I've found ATI cards to also work quite well. Nvidia cards had some issues with Unreal. I know it's best played with a 3dfx, but ATI played it smooth, whereas some Nvidia cards stuttered. UT was fine though. Incoming has that colour palette issue on later NV cards, works fine on TNT2. That game also works fine on ATI.

One problem I have is a lack of games to test to really make a call. It seems that one way or another, try enough games, and you will run into some issue.

Doesn't the FX card have benefits with enabling AA and still having enough performance? This is mentioned in the VOGONS wiki, but I've never played around with it.

I know that with AMD cards, you can set the quality slider all the way to the right, which will enable 2x AA and other features.

In the reviews from way back then I remember that enabling AA had a bigger performance impact on the 4Ti cards compared to FX. IMHO the biggest issue when comparing them is the numbering. The Geforce4 Ti 4200 (mid-high end) is faster than any FX 5200 (budget card), since the 5200 had the same target audience of the MX cards.

With Unreal D3D you sometimes needed to swap in a different d3ddrv.dll file for better performance. Wheel of Time has this issue too and there is even a patch specifically to improve GeForce/TNT performance. It took Epic a long time to get that engine working acceptably with D3D.
https://web.archive.org/web/19991128234143/ht … .epicgames.com/?

NV claims that GeForce FX 5600 and higher have color compression to improve AA performance. But other than 5800/5900, I don't think any will outperform a GF4Ti in most cases.

DirectX performance for Unreal is smooth for me. It should be faster but like you said this game was built for Glide.

Unreal Gold 226
1024x768 32-bit (1280x1024 wasn't an option)
57-FPS (about same as Voodoo3 3000)

goes down to 28-FPS with 4x AA and 2x anisotropic, which I won't be using.

Thanks for the info swaaye.

At least for benchmarking, UT worked well across all the cards I tried. I ended up removing Unreal from the list of benchmarks, it was just not reliable enough.

I have seen a lot of conflicting information about the core config on this card so I have been hesitant to say this but I now feel pretty confident in saying the Quadro4 380XGL stomps the GeForce4 MX440 8x's ass. They use the same chip in two different configurations. The GeForce4 MX440 8x's NV18 core is configured 2:4:2 and the Quadro4 380XGL's NV18GL core is configured 2:4:4 (Pixel Shader : Texture Mapping Unit : Render Output Unit). I was seeing conflicting information that showed they may have been configured identically but I feel that information is false. I do not have a MX440 8x with 128-bit memory bus for a head-to-head shootout but I will try to get one. This is definitely another reason to prefer this card over the similar GeForce. Their PassMark scores even differ.

PassMark
www.videocardbenchmark.net/compare.php? ... cmp[]=1496

Wikipedia
https://en.wikipedia.org/wiki/List_of_Nvidia_ … ng_units#Quadro
https://en.wikipedia.org/wiki/List_of_Nvidia_ … GeForce4_Series

Techpowerup
https://www.techpowerup.com/gpudb/1358/quadro4-380-xgl.html
http://www.techpowerup.com/gpudb/2132/geforce … -mx-440-8x.html

So for me, this is definitely my favorite DirectX 7.0 video card now! Ok, i have to say my favorite DirectX 7.0 AGP video card. I am still looking at PCI video cards.

EDIT:

I did not link any of the conflicting information

That is the thing with wikipedia. Somebody doubles ROP count of Quadro 4 series and the disinfo spreads.

When trying to find old cards, I changed my method a little bit. I kept looking for the model (e.g. TNT2), but now I just find a few specific brand / model and look for that. That way I was able to finally get some proper TNT2 cards, not one of the thousands of TNT2 M64 cards that can be found on eBay 😀

Well I am now armed with a pixel fill rate test. Seems to be accurate as it shows roughly 500 megapixels for the Quadro 100 NVS. I will test the Quadro4 380 XGL tomorrow and know for certain. Not sure why I didn't think of this before 🤣. Anyways with a core speed of 275-MHz and 4 ROPs we should get 1100 megapixels, if there are only 2 ROPs then we will only get 550 megapixels. I'm betting on 1100. I really don't know why I didn't think to just physically test the fillrate... doh.

