Are these benches for real?

[PingSpike]

Text

I would have thought the newer cards would still fair better in older games, even though those games don't use their new features.

 

[SonicIce]

I think that benckmark is WAY too CPU limited.

 

[Matthias99]

Originally posted by: SonicIce
I think that benckmark is WAY too CPU limited.

Ya think? It's running on an XP3000+ with a VIA KT333-based chipset. Any differences you see in older games like Quake3 or RTCW (especially at the low settings they're using) are because of tiny differences in driver efficiency, since they're totally CPU-limited on any newer hardware. If they ran the benches at 1600x1200 with AA and AF on, you'd see the results you'd expect; the newer cards would be far, far faster than the older ones.

 

[uOpt]

The GeForce T 4x00 series is actually very fast when it comes to just push polygons with textures around and add some hardware accelerated transform and lightning.

Only when you want avanced texture filters and/or if your application is making use of new vertex or pixel shaders are the newer cards actually better.

 

[uOpt]

Also, the game is probably drawing a lot of textures in layers, given the results.

In the GeForce 4x00 and 5x00 each pixel pipleline has two hardware units to draw textures. Newer cards drop that feature in favor of more hardware support for shaders. If you game doesn't use the shaders but does extensive texture drawing - well there you go.

 

[Matthias99]

Originally posted by: MartinCracauer
The GeForce T 4x00 series is actually very fast when it comes to just push polygons with textures around and add some hardware accelerated transform and lightning.

Only when you want avanced texture filters and/or if your application is making use of new vertex or pixel shaders are the newer cards actually better.

Uh... only if you totally discount the unbelievable fillrate and bandwidth advantages of the newer cards.

6800GT -- 16 pipes * 350Mhz = 5600MPixels/sec.
9800Pro -- 8 pipes * 380Mhz = 3040MPixels/sec.
Ti4600 -- 4 pipes * 300Mhz = 1200MPixels/sec.

6800GT -- 1000Mhz * 256-bits/clock = 32GB/sec. bandwidth
9800Pro -- 680 * 256-bits/clock = 21GB/sec. bandwidth
Ti4600 -- 650 * 128-bits/clock = 10GB/sec. bandwidth

The 4600 starts to fall apart in newer games at higher resolutions, even without fancy shaders and AA/AF. Newer cards have a lot more headroom.

That said, if all you want to do is play older games (like Counter-Strike, Quake3, RTCW, etc.), or don't mind dropping to 1024x768 on newer games (and having cruddy shader performance), a card faster than a Ti4600 is obviously overkill.

Edit: You do have a point about multitexturing; the 4600 has 8 texturing units, so it runs pretty decently when multitexturing (2400MTexels/sec.) unless it starts running out of memory bandwidth.