Very nice benchmarks, i know how much work it is and to actually finish them and make the graphs takes a lot of effort.

I personally think AMD does an amazing job with Ati up to this point. Ati wasn't in the best place when AMD bought it and i was afraid that AMD would screw up and Ati division would be gone.
They are still here and made some great products.

Right - so I merged the first and 2nd posts to make room for centralized charts - overall results, with all brands / models, sorted by performance:

The results speak for themselves. It's worth noting that nvidia card results are heavily influenced by driver version, with performance figures variations of as much as 25% for some cards. This is most notable in unreal. It's possible that the higher end / newer nvidia cards would have gotten better performace with the same drivers at a higher resolution, but do consider the sheer ammount of time testing and gathering the results takes - I'm not looking foreward to re-doing them.

I will be updating this thread much much later with 32 bit 1600x1200 results in all games, and I will just be testing the faster cards - but first I have to find some time to test the 3dfx cards as well...

I think drivers are always a problem when benchmarking a lot of cards. It's never right or perfect.

Example: I did some benchmarking of my fastest agp cards (windows xp). I wanted to bench the fastest from each series from Nvidia (TNT, TNT2ultra, Gf3Ti, Gf4Ti, Gf5ultra etc. etc.) but i could only find one driver for cards from gf4 up to gf7. I wanted to use a single driver just to make it easier and to keep things equal. Older driver would drop support for gf7 and later drivers had no support for gf4.... 😐

So i dropped gf3 and lower and went testing. All went well but after the results (which looked fine) i kept thinking: gf4 might be faster with older drivers and gf7 with later ones.... 😢
So i tested an older driver and indeed the gf4 was quicker... i didn't test it with all the other cards but probably every single series has their own ''perfect driver'' (would be logical anyway).

So to run the ''perfect benchmarks'' i would have to figure out first which driver is the absolute best overall. It needs to be done for every single card to give them the most speed. 😵
That is a LOT of work but i think it is the only way to do it right, am i wrong?
The only problem with it maybe is that a driver for example can be the absolute fastest in 3dmark99 and it can be a bad driver for 3dmark2000 (or any other title).

Also, a too high resolution is a bit unfair for an older card like for example a TNT1 but a low resolution creates a bottleneck for a higher end card and it can not stretch it's legs.

Since we all are using and testing cards that have many years between them this will always be a problem driver-wise.

meljor wrote:

I think drivers are always a problem when benchmarking a lot of cards. It's never right or perfect. […]
Show full quote

I think drivers are always a problem when benchmarking a lot of cards. It's never right or perfect.

Example: I did some benchmarking of my fastest agp cards (windows xp). I wanted to bench the fastest from each series from Nvidia (TNT, TNT2ultra, Gf3Ti, Gf4Ti, Gf5ultra etc. etc.) but i could only find one driver for cards from gf4 up to gf7. I wanted to use a single driver just to make it easier and to keep things equal. Older driver would drop support for gf7 and later drivers had no support for gf4.... 😐

So i dropped gf3 and lower and went testing. All went well but after the results (which looked fine) i kept thinking: gf4 might be faster with older drivers and gf7 with later ones.... 😢
So i tested an older driver and indeed the gf4 was quicker... i didn't test it with all the other cards but probably every single series has their own ''perfect driver'' (would be logical anyway).

So to run the ''perfect benchmarks'' i would have to figure out first which driver is the absolute best overall. It needs to be done for every single card to give them the most speed. 😵
That is a LOT of work but i think it is the only way to do it right, am i wrong?
The only problem with it maybe is that a driver for example can be the absolute fastest in 3dmark99 and it can be a bad driver for 3dmark2000 (or any other title).

^Exactly

meljor wrote:

Also, a too high resolution is a bit unfair for an older card like for example a TNT1 but a low resolution creates a bottleneck for a higher end card and it can not stretch it's legs.

Well, no, I don't think it's unfair since we are talking about games here, and the better they look, the happier most gamers will be. Sure, 1600x1200 is a stretch for a TNT2, but it doesn't perform too well at 1280x1024 either in some games, and that was the go-to "high" resolution back in the day. You make valid points everyware - benchmarking at a low resolution creates CPU bottlenecks, not allowing newer cards to perform as well as they shoud - but I as a retro gamer find it relevant since I play as many games as I can at the highest resolution the game and my monitor will support.

Unfair was not the right word, but you know what i mean: low res bench is good for old cards but bad for much faster, newer ones and vice versa when using very high res.

Looking forward to the 3dfx tests. Driver choices can be a fair bit more straight forward with them unless you use 3th party ones. Lots and lots of them... 🤣

Good luck and keep up the good work.

The passive cooled only 64Bit memory-bus GF6200 performs VERY well.