AGP 4x to 8x upgrade possible?

[singularity2006]

I've never seen that type of upgrade. You will most likely need a new board. The issue with many of these upgrades is that convertor slots end up giving u a bottleneck by the maximum capacity of the board specs anyway.

[harmonic01]

:sad: That makes me a sad person. Thanx for the info, though.

[kex]

4 x / 8 x make no diffrence at the moment as software dosnt use the hardware . and a new mother board would cost maybe less or more than an converter and a new motherboard mostley likely other new and improved performance.

[harmonic01]

Hmm, I thought AGP speed is software independant.

[kex]

no its a combination of software, motherboard, and graphics card the lowest value will be what they will all perform at ;)

[harmonic01]

Thanx, I'm sure Half Life 2 will take advantage of AGP at 8x. :-)

[elvis]

AGP bandwidth specifications were a great idea back in the day when high-speed video ram cost an arm and a leg, and manufacturers wanted texture storage offloaded to the system ram (accessible via the system memory bus, which is where high-speed AGP came in useful). this way no card would ever run out of video memory.

the problem with AGP to system texture storage is it's anywhere from 10 to 100 times slower than storage on the video card's memory (even in old cards). for modern day cards, the retrieval of local textures is around 1000 time faser than from the system memory.

however, on the flipside: ultra fast memory has been coming down in price for a while now. you can buy an off the shelf video card with 128MB or 256MB of video memory for not much more than a 32MB card would have cost you 2 years ago.

now consider this: you can set your AGP apature to a maximum of 256MB. this means that on any modern system, the maximum amount of system memory resources that an AGP card can borrow for texture storage is 256MB. this was great for 32MB cards, but for 128 and 256MB cards this is barely usable.

any comlexly textured game using newer shader features will certainly take advantage of the fastest AGP soec available. but honestly speaking, to upgrade an entire system just to get AGP8X is a waste of time, effort and most importantly: your money.

you'll get a lot more benefit out of a geforceFX5900 or ati radeon 9800 with 128 or 256MB of memory than you ever would an AGP8X motherboard. even huge games like unreal tournament barely need 128MB of video memory. half-life 2 is certainly impressive, but will be far more CPU and GPU/VPU intensive than craving of video memory.

[harmonic01]

Ah, thanx for clearing this up Elvis. :beer:

[CgFX]

AGP memory is used by the graphics driver and applications for more than just textures. Its primary role now days is based on the fact that the GPU can access it with minimal or no CPU intervention. Even with 128MB-256MB graphics cards it is still used quite heavily and it should always be set to the max.

CPU memory bandwidth performance will have more of an impact on a modern graphics cards performance that AGP4x vs. AGP8x. However, that being said, it is yet another link in the chain and upgrading the CPU often requires upgrading the motherboard. Even if it doesn't with the cream of the crop Asus nForce2 Deluxe going for $125, I don't see any reason not to.

[elvis]

Originally posted by CgFX
Even with 128MB-256MB graphics cards it is still used quite heavily and it should always be set to the max.

in pro 3d applciations it certainly does. there's hardly any games on the market that would need a whole lot of system ram for storage. again: i wouldn't upgrade JUST for AGP8X. if the board you bought happened to have it, then that's your bonus. (it's not like i'd buy a board to AVOID AGP8X! :) )

[Aearon]

there simply is no game where agp 4x is a bottleneck

this won't change with doom3 or hl2..

as some already said it might make a difference in pro apps, i remember nvidia using the final fantasy movie as an example, they rendered it in realtime on their hardware and agp 8x made a big difference.. talk about real-world benefits ;) ;)

[harmonic01]

well, of course I won't be just plaing games. Most of my time I spend in 3dsmax anyways. It's kind of a bummer that I got this dell less than a year ago and allready thinking of an upgrade.

[kex]

just how meny processors did they have in that bitch of a machine when they hardware rendered final fantasy the movie :bounce: :bounce: :bounce:

[Aearon]

kex: imo it was just the models without fancy special effects of course, it ran on standard hardware

harmonic01: i don't know where you got the idea, but agp 8x is really no feature you'd be upgrading for, if you did buy 8x capable mobo & gfx card basically non of the speed improvent you'd get would be caused by the agp interface

8x is really just a marketing feature right now, nvidia did a really good job selling this with their gf4 8x line, pure nonsense

[elvis]

features like hardware shader support will far outweight AGP specs in the next generation of cards. massive textures aren't an issue any more as effects don't need to be pre-rendered to texture. things like cloth, materials, hair, woodgrain, metal, etc can all now be simply textured and have shader scripts to make them look more realistic. this is what you'll find in doom3 and HL2 moreso than MASSIVE textures needed some silly AGP spec.

[EvilBasterd]

Don't forget that the memory on an AGP card is also being used as a frame buffer (for building an image based on the information from poly's, shaders, etc). Most modern cards create a buffer for 3 frames. This is raw information without compression. This also hogs up a lot of memory when working on high resolutions. When using FSAA this only gets bigger, because the image is being enlarged to the size of the current resolution times the FSAA factor. So at 1024 x 768 it will be scaled up to a resolution of 8192 x 6144 (at a FSAA factor of 8)

Imagine the memory "consumption"
:eek:

[elvis]

i too can run my current games at 4X FSAA with full anisotropic filtering at 1600x1200 and complain that they are slow.

i'm not saying it's not easy to slow a card down. i'm saying that the "features" you are focussing on are secondary.

[EvilBasterd]

I was not talking about slow... I was talking about memory usage

[kex]

okz lol i wander how long final fantasy the movie took to render in maya :surprised , i dunno if it war rendered in maya but i know it was made in it.

[EvilBasterd]

It was rendered in Pixar Renderman if I remember correctly:drool:

[3Dfx_Sage]

EvilBastard- you are incorrect in your FSAA calculations. Firs of all, the only time that FSAA is truely rendered in a multiple of the base resolution is when supersampling is used, which is not what is used in almost all current hardware. Also, the framebuffer is the size of the base resolution; the colour pixels is calculated on the GPU at rendertime and not stored in a finished framebufer. Subpixels are stored temporarily, but only for the frame that is currently being rendered.

Also, if you experience performance loss using tripple buffering then it is a simple matter to switch to doubble buffering.

Also, current drivers are not even optimized to take full advantage of AGP4x.

Also, many effects do still have to be prerendered as hardly any GPU is currently able to perform shader operations fast enough to replace them entirely. However, as shaders to become longer they can take up massive ammounts of storage space and memory bandwidth if they have to be transfered to the GPU for every pixel to which they apply.

Also, the geometry information that has to be transferred over the AGP bus is beginning to reach a plateau as many GPU's are now fast enough to perform specialized shaders and bump mapping to geometry to increase the apparent complexity.

And, lastly, PCI Express 16x is right around the corner... and its FAST.

[CGTalk Moderation]

This thread has been automatically closed as it remained inactive for 12 months. If you wish to continue the discussion, please create a new thread in the appropriate forum.