AGP 4x v. 8x

[Lunaryn]

I'm looking at selling a desktop system to a fellow FFXI player whose system is dying. Unfortunately, this will involve shipping, so I can't do a lot of comparative testing between hardware I have to send and hardware he already has on hand. I'm trying to get a sense of the performance comparison between his existing video and mine, particularly as regards FFXI.
The cards in question:
ATI Radeon 8500 LE (R200)
nVidia GeForce FX 5500
This is essentially a question of comparison of a value (but not quite bottom-of-the-line) card to a high-end (but not quite top-of-the-line) card of a prior generation. It has been noted that the 8500 is roughly equivalent in power to the GeForce 4 Ti 4200.
There seem to be two key issues that look like they'd impact performance based on what I do know from comparisons of different video chipsets: Fill rate and AGP bus speed (Memory bandwidth may also matter but is harder to judge since one of the chipsets I've looked at uses the system memory bus). The 8500 has about twice the fill rate as the FX 5500 (2000 vs 1080), and has a higher memory bandwidth (8.0GB/s vs. 6.4 or 3.2 depending on bus width -- Is this just a difference between the AGP 8x and PCI versions of the card?), but it only supports AGP 4x, which has been shown to make a difference in games making heavy use of textures.
I'd like to know if other people who have experience using FFXI with AGP 4x cards can shed some light on the nature and extent of the performance difference between AGP 4x and 8x, and how much that might offset the extra power of the older card.
For reference, I can post some Vana'diel Bench 3 results I've gotten for various systems, but I can't avoid extra factors leaking in since I don't have a single system that can use each chipset.
Emily - Pentium D 805, Radeon 8500 LE: 2908-H
Emie - T2400 @ 1.83GHz, Mobility Radeon X1400: 3249-H
Meredith - AMD Athlon X2 6000+ w/ PCI-E:
w/ Onboard ATI RS690 (X1250): 3125-H
w/ GeForce 7600 GT KO: 8336-H
Note the similar fill rates between the first 3 setups (The Mobility X1400 has a fill rate of 1728 and the RS690 has a fill rate of 1600). Despite differing CPU speeds, each squeezes out nearly the same performance in FFXI (with Emily suffering from some notable sluggishness in some situations, admittedly). The specs for the 7600 GT KO are several times higher than most of the chipsets shown here, but does demonstrate the fact that a system which squeezes out comparable performance with a value Radeon chipset scales up dramatically with high-end hardware, despite the usual claims that CPU will matter more than video hardware.
Also, note that aside from the 8500, all of those scores come from hardware on a PCI-E bus. This may be more significant than the CPU in terms of Emily scoring lower on these benchmarks, but it's difficult to say exactly how much of an impact there is without being able to compare to, say, the AGP 8x rerelease of the GeForce 4 Ti 4200.
Unfortunately, the system with the FX 5500 is apparently incapable of running Vana'diel Bench 3, even though it can handle FFXI itself.
Wikipedia on ATI GPUs
Wikipedia on nVidia GPUs

[Feba]

the ATI will probably be better, simply because nVidias don't play nice with FFXI.

[DakAttack]

Mine always have, but I've heard horror stories.

Also, wall of text. The eyes, they do nothing.

[Feba]

The eyes,

wut?

[Mhurron]

The eyes, they do nothing.

Fail

[Lunaryn]

Well, fortunately, it seems a coworker has a card I can borrow (either a FX5200 or FX5500, he's not quite sure which), so hopefully I'll be able to test this myself in a few days.

[Lunaryn]

The verdict is in. The AGP 8x FX 5200 shows a marked degredation in framerate over the AGP 4x Radeon 8500 LE, clearly indicating that the AGP bus speed is a lesser factor compared to the higher specs of the 8500. Note in particular the most substantial difference is the Fill Rate, as mentioned before; the 8500 LE has twice the maximum fill rate of the FX 5200. There is also a 25% improvement in memory bandwidth (8.0Gb/s vs. 6.4 Gb/s). The clock speed of these two cards, on the other hand, is identical.

