Parhelia doesn't need Occlusion Culling....

[Electric Amish]

Depth-Adaptive Tessellation is an advanced tessellation scheme that tessellates meshes using multiple levels of detail (LODs) to maximize the geometry detail of a 3D scene while maintaining high levels of performance. By using LOD-based tessellation, the graphics processor avoids unnecessarily processing triangles that would otherwise not contribute significantly to the visual quality of the final rendered image.
Lower levels of detail are acceptable when the object being rendered is further back in the scene. Because it appears smaller, it is rendered
using a lower number of screen pixels. In fact, depending on the distance, increasing the number of triangles beyond a certain point may have
little or no effect on an object's appearance. The ability to reduce the LOD for distant objects provides considerable savings on
transformation, lighting, setup and rasterization, leaving a higher triangle budget for objects that are up-close.
LODs can also be applied to an object whose mesh spans a significant portion of the scene's depth on the display. A good example of such
an object is a terrain. For such objects, Depth-Adaptive Tessellation ensures that no cracks are seen on the seams where changes in LODs occur.

If some people would read the white-paper on Hardware Displacement Mapping they would see the section about Depth-Adaptive Tesselation. (Page 6)

They have a scene rendered with and without Displacement Mapping/Depth-Adaptive tesselation.

with= 17, 794 triangles

without= 165,150 triangles.

amish

 

[Adul]

I don't think some people actually read these articles. Even though it may pass by their eyes, I question how their brain perceives it. And I would not go has far has to say it doesn't need it, but it is certainly better off then the Geforce 2 was when it comes to saving some bandwidth. This card will be very competitive for a while and i see no reason why it shouldn't be . now if I can only afford it. I would love the at least dual monitor option

 

[Electric Amish]

Unfortunately our fearless leader didn't correlate the Occlusion Culling and the Depth Adaptive Tesselation....

amish

 

[kuk]

Originally posted by: Electric Amish
Unfortunately our fearless leader didn't correlate the Occlusion Culling and the Depth Adaptive Tesselation....

amish

That means: ANAND, wake the hell up!


Has any reviewer picked up this little piece of info?

 

[NFS4]

Originally posted by: Electric Amish
Unfortunately our fearless leader didn't correlate the Occlusion Culling and the Depth Adaptive Tesselation....

amish

This is something totally different AFAIK. This is NOT Occlusion Culling or anyway related to it IMHO. Anand DID talk about DAT, but what this does is reduce the number of triangles in a given scene. ATi and NVIDIA don't render them at all. This if from Anand's article:

This concept is very similar to mip-mapping when it comes to textures but simply applied to displacement maps instead. Matrox has licensed this technology to Microsoft for use in DX9 and you will definitely see other vendors implement similar functions into future GPUs.

A major benefit of HDM is that using technologies such as Depth-Adaptive Tessellation you can produce a very detailed terrain using a low polygon count base mesh and a very small displacement map (multiple KBs in size). This saves traffic across the memory and AGP buses while allowing for extremely detailed scenes to be produced.

Similar to mip-mapping. When I think of mip-mapping, I don't think Occlusion Culling.

This technology is about lowering or raising detail levels of scenes. Occlusion Culling is about removing unneeded ploygons all together. Big difference IMHO.

Also, I'd be more inclined to go with Anand's interpretation of the facts instead of yours...no offense

 

[Pariah]

This topic is certainly not deserving of a new thread.

Lack of any bandwidth saving measures will certainly not be the death of this card just as lack of T&L did not kill the V5 like all the doomsayers predicted. Beyond the basics there are no "have to have" features for top notch performance depending on how the rest of the card is designed. The Parhelia will do fine just as the V5 did because it appears it will be able to brute force it way to top notch performance. There is no game available today that will stress this card with everything turned on, heck, there isn't anything available that stresses the Radeon 8500 of Geforce 4 either. 2 years from now lack of this feature may catch up to it, like lack of T&L has recently hurt the V5, but that is well down the road and by the time that rolls around, we should be done with the Parhelia Max and Parhelia 550 and be looking at a new generation from Matrox.

 

[Anand Lal Shimpi]

HDM and Z-Occlusion Culling are two separate things. The Matrox whitepaper on HDM is talking about the polygon savings when using depth-adaptive tessellation vs. the same scene with a higher tessellation factor. Occlusion culling is a different technology that allows occluded pixels (those not visible to the user) to be discarded before they are textured.

