)
33Quake 2 1280 x 960 x 32: 187.2 FPS (demo1.dm2 / Default settings / 4x AA + 8x AF)
34Quake 2 1280 x 960 x 32:  62.3 FPS (demo1.dm2 / Default settings / 8x AA + 8x AF)
35
36
37
38Quake 3  640 x  480 x 16: 331.6 FPS (demo four / Default settings)
39Quake 3 1024 x  768 x 16: 328.1 FPS (demo four / Default settings)
40Quake 3 1024 x  768 x 32: 327.4 FPS (demo four / Default settings)
41Quake 3 1280 x 1024 x 16: 316.1 FPS (demo four / Default settings)
42Quake 3 1280 x 1024 x 32: 311.9 FPS (demo four / Default settings)
43Quake 3 1280 x 1024 x 32: 294.1 FPS (demo four / Full details + Trilinear filtering)
44Quake 3 1280 x 1024 x 32: 160.9 FPS (demo four / Full details + Trilinear filtering + 4x AA + 8x AF)
45Quake 3 1280 x 1024 x 32:  52.4 FPS (demo four / Full details + Trilinear filtering + 8x AA + 8x AF)
46
47
48
49MDK2  640 x  480 x 16: 425.01 FPS (Default settings / EAX sound)
50MDK2 1024 x  768 x 16: 422.08 FPS (Default settings / EAX sound)
51MDK2 1024 x  768 x 32: 422.49 FPS (Default settings / EAX sound)
52MDK2 1280 x 1024 x 16: 390.86 FPS (Default settings / EAX sound)
53MDK2 1280 x 1024 x 32: 370.72 FPS (Default settings / EAX sound)
54MDK2 1280 x 1024 x 32: 359.62 FPS (Full details + Trilinear filtering / EAX sound)
55MDK2 1280 x 1024 x 32: 152.84 FPS (Full details + Trilinear filtering + 4x AA + 8x AF / EAX sound)
56MDK2 1280 x 1024 x 32:  50.45 FPS (Full details + Trilinear filtering + 8x AA + 8x AF / EAX sound)
57
58
59Expendable 1280 x 1024 x 32: AVG 207.69 FPS / HIGH 276 FPS / LOW 149 FPS (EAX SOUND)
60Expendable 1280 x 1024 x 32: AVG 181.00 FPS / HIGH 244 FPS / LOW 134 FPS (EAX SOUND / 4x AA + 8x AF)

61Expendable 1280 x 1024 x 32: AVG  70.68 FPS / HIGH  87 FPS / LOW  39 FPS (EAX SOUND / 8x AA + 8x AF)
62
63
64Unreal Tournament 1280 x 1024 x 32: AVG 86.63 FPS / HIGH 137.30 FPS / LOW 50.31 FPS (Full details / utbench)
65Unreal Tournament 1280 x 1024 x 32: AVG 58.75 FPS / HIGH  98.07 FPS / LOW 28.05 FPS (Full details / utbench / 4x AA + 8x AF)
66Unreal Tournament 1280 x 1024 x 32: AVG 36.93 FPS / HIGH  66.59 FPS / LOW 17.61 FPS (Full details / utbench / 8x AA + 8x AF)
67
68
69Unreal 1280 x 1024 x 32: AVG 214.60 FPS / HIGH 430.43 FPS / LOW 93.99 FPS (Full details)
70Unreal 1280 x 1024 x 32: AVG 159.21 FPS / HIGH 254.30 FPS / LOW 83.15 FPS (Full details / 4x AA + 8x AF)
71Unreal 1280 x 1024 x 32: AVG  60.01 FPS / HIGH  86.12 FPS / LOW 38.62 FPS (Full details / 8x AA + 8x AF)

bloodem wrote on 2022-08-08, 12:24:

I was finally able to run the GeForce FX 5900XT benchmarks as promised. Sorry for the delay, but... life happened and I had to return to my hometown.
I initially started doing the benchmarks on a Pentium Dual Core E5800 build, but I now switched to an Athlon 64 build that I have in this other location.

Thanks! This will give me a nice baseline for testing my own 5900XT. If it ever passes customs I guess. It will also be interesting to see how much faster the A64 Venice will be than the NewCastle core which I'm now using.

