Xbox's NV2A, which of these features does support?

[Airwave]

I read a lot of interesting discussions on this forum about the gpu in the xbox, but there are still some things that aren't completely clear to me. For Example i was wondering which of the geforce 4 family cards features, are actually in the the NV2A. For example, does the xbox's gpu supports all of the features of the so called "lightspeed memory architecture 2" or only some of them?. And what about the accuview antialiasing? does nv2a supports it?. Speaking of antialiasing, which was the most used kind of AA in xbox games? can you make some examples of games that actually uses it?
I thank already, everyone who'll contribute to this topic , i just hope it doesn't get snobbed, only because it's about an older consolle :smile:.

BTW i apologize for my occasionaly raw english, since it's not my mother language (i'm from Italy)

 

[swaaye]

The only game on Xbox that uses AA, that I know of, is apparently Blood Wake. I'm sure Xbox has the GeForce 3 AA features (I've read that here before), but I also know it has memory bandwidth issues and AA eats up bandwidth among other things. AA is neglected on the modern consoles too unfortunately.

For more hardware details, this is a good article.
http://www.extremetech.com/article2/0,2845,1152647,00.asp

As for the Xbox memory interface, it runs at 200MHz (400MHz effective clock rate since it's talking to DDR memory), which puts its peak data rate at 6.4GB/sec. The XGPU uses a four-node memory crossbar that's been modified to support UMA and CPU accesses, and the XGPU incorporates a Pentium-III Front Side Bus (FSB) to connect to the CPU. According to nVidia, this modified memory crossbar supports many memory "clients" (texture, color, z, etc.), as well as supporting a high-priority client to service CPU memory traffic. nVidia also stated that the crossbar memory controller is more than capable of fully saturating the P-III's 133MHz front side bus interface.

 

[Urian]

If I am not wrong NV2A is a NV20 (GeForce3) with an extra Vertex Shader, it isn´t fully considered a NV25 since it lacks some of the functions included in GeForce4.

 

[AlNom]

PGR2 used 2x. Unreal Championship pre-patch used AA as well, but I can't remember what level (probably 2x).

There are some NDA'd extra features that made its way into the FX, but I'm not sure which... There's very little info for the public there, but that's what I recall from along time ago.

 

[Airwave]

yeah the nv2a seems like one of the least documented and most "mysterious" gpu ever, that's why i decided to post here, in the hope of knowing more. So it seems that accuview is definetely not supported, and the fact that FSAA looks worse than on geforce 4, and that it takes a much stronger hit from enabling it, seems to strenghten this point don't you think? . So if i'm not wrong, the AA feature supported by the nv2a are the classic multisampling (2x, 4x) and the quincunx, right? . Swayee thanks for posting that article too, i'll be eager to read that as soon as i can .

On another site, i spotted, a user that claims that the xbox actually utilizes all of the features of the LSMA2, minus some that however were not essential on the xbox since it has a shared memory system. I have that page saved in the favourites , when i'll be back home i'll post it, so that we can discuss that.

So about the most used kind of AA seems to be the multisampling 2x, i tought, it was the quincunx, for i read that 2x just wasn't worth the performance hit, since on regular crt tv sets, the advantage of such level of antialiasing were barely appreciable. Quincunx on the other side is considered a much better choice, since it offers near 4x quality with a performance hit nearer to 2x levels. Please Can you make me other examples of xbox games using FSAA? does doom 3 uses any kind of AA?

 

[sebbbi]

...FSAA looks worse than on geforce 4, and that it takes a much stronger hit from enabling it, seems to strenghten this point don't you think?

...minus some that however were not essential on the xbox since it has a shared memory system.

Without knowing any of the dirty details, my prediction is that the Xbox unified memory architecture was the biggest reason why the antialiasing was not used that much in games. Enabling 2xMSAA doubles the back buffer bandwidth requirement during rendering, blending and during the resolve operation (copy MSAA back buffer to front buffer or another texture). And 4xMSAA quadruples the BW requirements.

In comparison to the newer consoles: PS3 graphics chip has it's own DDR graphics memory (much higher available graphics memory BW), and Xbox 360 graphics chip has it's own EDRAM (to cover all backbuffer/blending/AA BW requirements).

 

[Airwave]

Without knowing any of the dirty details, my prediction is that the Xbox unified memory architecture was the biggest reason why the antialiasing was not used that much in games. Enabling 2xMSAA doubles the back buffer bandwidth requirement during rendering, blending and during the resolve operation (copy MSAA back buffer to front buffer or another texture). And 4xMSAA quadruples the BW requirements.

