Agp 8x only really showed improvement on much newer cards.

6800 ultra, x800xt or faster can utilize the bandwidth but ,4x doesn't cripple any of the agp cards until we get to say hd3850 and hd4650.

A ti 4200 will show the same performance in agp2x as it does 4x as it doesn't require the bandwidth.

LittleBubble wrote on 2020-02-05, 06:56:

Same as with the AGP voodoo3 card which was the same as the PCI in speed

Because bus speed and graphics card speed are two different things. It still makes a difference, as the faster the bus is, the shorter the load on it is to transfer a chunk of data.

candle_86 wrote on 2020-02-05, 07:11:

A ti 4200

Had to read that twice.😅

derSammler wrote on 2020-02-05, 08:48:

LittleBubble wrote on 2020-02-05, 06:56:

Same as with the AGP voodoo3 card which was the same as the PCI in speed

Because bus speed and graphics card speed are two different things. It still makes a difference, as the faster the bus is, the shorter the load on it is to transfer a chunk of data.

Not quite, if the card can't make use of the bandwidth it's not going to. Anything pre 6800 or x800 has no chance of saturating the agp 4xbus, they simply are not powerful enough, they can't process that much data at once period. If the bus was a bottleneck you'd see an improvement as bus speed increases.

I've noticed that the CPU tests in 3DMark 2003 and 2005 really benefit from more bus bandwidth. Yet the actual graphics tests don't benefit much at all, at least once you're past PCI. Strange stuff.

Anyway on a PIII-S at 1628MHz with Apollo Pro 266T, DDR-310 @ 2-2-2-5, FSB at 155, AGP at 77.5, and PCI at 38.75.

3DMark03 CPU score:
GeForce 6800GT @ PCI: 160
GF 6800GT @ AGP 2X: 420
GF 6800GT @ AGP 4x: 536

3DMark03 Overall score:
GeForce 6800GT @ PCI: 6564 <-Wings of Fury took the largest hit, by far.
GF 6800GT @ AGP 2X: 9910
GF 6800GT @ AGP 4x: 10076

ATI's software Truform processing on non-R200 cards. Anything which "supports" Truform but isn't a Radeon 8500 or 9100. There is a huge increase in AGP traffic because the CPU is partially doing the geometry processing. You will see a substantial bus bottleneck at AGP 2x, some at 4x, but by 8x it will mostly be a CPU bottleneck.

Also, AGP readback performance improves with later AGP revisions and this is useful for things like Glide wrappers in some cases. It's still very slow compared to PCIe x16 however.

There's also the consideration that early AGP chipsets not from Intel tended to be problematic. So besides just improved specs, you have AGP and its various functions working better in general for VIA, ALI, NVidia, SIS, AMD. AGP4x chipsets were still often rather flakey but AGP8x boards tend to be solid. The main problems that occur with 8x are usually related to those cards with PCIe bridge chips AFAIK.

@ Standard Def Steve

Thanks for your 3Dmark03 values. Interesting the difference between PCI and AGP2x/4x.

I have benched my 6600GT already. Later, i will try also 6800(GT)...

Interesting the difference between PCI and AGP2x/4x.

That's not real PCI though. More like PCI-X with exclusive and fast access to RAM. The real PCI bus would be even worse.