- As you can see, in GLQuake the card is bottlenecking the CPU even at 640 x 480, so I've decided to also test MUCH lower resolutions just for fun (at least in some games) . 😀

Heh, figured as much. 😁 Even with a GeForce4 Ti4200 and a NewCastle 3400+ I'm currently getting 706 FPS in Quake 2 @ 640x480x32. BTW, were you using the 3DNow! optimized version of Quake 2 for your benchmarks or not?

- if you'd like me to specifically test something else, let me know and I'll do my best (I'll be near this PC until Saturday)

Just the standard UT99 flyby intro benchmarks, if you don't mind going through those. Like you said, the utbench demo is not indicative of in-game performance. Neither is the flyby intro, but I still find it to be a better match.

Joseph_Joestar wrote on 2022-08-08, 12:46:

Thanks! This will give me a nice baseline for testing my own 5900XT. If it ever passes customs I guess. It will also be interesting to see how much faster the A64 Venice will be than the NewCastle core which I'm now using.

The CPU is indeed very fast, especially when overclocked at 2.7 GHz. Not as fast as a Core 2 Duo / Pentium Dual Core, but... those 2000+ FPS in GLQuake speak for themselves. 🤣
At some point I might replace the CPU with a 4000+ (with 1 MB cache), I have a bunch of those and never got to even test them out (except one that I'm using in an nForce 3 build which is also VERY fast).
Another thing to note is that, even though the motherboard has Dual Channel support, I am using a single 1 GB stick of RAM (unfortunately don't have any spare memory in this location), so the memory bandwidth is probably half of what it should/could be). So there's a little bit of performance left on the table at low resolutions (but probably not much, at least not in Win98 games). Anyway, when you get to ~1000 FPS... all the extra performance is just for show. 😀

Heh, figured as much. 😁 Even with a GeForce4 Ti4200 and a NewCastle 3400+ I'm currently getting 706 FPS in Quake 2 @ 640x480x32. BTW, were you using the 3DNow! optimized version of Quake 2 for your benchmarks or not?

I never use the 3DNow! version of Quake 2 on anything newer than a K6-3+ CPU. Starting with the Athlon, the FPU got so powerful that there is usually very little extra performance when switching to 3DNow! (and the 3DNow implementation in Quake 2 was without a doubt specifically optimized for the K6-2/3 line of CPUs). Anyway, at some point I tested the 3DNow version on an Athlon XP and I think performance was actually 1-2 FPS slower. 😀

Just the standard UT99 flyby intro benchmarks, if you don't mind going through those. Like you said, the utbench demo is not indicative of in-game performance. Neither is the flyby intro, but I still find it to be a better match.

Will do!

bloodem wrote on 2022-08-08, 13:58:

Another thing to note is that, even though the motherboard has Dual Channel support, I am using a single 1 GB stick of RAM (unfortunately don't have any spare memory in this location), so the memory bandwidth is probably half of what it should/could be).

That actually works better for me, since my socket 754 based K8V-MX is of course only single-channel. So our performance should be somewhat similar, with socket 939 still having an edge due to 1000 MHz FSB and the Venice core CPU.

I never use the 3DNow! version of Quake 2 on anything newer than a K6-3+ CPU. Starting with the Athlon, the FPU got so powerful that there is usually very little extra performance when switching to 3DNow! (and the 3DNow implementation in Quake 2 was without a doubt specifically optimized for the K6-2/3 line of CPUs). Anyway, at some point I tested the 3DNow version on an Athlon XP and I think performance was actually 1-2 FPS slower. 😀

I think I got a slight performance boost on my AthlonXP + Voodoo3 system with that patch, but it's been a while since I checked. You're right that it doesn't seem to make much difference on the Athlon64 though. I just tested it and the 3DNow! vs. stock Quake 2 results are within margin of error.

Nice, clean build, enough powerfull for W98/ME. I build something similar, only based on nForce3 and with Turion CPU and Geforce 6800.

Ydee wrote on 2022-08-09, 11:29:

Nice, clean build, enough powerfull for W98/ME. I build something similar, only based on nForce3 and with Turion CPU and Geforce 6800.

Cheers!

Out of curiosity, how has your experience with the 6800 been in terms of compatibility with Win9x games? I briefly tested a 6200 AGP with 77.72 drivers (that particular card couldn't take lower driver versions) and I encountered graphical oddities in certain games like Dino Crisis.