Take care,
Anand

 

[Pocatello]

I hope the Parhelia will bring better fortune for Matrox than the voodoo 5 to 3dfx.

 

[Czar]

could someone clear this for me

Lets say we have a scene that has enough triangle count to push the parhelia to the limit, that is without hardware tesselation, will the card reduce triangles far away from the original mesh? that is where the knife is stuck in the cow (nevermind, icelandic expression)

 

[DeathByDuke]

You know that technique, Depth-Adaptive Tesselation, just why the hell did ATi and nVidia even bother creating Hidden Surface Removal? it just reminds me of when the G400 came out, was equal to GF256 on speed but had features that weren't available from nVidia til' GF3 (!) not to mention Radeon 1.... that's almost three generations of cards until support for what Matrox had came into being. Hell, it's a crazy world!

 

[DeathByDuke]

Come to think of it, Unreal Tournament kinda of did this in software... if you see the Unreal logo on a wall or a trophy, go as far away from it and you'll notice a drop in polygons in the models, get close and the polygons are visibly 're-filling' the model! Sometimes when you did the 'playersonly' command and went moving around, there was a kind of lag in level of detail on the frozen bots, i would get close and they'd look like weeds, (!!) then suddenly inflate! this helped my Voodoo2 on the demo a lot...

 

[NFS4]

You know that technique, Depth-Adaptive Tesselation, just why the hell did ATi and nVidia even bother creating Hidden Surface Removal?

They are two different techniques aimed at tackling two different scenarios. They are not competing technolgies so I'm not quite sure I see your point.

 

[kazeakuma]

Hrmmm.

LOD, HDM and Occlusion culling are all different things. UT has LOD, most games do, the further away an object is, the lower the detail. This has been in games for years. HDM (DAT actually) is a technique to create high poly meshes, very easily. A neat feature of DAT is the ability to raise/lower the LOD depending on distance. Occlusion culling (which has been explained by 5 people already but I'll say it again) is a technique to remove pixels not visible to the user. Thus saving on rendering time. While the Parhelia doesn't need OC, it's because it's supposed to overcome it with sheer brute force. DAT is something else entirely Amish.

 

[Pocatello]

I guess nvidia isn't the only one who likes brute force .

 

[FishTankX]

Yes, well, the place where it will need bandwidth saving techniques is when it has huge amounts of overdraw and pixel shading calculations. You see, Electric Amish, Memory bandwidth isn't the only thing that occlusion culling thing saves. You also must remember that it also saves [memory bandwidth that would otherwise be used to look up textures, this the main limitation in] Fillrate. And Pixel shader operations. What good is enough bandwidth if you're wasting bandwidth on texture reads and pixel shader operations?

And don't forget the pixel shaders, why shade what won't be seen? it's just plainly a huge waste. Mix tons of overdraw with pixel shader effects galore and you have a nightmare for any card not armed with occlusion culling. It's not just about memory bandwidth, man..

Shading all those pixels takes up computation time in the pixel shader pipes, right? Waste

It also wastes memory bandwidth (what you were talking about)

It also wastes fillrate itself. This will become aparent in high resoultions where the pixel/vertex shaders will get much more use.

Unfortunately our fearless leader didn't correlate the Occlusion Culling and the Depth Adaptive Tesselation....

Uh.. lemme see.. it's like this. He didn't corelate it to occlusion culling because it isn't. It does nothing to save on pixel shader operations or fill rate. Does it? Correct me if i'm wrong

-Fishtank

 

[FishTankX]

Czar

Adaptive tesselation will reduce traffic on the AGP bus. Thus, if the Parahelia was pushed to the limit it would drastically reduce the polygon count flying over the crowded 1.06GB/s AGP pipe or from the grahpics memory to the chip. Thus, it's more about saving bandwidth on a slower connection to AGP and to keep form having to pull all those polygons over from main memory onto the chip. . If it was bad enough to stress the AGP bus and graphics memory to the limit, Adapative Tesselation would definatley help. Very much so. Because instead of having to retrieve a load from polygons from memory, you would just retrieve a displacment map and a basic mesh and use something like Truform to put detail where it's needed and leave it out where it's not needed. Transfering a low detali model and a map of where detail is suposed to be added is alot easier than trying to pull a high detail model (think 10X bigger) over the memory bus and eases the stress on memory systems in general .

