AGP question

hi there,
I am a completly new to overclocking, but am keen to build my own machine and start experimenting
I've been educating myself a bit but have some queries, hope you can help me out a little.

I understand that the AGP bus is designed to run at 66mhz, and that it is 32 Bit, and thus 4 (byte) X 66mhz = 133 mb/s. Taking the latest AGPx8 factor, I arrive at 2133 mb/s effective AGP bandwidth (please correct me if I am wrong).

Looking at the very latest graphic cards, the memory bandwith they can achieve is well in excess of this (up to 2.7 Gb/s). As my limited understanding goes, their is a major latency (right terminology?) factor here - or am I missing a sizable chunk of the puzzle?

Any comments much welcomed.

Well, first of all 4 x 66 = 264, not 133. (Actually it's closer to 266 MB becuase the bus runs at 66.6666... Mhz )

But 8x AGP is about 2.1 GBps. If someone is claiming more across the AGP bus there could be a number of reasons, such as compression.

Also, I believe that the memory bandwith stated is a direct card memory -> GPU bandwith. Once you get the data over the AGP bus, the bus on the card should take over and move that data like no tomorrow. And since you don't really have to move that much data in one chunk (first time to fill up memory with textures and such), the AGP -> card bottleneck shouldn't really do much.

But I'm not up to date with video cards so I could be wrong...
JigPu

Not sure why I said 4 x 66 = 133?! Thanks for correction tho.
I suppose what im trying to do is get a grasp on the agp system, rather than establish maximum b/w over agp bus. Graphic card specifications can be quite confusing to a noob, when I fork out in excess of £200 for a card I wanna be pretty sure that not only is it fast, but that It will complement my system and bring the best performance in it. I've looked at the latest nVidia and ati offerings and they seem to be pushing their huge memory bandwidth as a selling point, but as far as I can understand, these factors are limited by the agp bus. Think I need to some more investigating.

Looking at the very latest graphic cards, the memory bandwith they can achieve is well in excess of this (up to 2.7 Gb/s). As my limited understanding goes, their is a major latency (right terminology?) factor here - or am I missing a sizable chunk of the puzzle?

Click to expand...

the 5900Ultra has 27GB/s not 2.7GB/s (I think thats what you were referring to there...) but yeah, it is the GPU -> memory bandwidth

Ah, me thinks I understand now. I've read alot of information on overclocking recently, and came across some info on increasing the agp bus speed (I understand it can be determined through the bios or jumpers on a mobo, in relation to the fsb). I was interested to know if upping the agp bus speed has any performance benefits to benchmarking. I know that this can cause instability or lock ups, but I see that system performance is generally improved from upping the fsb (amongst other things) and just wondered if this applies to the agp bus speed?

it actually does very little and i personally dont think its worth the potential problems it could cause.

got me head round it now, tnx for the help.