Parhelia!!!

[superfly]

Originally posted by Rags

Yes, only so much memory available is correct. But the action of the software making a request for the data, sending it down to the motherboard's memory, storing it, then the video card using AGP 8X as a means to request the stored data, fetch the data, then send the information back that the operation has completed is taking a whole bunch of memory bandwidth in order for that to even be remotely useful when compared to the software sending it to the card, the card storing and manipulating the data on board. With cards coming out with 256 MB of memory on board, this shouldn't be much of an issue and those that will perform this on board with local memory will be the ones with useful performance. There is no way, with current memory tech or even AGP 8X, the performance level will be acceptable when using the local memory to store the info.(EDIT: unless it is using execution at the motherboard's memory level, and we all know how tricky it is to perform DME with certain motherboard chipsets )


Rags

I never said otherwise either,it'll always be a better solution to have the ability to store as much data as possible in local memory and AGP 8x is only fully used if system memory is fast enough to feed it(IE rambus,DDR 400 sdram,etc...),while still having enough left over for other things like physics,AI,collision detection,sound,etc,but when you run out of video card memory,the joy ride's over...


My argument was always based on the assumption that the system was already equipped with memory fast enough to feed AGP 8x,which rambus is one solution which is pretty cheap and plentifull, but i should have made that clearer, i guess....


Let's imagine a situation where any given part of a level is loaded and you're walking about in it and enjoying all the eye candy since everything is loaded onbard the video card and running very smoothly and you decide to walk towards a door that connects to another room....At some point the graphics engine has to make a visibility determination,relative to your position and load the geometry and texture data as fast as possible....AGP 8x,assuming system memory is fast enough to feed it,is usefull in a situation like that,especially if that particular scene uses lots of polys and high quality textures which could otherwise saturate a standard AGP4x bus...



Btw...did anyone try the demo with a Radeon 8500?...it should run it just fine....

[Novdid]

Originally posted by Pace
Hope these are right:
AGP 2x - 266MB/s
AGP 4x - 533MB/s
AGP 8x - 1066MB/s

AGP 1x - 266MB/s
AGP 2x - 533MB/s
AGP 4x - 1066MB/s
AGP 8x - 2133MB/s

[Wombat]

And since P4's only have 1.6-3.2GB/s of bandwidth, and EV6 at 133MHz offers 2.1GB/s, those high AGP speeds are worthless at the moment. Don't forget that your system has other things to do, like service bus requests, and keep the processor fed.

[rubank]

Wombat,
your bandwidth claims are already history, or will be in a few weeks.

Your reasoning reminds me of a well known person claiming a while back that no one will ever need more memory than 640 kB

rubank

[Novdid]

Wombat said that the bus is "worthless at the moment" that means they will be necessary some time in the future.

That well known person said that we will "never" need more than 640k.

[superfly]

I for one believe that AGP 8x can be taken advantage of if you're using a rambus based p4 system,perhaps not to the fullest capabilies of AGP 8x(2.1 gig/sec),but certainly more than AGP 4x provides,and let's not forget that p4 systems will transition to a 533 mhz bus(4.2 gig/sec)and use rdram pc 1066 within the next 3 months and by the end of the year, intel have a dual ddr sdram chipset using pc 2100 memory(4.2 gig/sec) which is way more than enough to fully use agp 8x and still have plenty left over for other things,so memory and bus speeds aren't what's delaying the intro of AGP 8x


I believe that the implementation of AGP8x has more to do with fairly tecnical issues rather that waiting for bus speeds and memory technologies fast enough to fully use it...

Agp 2x systems were released when sdram pc 66 was all the rage(intel's LX chipset),which wasn't nearly fast enough to use it and the same thing happened with Agp 4x as well...You could only use it with either pc 133 or single channel Rdram systems,again clearly not fast enough,that's why we never really noticed more than a very small difference moving from AGP 2x to 4x...

[drzaius]

I want Ant to have my babies.

[xortam]

Originally posted by drzaius
I want Ant to have my babies.

That's just too freakin' wierd!

[drzaius]

when's he due? i mean everything is speculation until we actually see the beast.

[Marshmallowman]

yeah 8x will be nice but....

128M and 256M video cards, seems to indicate most video card manufacturers will not be relying on AGP for textureing, but I am sure the 8x will come in handy just loading up the textrues and pumping those polygons, I guess at 8x the video card should be taking half the main mem data bus time, not really bandwith but more likely reducing bus contention.

The X ISle demo ran quite well on my geforce ddr, a bit jerky but still ok even at 32bit. Its a very nice engine

and i could not get it to run on an ati 8500, cannot find hardware for gen 6, and just dosn't run for Gen 5 at all...probaly ati drivers

[cjolley]

From hints we've been getting and Carmack's comment about Matrox having "solved" the memory bandwidth problem.
It doesn't sound to me like AGP speed is going to have much to do with how this new thing performs.
chuck

[superfly]

Video card bandwith issues dont really have direct relationship with AGP speed,it will only help upload data faster..Video card bandwith will help you things like AA,anisotropic filtering,etc..., at high resolutions while keeping good frame rates....


Although i believe that video card bandwith isn't the big problem it once was,we have much faster memory available now,there's HSR(hidden surface removal)tech built into the latest chips,which for now isn't even being used yet,since in the case of Nvidia's and ATI's chips,it requires developers to change their graphics engines in a way that they render any given 3d scene from front to back,so that the chips can compare Z depth buffer values for each pixel and discard those that you won't see...so we don't even see the full performance that current cards have...


Only the kyro chips do HSR without any intervention from developers...


Then there's the crossbar memory tech built into the Geforce 3 and 4 line,which regradless if like or not Nvidia,is an awsome tech to get the most of the chip since they organised a 128 bit bus into 4 seperate 32 bit buses,precisely because most of the data going to the chip are made up of 32 bit data values(32 bit color,32 bit textures,32 bit Z buffer),potentially transfering up to 8 values for every single memory clock cycle.


I Can only imagine the performance if they decide to go a 192 bit version of that,using 6 seperate 32 bit buses(one for each rendering pipeline)and use 800 mhz DDR sdram,which samsumg is curently shipping in limited quantities,that can easily make 20 gigs/sec and that's without considering the benefits of the crossbar memory tech...

[superfly]

News flash...new UT2 movie(as in unreal tournament 2...),is available and it ownz...an exellent test for parhelia...,it's 55 megs btw,but the image quality isn't very good though.


Pay close attention to the overall detail and the physics are some of the best i've seen period,with everthing behaving the way it should....

http://ina-community.com/forums/showthread.php?s=&threadid=166709

[cjolley]

Originally posted by superfly
...the physics are some of the best i've seen period,with everthing behaving the way it should....

For us Geezers, that would mean it works like Quake 1
Chuck

[superfly]

Did you see it yet???...