 

[AnoTech]

Originally posted by: FishTankX
Yes, well, the place where it will need bandwidth saving techniques is when it has huge amounts of overdraw and pixel shading calculations. You see, Electric Amish, Memory bandwidth isn't the only thing that occlusion culling thing saves. You also must remember that it also saves [memory bandwidth that would otherwise be used to look up textures, this the main limitation in] Fillrate. And Pixel shader operations. What good is enough bandwidth if you're wasting bandwidth on texture reads and pixel shader operations?

And don't forget the pixel shaders, why shade what won't be seen? it's just plainly a huge waste. Mix tons of overdraw with pixel shader effects galore and you have a nightmare for any card not armed with occlusion culling. It's not just about memory bandwidth, man..

Shading all those pixels takes up computation time in the pixel shader pipes, right? Waste

It also wastes memory bandwidth (what you were talking about)

It also wastes fillrate itself. This will become aparent in high resoultions where the pixel/vertex shaders will get much more use.

Unfortunately our fearless leader didn't correlate the Occlusion Culling and the Depth Adaptive Tesselation....

Uh.. lemme see.. it's like this. He didn't corelate it to occlusion culling because it isn't. It does nothing to save on pixel shader operations or fill rate. Does it? Correct me if i'm wrong

-Fishtank

I believe a correlation between performance hits and increases for both technologies in comparison to a GPU with just an advanced OC would be nice. For instance:
Will the benifits of DAT be enough to overlook all those extra pixels being processed? Personally I don't think so. But we have to keep in mind future games also. Will the user be able to notice? I don't believe so, but it will be very interesting to see some statistics.


I think the most important thing to consider for buyers on the prowl for a new video card is DX9. Personally I don't want to have to upgrade sooner than 2 years, unless it's a major change in technology.

Also, maybe someone can clear this up: with Matrox not being able to handle 32-bit floating integers, does this mean it will not be DX9 compliant at all or will it just run poorly?
I see that the Vertex Shaders are claiming to be compliant?
Also, a .15 micro die size.

I guess what I really want clarification on, is how much of an effect will the larger die actually have on DX9 games?

 

[AnoTech]

Perhaps a better word choice would be "comparison" instead of corelation.

 

[Czar]

I personaly see only one problem with this card, its 0.15 micron when it seems to be designed to be 0.13 and then they just removed bits and peices so it would work as a 0.15 micron card

 

[PG]

Originally posted by: DeathByDuke
it just reminds me of when the G400 came out, was equal to GF256 on speed but had features that weren't available from nVidia til' GF3 (!) not to mention Radeon 1....

Wow, do you really believe that a G400 equals a GeForce256 in speed? :Q
...and had more features? :Q :Q

just remember to pass the crack pipe and share

 

[imgod2u]

Correct me if I'm wrong, but isn't DAT similar to ATI's TRUFORM in that it doesn't really have anything to do with the actual rendered polygons but rather, it's taking current polygons and tenselating it to create more detail. ATI's TRUFORM does this with all polygons rendered (it has to be turned on though). So I don't know what this has to do with occlusion culling with polygons that are used by the game engine.

 

[Nemesis77]

Originally posted by: Czar
that is where the knife is stuck in the cow (nevermind, icelandic expression)

Then that cow is in the ditch (a finnish expression ).

 

[Nemesis77]

Originally posted by: PG
Wow, do you really believe that a G400 equals a GeForce256 in speed? :Q
...and had more features? :Q :Q

just remember to pass the crack pipe and share

Dunno about performance but... When did GeForce get EMBM?

 

[Sunner]

Granted the GF256 didn't have EMBM, but OTOH the GF had many other features that the G400 didn't, so in the end, I'd say the GF definately wins in terms of 3D features.

Oh well, I'll probabaly pick up a 128 MB Parhelia if it performs good, and keeps up Matrox's tradition of ultimate 2D quality, which it looks like it definately will.

 

[HardWareXpert]

Originally posted by: Nemesis77

Originally posted by: PG
Wow, do you really believe that a G400 equals a GeForce256 in speed? :Q
...and had more features? :Q :Q

just remember to pass the crack pipe and share

Dunno about performance but... When did GeForce get EMBM?

GF3 upwards supported EMBM, not that it was a big deal because hardly any games supported it.