In comparison to the newer consoles: PS3 graphics chip has it's own DDR graphics memory (much higher available graphics memory BW), and Xbox 360 graphics chip has it's own EDRAM (to cover all backbuffer/blending/AA BW requirements).

I want to better understand this point, In which way did the unified arhitechture, prevented the use of antialiasing? how does divided banks of memory, would have helped instead? do you think that support to accuview, would have helped in that regard too (making the ipact of AA on performance less dramatic)?

 

[Shifty Geezer]

It's not so much unified RAM versus split RAM. It's really just a matter of available bandwidth. If you have a choice between 100 GB/s to a unified RAM pool shared between CPU and GPU, or 5 GB/s GPU-only VRAM, the former will be better. XB just lacked overall bandwidth for the graphics chip, especially when the CPU was consuming some of that bandwidth.

 

[Airwave]

It's not so much unified RAM versus split RAM. It's really just a matter of available bandwidth. If you have a choice between 100 GB/s to a unified RAM pool shared between CPU and GPU, or 5 GB/s GPU-only VRAM, the former will be better. XB just lacked overall bandwidth for the graphics chip, especially when the CPU was consuming some of that bandwidth.

ok i understand. However I tought that xbox was not that bandwidth limited, cause of the use of lower resolutions and trilinear filtering. seems like i was wrong. However, did the accuview antialiasing would have made the use of antialiasing more easy, and with less performance impact?

 

[brain_stew]

If I am not wrong NV2A is a NV20 (GeForce3) with an extra Vertex Shader, it isn´t fully considered a NV25 since it lacks some of the functions included in GeForce4.

That was always what I was led to believe. A GF3/GF4 hybrid if you will.

I want to better understand this point, In which way did the unified arhitechture, prevented the use of antialiasing? how does divided banks of memory, would have helped instead? do you think that support to accuview, would have helped in that regard too (making the ipact of AA on performance less dramatic)?

I'm far from an expert, but I believe he was simply making the point that since that bandwidth had to be shared between the GPU and CPU, the "effective" bandwidth for the GPU is much lower. You go ahead and saturate that pipe by enabling msaa and you could end up starving the CPU. If the Celeron had its own memory pool and the NV2A had that 6.4GB/s of memory bandwidth all to itself, enabling AA would have been much more feasible.

Edit: Oops already answered, sorry.

 

[Airwave]

That was always what I was led to believe. A GF3/GF4 hybrid if you will.




I'm far from an expert, but I believe he was simply making the point that since that bandwidth had to be shared between the GPU and CPU, the "effective" bandwidth for the GPU is much lower. You go ahead and saturate that pipe by enabling msaa and you could end up starving the CPU. If the Celeron had its own memory pool and the NV2A had that 6.4GB/s of memory bandwidth all to itself, enabling AA would have been much more feasible.

Edit: Oops already answered, sorry.

thanks for pointing out anyway , so it appears that altough the low resolutions, and simplier texture filtering used, the bandwith was not enough to warrant a "serene" use of Antialiasing.

 

[AlNom]

To put it in persepective, the GF3-Ti200 had just as much bandwidth for just the graphics alone (texturing, geometry, and framebuffer), and this card was already bandwidth limited. Now that 6.4GB/s is shared with the entire system...

 

[Airwave]

To put it in persepective, the GF3-Ti200 had just as much bandwidth for just the graphics alone (texturing, geometry, and framebuffer), and this card was already bandwidth limited. Now that 6.4GB/s is shared with the entire system...

i see, but i wondered if the xbox used a better memory management system (the features of the lightspeed memory architechture 2), so that it wasn't as bandwidth limited as the geforce 3 cards.

 

[AlNom]

I would be inclined to believe quite a bit of the work done for the NV2A made it's way into the Gf4. That said, the bandwidth is going to be terrible when alpha blending anyway. They'd need several times the bandwidth just to be safe, and that won't be solved in some magical component of the chip logic.

 

[Airwave]

I would be inclined to believe quite a bit of the work done for the NV2A made it's way into the Gf4. That said, the bandwidth is going to be terrible when alpha blending anyway. They'd need several times the bandwidth just to be safe, and that won't be solved in some magical component of the chip logic.

so you think that, nv2a actually supports the same bandwidth saving features (Light speed memory architechture 2), of the geforce 4? this is matter is of great interest to me, since, i found a lot of contraddictory informations over the internet. As far as i have understood, it seems that Fast z clear, crossbar memory controller and Lossless Z-buffer compression are there, but what about , the Quad Cache memory caching subsystem, auto pre charge and visibility subsystem?