Yes, I read it before, but personally I'm still "under construction" - I wanted the strongest possible GPU with official drivers for W98SE because I have a dualboot with WXP.
If I run into problems with older games, I still have the FX5700 and GF4Ti4200 (like you) as backup. When I'm done, I'll start a thread here and share the experience.

Joseph_Joestar wrote on 2022-08-08, 12:46:

Just the standard UT99 flyby intro benchmarks, if you don't mind going through those. Like you said, the utbench demo is not indicative of in-game performance. Neither is the flyby intro, but I still find it to be a better match.

Ready, see the benchmarks below.

Again, some observations:
- the flyby demo is generally faster than utbench, however it's still usually below what I'm seeing during 30 - 60 minutes of actual gameplay with 16 bots on a few maps that I tested.
- at 4x AA and 8x AA, there are "symptoms" of VRAM exhaustion (stuttering, very bad frametimes, no real difference between the card running at default and FX5950 Ultra clocks, etc), although it might also be some other game/driver bug related to AA in this game.

In the following days I will try to do a few tests with driver version 56.64, I'm curious to see how it compares to 45.23 (I remember someone mentioned that it should be faster, though I remember it being slower). What I do remember about 56.64 is that the Coolbits registry tweak doesn't work anymore, and even though the driver has general and per-game vsync settings... those rarely work.

1#######################################################################################################################
2Leadtek Winfast A350XT GeForce FX 5900XT 128 MB (default clocks: 390 MHz Core / 700 MHz memory)
3#######################################################################################################################
4
5Unreal Tournament  640 x  480 x 16: AVG 171.36 FPS / HIGH 277.94 FPS / LOW 105.46 FPS (Full details / flyby intro)
6Unreal Tournament 1024 x  768 x 16: AVG 168.54 FPS / HIGH 278.03 FPS / LOW 101.22 FPS (Full details / flyby intro)
7Unreal Tournament 1024 x  768 x 32: AVG 163.44 FPS / HIGH 271.03 FPS / LOW 100.25 FPS (Full details / flyby intro)
8Unreal Tournament 1280 x 1024 x 16: AVG 164.06 FPS / HIGH 271.46 FPS / LOW  98.72 FPS (Full details / flyby intro)
9Unreal Tournament 1280 x 1024 x 32: AVG 163.10 FPS / HIGH 271.78 FPS / LOW  98.05 FPS (Full details / flyby intro)
10Unreal Tournament 1280 x 1024 x 32: AVG  92.01 FPS / HIGH 240.49 FPS / LOW  18.95 FPS (Full details / flyby intro / 4x AA + 8x AF)
11Unreal Tournament 1280 x 1024 x 32: AVG  43.53 FPS / HIGH  81.11 FPS / LOW  11.26 FPS (Full details / flyby intro / 8x AA + 8x AF)
12
13
14#######################################################################################################################
15Leadtek Winfast A350XT GeForce FX 5900XT 128 MB OC @ 5950 Ultra speeds (475 MHz Core / 950 MHz memory)
16#######################################################################################################################
17
18Unreal Tournament 1280 x 1024 x 32: AVG 181.82 FPS / HIGH 309.05 FPS / LOW 105.48 FPS (Full details / flyby intro)
19Unreal Tournament 1280 x 1024 x 32: AVG  94.36 FPS / HIGH 271.13 FPS / LOW  19.86 FPS (Full details / flyby intro / 4x AA + 8x AF)
20Unreal Tournament 1280 x 1024 x 32: AVG  43.74 FPS / HIGH  81.16 FPS / LOW   9.84 FPS (Full details / flyby intro / 8x AA + 8x AF)

bloodem wrote on 2022-08-09, 17:36:

- the flyby demo is generally faster than utbench, however it's still usually below what I'm seeing during 30 - 60 minutes of actual gameplay with 16 bots on a few maps that I tested.

Cheers! The flyby benchmarks without AA and AF look fairly similar to what I'm currently seeing on my Ti4200, so I'm guessing the game remains CPU bound. And yeah, actual gameplay is faster on most maps, especially the smaller ones.

- at 4x AA and 8x AA, there are "symptoms" of VRAM exhaustion (stuttering, very bad frametimes, no real difference between the card running at default and FX5950 Ultra clocks, etc), although it might also be some other game/driver bug related to AA in this game.

This is very interesting. When AA and AF are cranked up, it looks like we start hitting the limit of what's possible with 128 MB VRAM. I guess ATi was smart for putting 256 MB on many Radeon 9600 Pro/XT cards from that era.