Emily returned the following results for a GeForce FX 5200 (AGP 8x):
1768-H
3398-L
Graphics were notably jerky, although this may in part be an issue of low monitor response time, as the Radeon benchmark was a little more jumpy than usual, though not nearly as much as these runs.

Emily returned the following results for the Radeon 8500 LE (AGP 4x):
2987-H
4816-L
Even allowing for a decent margin of error (benchmark scores often vary within a range of close to 100 points), there is a substantial improvement in framerate. The AGP 4x bus doesn't seem to be bottlenecking performance at all relative to the lower-spec card. The framerate seems to be about 50% higher in low resolution setups and about 80% higher in high resolution setups.

Hopefully this will be of use to anyone looking at using older cards with FFXI. It certainly answers my own question as to which card the system should use once deployed.

[InuTrunks]

Just wondering on these tests, was the only thing you swapped out is the video card? or did you change other specs? Also, did you completely wipe out the old videocard drivers? If not your tests are flawed.

Also, what motherboard did you use? Did you make sure you had an AGP8x MB?

[Lunaryn]

This is an AGP 8x board. I don't have the model number off-hand but it's a fairly recent Via-based ECS board running a Pentium D 805 processor. The only hardware changed was the video card. The ATI drivers were removed from the system before testing the nVidia hardware. I was unable to remove the nVidia drivers from the system before running the Radeon scores shown below, but said scores are in line with previous benchmark scores for the same hardware, allowing for around 100 points margin. I'm only particularly interested in the magnitudes of the scores, as represented by the rough percentages listed.

It would be rather enlightening if someone could post comparisons between an AGP 4x and 8x release of the same card, to determine how much difference the bus speed makes when all else is equal, though. I unfortunately lack the appropriate hardware.

Edit: The motherboard is an ECS P4M800PRO-M

[InuTrunks]

http://www.tomshardware.com/

I think that's the link anyway.

Edit: Helps if I can spell.

Edit 2: they changed their URL.

[Aeni]

This is pretty much given that you know this, but 8x AGP does not necessarily give better performance than 4x AGP. It all depends on the kinds of hardware you have in your system, how they are configured and how optimized the drivers enabling all this are.

I remember seeing something similar for WoW in the tech support forum that Blizzard's employees would post in. Pretty much the de facto working standard as far as AGP rate is concerned is 4x. But most people go for the higher setting as bigger number always seem to indicate better performance results.

If you have the exact same setup and hardware that a company running benchmarks have, then by all means, go for the higher settings everytime. But not everything in life is this simple. Just know that higher settings does not necessarily mean you'll get better results.

This is also true when buying hardware. Higher numbers like clock speed can mean different things to different items and does not always translate to "better."

[Lunaryn]

The point of this thread was to compare the impact of specific hardware specs on FFXI itself. There have been points made about 8x AGP being more bus bandwidth than most graphical applications can make effective use of, but there are also questions about FFXI's unusual methodology for rendering graphics. I was hoping to determine whether the bus width was a significant, moderate, or minor influence on FFXI's performance, and I personally am satisfied that it is a fairly minor influence, even for some of the best cards out at the time that AGP 8x was introduced.

[Aeni]

The point of this thread was to compare the impact of specific hardware specs on FFXI itself. There have been points made about 8x AGP being more bus bandwidth than most graphical applications can make effective use of, but there are also questions about FFXI's unusual methodology for rendering graphics. I was hoping to determine whether the bus width was a significant, moderate, or minor influence on FFXI's performance, and I personally am satisfied that it is a fairly minor influence, even for some of the best cards out at the time that AGP 8x was introduced.

The point I was making was irregardless of the software taking advantage or what kind of graphics card you owned, it comes down to bottlenecks within the main system aka motherboard. You could probably get better performance OC'ing the bus, bridge processes, any DSPs along the way as well as the drive controllers etc ... all the way to the timing and speed of the memory, CPU and cache controllers.

In other words, if you want to do a real test, then you'll need to make sure you test everything and not just play musical chairs with the graphics cards.