 

[Airwave]

http://forum.teamxbox.com/showthread.php?t=198275&page=2

here is the webpage of the forum, i was talking about before, according to this user (X-2001), nv2a is even more similar to nv25 cards that I tought. He states that nv2a supports all of the LMA 2, except for the vertex cache (that according to him, is not useful since xbox uses a shared memory architechture), and maybe the auto pre charge. On Nvidia's site, i read a pdf that confirms that at least fast z clear, visibility subsystem, crossbar memory controller, and loseless z buffer compression were already supported by the original geforce 3, no reason to doubt they are present on nv2a too Maybe, on the geforce 4, these features are more twaked/advanced than on nv2a or they are the same? Another thing that he states, is that nv2a supports the 4xS FSAA, that i tought introduced with the geforce 4. What do you think about?

 

[brain_stew]

While we're on the subject of the NV2A have any of the pixel counters had a look at the bunch of "720p" Xbox titles? How many of these titles actually rendered natively at the full 1280x720 resolution (I'd wager not many) and how many just relied on scaling trickery like we see in a bunch of titles today?

 

[Neb]

On Nvidia's site, i read a pdf that confirms that at least fast z clear, visibility subsystem, crossbar memory controller, and loseless z buffer compression were already supported by the original geforce 3, no reason to doubt they are present on nv2a too Maybe, on the geforce 4, these features are more twaked/advanced than on nv2a or they are the same? Another thing that he states, is that nv2a supports the 4xS FSAA, that i tought introduced with the geforce 4. What do you think about?[/COLOR]

IIRC Geforce4 had improved fast z clear, loseless z buffer compression etc over Geforce 3. I assume they refined other parts to to improve AA, texturing perfomance etc. Maybe some of that stuff got into nv2a or perhaps not considering when the chip was done vs release dates and preparation. Also factor in the Geforce 4 had upto 10.4GB/sec VRAM throughoutput. Geforce3 ranging from ti200 (6.4GB/sec), regular (7.36GB/sec) and ti500 (8GB/sec)

Another question would be how efficient Xboxs memory system was as I remembered it to be 2x3.2GB/sec. Nforce2 chipset also sported this on PC although average never really reached anywhere close to maximum.

About FSAA Geforce2 could also do it. I actually played Kingpin FPS at 800x600 with 2xFSAA back in the days on my Geforce 2 GTS. Dont think they removed this possibility on Geforce 3 or 4. Maybe from control panel but not in drivers.

 

[Airwave]

Regarding FSAA, i have no doubt that it is in nv2a since it has been supported by older geforce cards too (as also you were noting). My question is if nv2a supports, the new kind of antliasing called 4xS, that is supposed to be first introduced in the geforce 4 family cards. Since I read developers lamenting the fact that FSAA didin’t look as good and wasn’t as efficient as on geforce 4, I suppose that it is not supported by nv2a, but it’s only my humble ide, that I can’t prove. To make more clear what I’m talking about, I’ll quote an extract, from another nvidia pdf that I read.
4x and 4XS Modes: Higher Level of Texture Quality
4XS mode is a new high-quality mode that delivers improved subpixel coverage and
a higher level of texture quality.
When the final color of a pixel is determined, all of the subpixels that can contribute
to the final pixel color are tested. They are either .in. or .out. (that is, they
contribute by being part of the object and that object covers the subpixel, or they
don.t). The final pixel color is then reconstructed by implementing a weighted
summation of all of the subpixels. 4XS mode delivers 50 percent more subpixel
coverage than previous modes. This translates into a finer gradation in final pixel
color values at the edge of an object, resulting in a smoother antialiased edge.
Additionally, 4XS mode delivers a higher-quality image by delivering more texture
samples per pixel.

 

[AlNom]

4xS is just a mixture of 2x super sampling and 2x multisampling. The higher level of texture quality is due to the second texture sample whilst sub-pixel coverage is due to 2 multisamples being calculated per texture sample (4 sub-samples total).

So yes, NV2A probably supported it, but would have been far from being able to use it due to severe bandwidth and fillrate limitations (unless the game were extremely simple).