What I do remember about 56.64 is that the Coolbits registry tweak doesn't work anymore, and even though the driver has general and per-game vsync settings... those rarely work.

Personally, I only ever used 56.64 with my GeForce FX 5700LE or while doing specific compatibility tests. But I do remember that I needed to run RivaTuner for even the simplest tweaks with that driver, such as disabling VSync. That version always felt kind of unfinished to me, in terms of available settings.

Joseph_Joestar wrote on 2022-08-09, 18:07:

Cheers! The flyby benchmarks without AA and AF look fairly similar to what I'm currently seeing on my Ti4200, so I'm guessing the game remains CPU bound. And yeah, actual gameplay is faster on most maps, especially the smaller ones.

I'm starting to think that it's more like "bad-coding bound" than CPU bound. 😁 I think that faster CPUs partially fix the effect, not the actual cause (and even with VERY fast CPUs, utbench is still way too slow). So there must be some other software bottleneck at play.

Joseph_Joestar wrote on 2022-08-09, 18:07:

This is very interesting. When AA and AF are cranked up, it looks like we start hitting the limit of what's possible with 128 MB VRAM. I guess ATi was smart for putting 256 MB on many Radeon 9600 Pro/XT cards from that era.

Indeed! I'm still not entirely sure that this is actually the issue, but it sure looks like it.

Joseph_Joestar wrote on 2022-08-09, 18:07:

Personally, I only ever used 56.64 with my GeForce FX 5700LE or while doing specific compatibility tests. But I do remember that I needed to run RivaTuner for even the simplest tweaks with that driver, such as disabling VSync. That version always felt kind of unfinished to me, in terms of available settings.

So far, 56.64 does appear to be slightly faster in certain benchmarks (particularly 3DMark 99 / 2000 / 2001 - this might just be good ol' nVIDIA being... nVIDIA and cheating their way "to the top").
In games, it seems to be more or less the same as 45.23 (sometimes a bit slower).

Regarding vsync, disabling it technically works from within the nVIDIA control panel, however this part is clearly buggy (as you said, it feels unfinished).
I'm not even sure how I managed to disable vSYNC, but here's basically what I did:
- accessed the nVIDIA control panel --> Performance & Quality Settings
- funnily enough, vsync appears to actually be disabled by default in the Global section (but this is not really the case)
- I enabled vsync & rebooted
- accessed the nVIDIA control panel again --> Performance & Quality Settings
- I disabled vsync & rebooted

After that, vsync was disabled in most titles (3DMark 99, GLQuake, etc), however it was still enabled in Quake 2.
I found a workaround (by accident), though: as mentioned, when you first launch Quake 2, vsync is enabled. However, as soon as you change the resolution, vsync is disabled until you exit the game. 🤣

bloodem wrote on 2022-08-10, 04:52:

After that, vsync was disabled in most titles (3DMark 99, GLQuake, etc), however it was still enabled in Quake 2.
I found a workaround (by accident), though: as mentioned, when you first launch Quake 2, vsync is enabled. However, as soon as you change the resolution, vsync is disabled until you exit the game. 🤣

Heh, I think that's pretty much what I saw while I was testing that driver on my 5700. The behavior is pretty inconsistent unless RivaTuner is used.

I noticed that the reviewers back in the day used 52.16 for the 5900XT and apparently had good results. I don't think I ever tried that version (yet), so I'm not sure if the interface situation is any better.

Joseph_Joestar wrote on 2022-08-10, 06:44:

I noticed that the reviewers back in the day used 52.16 for the 5900XT and apparently had good results. I don't think I ever tried that version (yet), so I'm not sure if the interface situation is any better.

I had never heard about this driver... and now I know why. 😁
I've just tried it and... it didn't go well, it's a complete BSOD/freeze fest.

My guess is that reviewers were using that driver on WinXP, since by that time nobody cared about Win98 anymore.

So, after a bit of testing, I've again concluded that, even if a forced installation is required, v45.23 is still miles better for GeForce FX cards that work with it (because not all do, but it just so happens that 5900XT works absolutely fine 😀 ).

