Performance increases for new generation cards (Nvidia)

silent_guy

Veteran
Supporter
I was a little bored tonight and since I had been wondering for a while by how much performance of GPU's has increased when they go from one generation to the next, this was the moment to dig up some numbers.

Here are the criteria that I used:
- use data on anandtech.com, since they have existed for a long time and results are fairly easy to search.
- find the introduction article for a particular GPU
- take a benchmark at a particular resolution that was typical for that time.
- divide number of the highest performing newcomer and divide by the performance of highest performing previous generation card
- Sometimes arbitrary choices had to be made. When there were results for both 16-bit mode (remember those?) and 32-bit mode, I choose 16-bit. For later models, I chose higher quality settings (e.g. 4x AA/AF).

I started with Nvidia since they have had the most consistent roadmap over the years. The benchmarks used are mostly based on an id Software engine.

Here are the numbers:
TNT / 128: 2.25 (Forsaken 800x600)
TNT2 / TNT: 1.53 (Quake 2 800x600)
Gf 256 / TNT2 Ultra: 1.52 (Quake 3 1024x768)
Gf 2 GTS / Gf 256 DDR: 1.75 (Quake 3 1028x1024)
Gf 3 / Gf2 Ultra: 1.25 (Quake 3 1600x1200)
Gf 4 Ti 4600 / Gf 3 Ti 500: 1.30 (Quake 3 1600x1200)
Gf FX 5800 / Gf 4 Ti 4600: 1.47 (Quake 3 1600x1200)
Gf 6800 Ultra Extreme / Gf FX 5950: 1.57 (Wolfenstein 1600x1200)
Gf 7800 GTX / Gf 6800 Ultra: 1.18 (Wolfenstein 1600x1200)
Gf 7900 GTX / Gf 7800 GTX512: 1.04 (Quake 4)

* The Gf FX 5800 was soo late that it's predecessor was completely obsolete. Anandtech didin't bother to compare them. I used number from digit-life.com. I should have compared with the Ti4800 because I believe that one was faster, but I didn't find numbers.
* I added the 7900 because it was the last one released...

All numbers geometrically averaged, not including the 7900, the average speed increase between generations is x1.50.

It will be interesting to see what's going to happen with g80. ;)

Next up... ATI.
 
silent_guy said:
* The Gf FX 5800 was soo late that it's predecessor was completely obsolete. Anandtech didin't bother to compare them. I used number from digit-life.com. I should have compared with the Ti4800 because I believe that one was faster, but I didn't find numbers.
Ti4800 = 4600 AGP 8x, there shouldn't be any speed differences.

While you're on it, could you throw some numbers on the D3D side too, as your choices (perhaps excluding Forsaken, can't remember if it supported D3D or OpenGL besides Glide) are all OpenGL?
 
I think I would not be using 1600x1200 as the touchstone until, oh, maybe this last generation. That brings bandwidth too much into play, especially for some generations that didn't have nearly what we have today.
 
What about a comparison via games designed to take advantage of ATi's architecture for the r580? Or am I mistaken in thinking that it doesn't take the usual pure polygon pushing approach?
 
There's a good idea here...

This thread can be (could have been?) interesting, if there's more, a lot more, games in the comparison and with different resolutions.

As it is now, this comparison is pretty much useless, to be honest.

If someone has the time and the courage to compile a more complete comparison, we might be able to see some trends.
 
Vysez said:
This thread can be (could have been?) interesting, if there's more, a lot more, games in the comparison and with different resolutions.

As it is now, this comparison is pretty much useless, to be honest.

If someone has the time and the courage to compile a more complete comparison, we might be able to see some trends.

I'm working on it... (but I haven't been too bored the last 2 days.)
The problem is finding the right data: everytime a new card is released, websites like to tweak benchmarks over and over, so it's almost impossible to use data from different articles.
 
just to let you know on a gf6800gt
1280x1024 32bit no fsaa no af (because thats how the game was played)
i get 390fps

i found a bench for the voodoo3 3000 at same settings = 56fps
 
Vysez said:
This thread can be (could have been?) interesting, if there's more, a lot more, games in the comparison and with different resolutions.

As it is now, this comparison is pretty much useless, to be honest.

If someone has the time and the courage to compile a more complete comparison, we might be able to see some trends.

I did something like this a couple of years ago. I collected about 4,000 sample points from something like 50 different hardware reviews and built model trees (decision trees with statistical components at the leaf nodes) to predict framerate based on various attributes (CPU type, CPU speed, GPU type, gpu speed, memory speed, resolution, AA level, anisotropic level, etc). I think I included something like 7 or 8 games, but some of them didn't have enough sample points to really be meaningful.

I wanted to take the work farther, but I got side tracked with life/job/etc. I think I still have my paper and slides around if you are interested in seeing them.

Nite_Hawk
 
  • Like
Reactions: Geo
Nite_Hawk said:
I did something like this a couple of years ago. I collected about 4,000 sample points from something like 50 different hardware reviews and built model trees (decision trees with statistical components at the leaf nodes) to predict framerate based on various attributes (CPU type, CPU speed, GPU type, gpu speed, memory speed, resolution, AA level, anisotropic level, etc). I think I included something like 7 or 8 games, but some of them didn't have enough sample points to really be meaningful.

I wanted to take the work farther, but I got side tracked with life/job/etc. I think I still have my paper and slides around if you are interested in seeing them.

Nite_Hawk

Dude, that's hardcore. :D A link to that sounds like funsies to me.

Btw, Micro Center for teh win on local bricks and mortar <Twin Cities shout out to the homey>
 
Back
Top