Here is the test I found

GfxBench

http://www.graphics.stanford.edu/courses/cs44 … ll/gfxbench.zip

GLUT 3.6 (required, just unzip into same directory as GfxBench)

https://www.opengl.org/resources/libraries/gl … /glutdlls36.zip

Good idea. Question regarding 3D Mark, don't these have fill rate test built in?

Also how accurate is using GPU-Z for example?

tried 3D Mark 2000 and its fill rate tests only deal with megatexels, 3D Mark 99 was the same (in that test at 640x480x32 the 128-bit Quadro of course smashed the 64-bit GeForce by like double)

GPU-Z does not run win Windows 98 and I'm not sure if it actually measures anything or just grabs a hardware ID and looks it up in a database

GfxBench was a disappointment and I think it shows there may be only 2 ROPS 🙁

I would love to hear some better ideas on testing pixel fillrate!

I think I found a proper 128-bit GeForce4 MX440 8x for a direct head-to-head comparison so I will post those results in a couple weeks

I will do some more testing tonight probably. While looking for testing applications I ran across this site. Looks to have a lot of useful tools.

http://www.benchmarkhq.ru/english.html?/be_video.html

I usually just plug in another drive with XP and run GPU-Z 😀

My thought with 3D Mark was more to compare two cards, rather than get cold numbers. E.g. to see if a card has significantly higher fill rate or not.

I will try this when comparing a full Radeon 9200 to a 9200 LE, which has half the memory width.

From my internal tables with 3DMark2001-SE:
numbers after the / are Fillrate Single, Multi, Polycount 1, Polycount 8, DOT3, Vertex Shader, Point Sprites
Geforce4 MX440-SE NV17 AGP 4x Gainward 64 MB DDR 128 Bit TV Out, 270/200(400) / 324,6 603 24 4,3 38,6 40,7 7,6
Geforce4 MX460 NV17 AGP 4x Medion/MSI MS-8863 64 MB DDR 128 Bit, VIVO, 300/268(537)/ 562 1057 34 7,2 80,1 56,4 10,3
tested in P4 3GHz with driver 71.89

elianda

I am downloading 3DMark2001-SE now and I will post results. Thanks!

K6-2+ 400 (6x66-MHz)
128MB RAM
Quadro4 380 XGL
Driver 45.23

Painfully slow for games demos 🤣, need more CPU!

EDIT: 3DMark2001 SE

Fill Rate Single: 533.4
Fill Rate Multi: 1049.0
Poly Count 1: 7.4
Poly Count 8: 4.0
DOT3: 29.7
Vertex Shader: 4.1
Point Sprites: 10.1

EDIT:

Quadro4 380 XGL as reported by RivaTuner
256-MHz, 513-MHz effective, 128-bit DDR Memory
275-MHz Core clock

I found a benchmark I think I trust

FillrateBenchmark 2004 v0.92
requires VB6 Runtime and DirectX 9.0

Shows pixel fill rate of 550 megapixels... damn

Could you please put this into some context? Is this good / bad / not what you expected?

I am not sure what you try to accomplish. The average DX7 card has no hardware support for vertex shaders, so the score is a good indicator for CPU power.
With the K6-2+ you score there 4.1. A normal Riva TNT with 90/110 clock is able to score 14.7 with a fast CPU in 3DMark2001-SE.

To get the performance from the late DX7 cards you better go for 2+ GHz CPUs.

I was simply trying to get a definitive answer to what the fill rates are for this card. This is for a time machine with a K6-3+ 450 and 3DMark2001 SE has nothing to do with what will be played on the system, just comparing to elianda's numbers on the same benchmark. Anyway it is a great card that will be going in the new time machine box soon, hopefully, and it will be paired with one or two Voodoo2s.

Now if you want to see this card fly I have run it on Pentium III 850 with great success and I tried it on the Core2 3,2 as well.