My Leadtek GeForce FX 5900XT finally decided to show up. The card seems to be in excellent condition and the stock fan is quiet and efficient. I haven't done any detailed performance testing yet, but my preliminary results are a bit unusual. Quake 2 at 1280x960 is actually slightly slower than it was on my GeForce4 Ti4200. On that card I had 302 FPS, while I only have 292 FPS on the FX5900XT. This is also below what @bloodem had on his 5900XT. I'll have to investigate this further. On the other hand, 3DMark scores seem to be mostly ok i.e. 15552 in 3DMark 2001SE with 45.23 drivers (forced).

I also got this 19" LG Flatron L1953HR monitor. Same brand as my old one, same series, just bigger. Back when I bought the 17" version (around 2006 I think?), I always felt like I missed out for not getting the 19"model. It was out of my budget range at that time, but it's fairly affordable now, and I'm very happy with having a bit more real estate on my retro rig. Oh, and in case anyone's wondering, that fairy is from Nvidia's Dawn demo.

More benchmarking, tweaking and pics will likely follow during the next couple of weeks, as time permits. I'm currently in the process of moving so things are a bit hectic.

Nice, enjoy your new card! 😀
It's good that the fan is still working perfectly. I had to replace it on both of my A350XT cards (bought three original fans from a chinese seller a couple of years ago).

I think I've mentioned this in the past, but as crazy as it sounds, make sure to restart the OS multiple times and do the same benchmark.
It's a weird behavior that I noticed with some of these late VIA chipsets: the 3D performance can sometimes be ~ 10% slower once in a few restarts (not entirely sure what's causing it, but never bothered investigating it further - since 400 FPS vs 450 FPS is not that big of a difference 😁 ).

Having said that, you are not wrong, I've also noticed that the GeForce 4 Ti cards can sometimes be faster than the GeForce FX 5900 (especially with the older 30.82 drivers that I'm using with those cards).

For comparison, here's some benchmark results from a Core 2 Duo platform + a GeForce 4 Ti 4400 (driver version 30.82).
In situations where the game is mostly CPU bottlenecked (Expendable), the difference between the Athlon 64 and Core 2 Duo results does make sense. However, there are GPU bottlenecked scenarios (Quake 2 at higher resolutions), where the GeForce 4 Ti 4400 is faster than the GeForce FX (or way too close). This is probably either an architecture limitation, or maybe a driver issue that affects certain games (related specifically to the GeForce FX series).
Looking at the actual specs of the cards, when it comes to both memory bandwidth and fillrate, the FX 59xx series should absolutely destroy the 4 Ti series at higher resolutions, but this is not always the case.
However, when using AA and AF, the FX series always comes out on top (sometimes by a very large margin), so there's that...

13DMark2000: 20242 points
23DMark2001: 14743 points
3
4GLQuake @  640 x  480 x 16: 879 FPS
5GLQuake @  640 x  480 x 32: 808 FPS
6GLQuake @ 1024 x  768 x 32: 413 FPS
7GLQuake @ 1600 x 1200 x 32: 194 FPS
8
9Quake 2 @  512 x  384 x 32: 975 FPS
10Quake 2 @  640 x  480 x 32: 918 FPS
11Quake 2 @ 1024 x  768 x 32: 510 FPS
12Quake 2 @ 1600 x 1200 x 32: 238 FPS
13
14Quake 3 @  512 x  384 x 16 (Default details): 803 FPS
15Quake 3 @  640 x  480 x 16 (Default details): 685 FPS
16Quake 3 @ 1024 x  768 x 32 (Default details): 313 FPS
17Quake 3 @ 1600 x 1200 x 32 (Default details): 134 FPS
18Quake 3 @ 1600 x 1200 x 32 (Full texture detail / trilinear filter): 118 FPS
19
20
21NFS 4 @  640 x  480 x 16 (Max details): Avg: 63.999 / Min: 64 / Max: 65
22NFS 4 @ 1024 x  768 x 16 (Max details): Avg: 64.000 / Min: 63 / Max: 66
23NFS 4 @ 1600 x 1200 x 16 (Max details): Avg: 63.549 / Min: 56 / Max: 65
24NFS 4 @ 1920 x 1440 x 16 (Max details): Avg: 60.363 / Min: 42 / Max: 65
25
26NFS 5 @  640 x  480 x 32 (Max details): Avg: 86.624 - Min: 73 - Max: 102
27NFS 5 @ 1024 x  768 x 32 (Max details): Avg: 86.579 / Min: 72 / Max: 102
28NFS 5 @ 1600 x 1200 x 32 (Max details): Avg: 76.614 / Min: 58 / Max:  97
29NFS 5 @ 1920 x 1440 x 32 (Max details): Avg: 62.745 / Min: 40 / Max:  88
30
31
32Expendable @  512 x  384 x 16 (Default details): Avg: 552.40 / Min: 413 / Max: 742
33Expendable @  640 x  480 x 32 (Default details): Avg: 535.85 / Min: 395 / Max: 699
34Expendable @ 1024 x  768 x 32 (Default details): Avg: 353.83 / Min: 277 / Max: 446
35Expendable @ 1600 x 1200 x 32 (Default details): Avg: 160.01 / Min: 116 / Max: 215
36Expendable @ 1920 x 1440 x 32 (Default details): Avg: 114.26 / Min:  83 / Max: 167
37
38Re-Volt @  640 x  480 x 32 (Max details): Avg: 840.920 / Min: 732 / Max: 1024
39Re-Volt @ 1024 x 1768 x 32 (Max details): Avg: 401.396 / Min: 283 / Max:  592
40Re-Volt @ 1600 x 1200 x 32 (Max details): Avg: 178.754 / Min: 136 / Max:  253
41Re-Volt @ 1920 x 1440 x 32 (Max details): Avg: 122.995 / Min:  86 / Max:  179
42
43Unreal @  640 x  480 x 32 (Max details): Avg: 294.36 / Min: 129.88 / Max: 644.17
44Unreal @ 1024 x  768 x 32 (Max details): Avg: 230.67 / Min: 120.01 / Max: 371.66
45Unreal @ 1280 x 1024 x 32 (Max details): Avg: 170.35 / Min:  98.97 / Max: 250.29

That leadtek is a nice looking card. I like how the cooler has a dust filter built in.

chrismeyer6 wrote on 2022-08-12, 11:51:

That leadtek is a nice looking card. I like how the cooler has a dust filter built in.

Yeah, it's pretty nice.

I like the automated fan control too. The card is nearly silent at the desktop and it ramps up as needed during 3D gaming. But even at full load, the noise levels are quite reasonable.

bloodem wrote on 2022-08-12, 11:49:

Nice, enjoy your new card! 😀
It's good that the fan is still working perfectly. I had to replace it on both of my A350XT cards (bought three original fans from a chinese seller a couple of years ago).

Cheers! 😀 BTW, I have a question with regards to forcing the 45.23 driver. Currently, I'm manually installing the 5900XT as an 5900 Ultra, see pic:

However, this results in the driver info screen listing the GPU as "NVIDIA (Unknown)" and games which use this info (e.g. Quake 2) also list the card as such. I've also tried adding the following lines to NVAML.INF in their respective sections:

1%NVIDIA&DEV_0332.DeviceDesc% = NV30,      PCI\VEN_10DE&DEV_0332
2
3NVIDIA&DEV_0332.DeviceDesc="NVIDIA GeForce FX 5900XT"

Those lines were taken from the 56.64 driver. And by doing that, I can install the card as an 5900XT, but the driver info screen still lists it as an "NVIDIA (Unknown)" GPU. Can this be fixed and does it even matter? I gather it's probably just a cosmetic issue.

How are the temps with that cooler?

chrismeyer6 wrote on 2022-08-12, 18:26:

How are the temps with that cooler?

At stock clocks, around 38C idling at the desktop and 59C under full load.

This is with an suboptimal environment though. My new place doesn't have air conditioning (yet) and my retro rigs are packed too closely together. I also plan on adding an intake and an exhaust fan to the case, since I'm now using more powerful hardware.

Joseph_Joestar wrote on 2022-08-12, 17:45:

Those lines were taken from the 56.64 driver. And by doing that, I can install the card as an 5900XT, but the driver info screen still lists it as an "NVIDIA (Unknown)" GPU. Can this be fixed and does it even matter? I gather it's probably just a cosmetic issue.

This has also "bothered" me, but I confirm that it's purely a cosmetic issue, indeed.
As far as I can tell, it has nothing to do with the INF file, it seems to be something more low level (a GPU core identification routine built into the driver itself). Now, what I find odd is the fact that both the GeForce FX5900(Ultra) and the GeForce FX5900XT are NV35 chips, so you'd think that the driver should just identify the XT as a FX5900. But, no, it seems there are also other differences that the driver is able to somehow spot.
Either way, it's not something that affects game performance/compatibility in any way.