The Game Technology discussion thread *Read first post before posting*

[ihamoitc2005]

When you say CPU what are you referring? Physics calculations? There are other games that have more than 8 cars on 360, Need for Speed Shift for example supports up to 16 cars all of them starting the race at the same time, but Need for Speed is 30fps so I guess is easier to have more cars, but my question is what does that have to do with the CPU.

Other developers on this forum have said that cpu load of each car (physics/AI) is very low and beyond certain degree, marginal cost is lower (even for rendering because you either can't see car or it's far away for mnimum LOD), so I do not think CPU s the issue. Even PS2/Xbox had game with 20+ cars. I feel that therefore this is mostly a rendering issue because sometimes the faraway cars may be nearby and must have higher LOD.

 

[Neb]

Plus, Digital Foundry writes that GT5 Prologue has more sophisticated (looking) shaders and more effects at 50% more resolution with 2xAA with much more over-draw situations (interior rendering, smoke, dust).

Maybe even stones have blood.

But my friend your highlighting superior stuff in one game while neglecting superior things in the other game to drive home your point. How about Forza 3 having clearly more detailed vegetation and tracks or how cars around you are reflected on the car body. How about 'soft flesh' rendering for the tires to have them flex. Forza 3 has tiremarks that stack up on the track how does it go for GT5p?

How will this affect perfomance in total?

Also interiors are rendered in Forza 3 as already showcased by a forum member. Another question is how you can conclude GT5p has more over-draw situations as in smoke dust?

I mean just becouse one car can create a decent amount of smoke particles doesn't mean all cars can do it at the same time. There is a particle count cap for each effect. If exceeded the oldest particles are removed. The question would be what is the cap for each game?

And... sorry but painting a stone red doesn't mean it has blood.

 

[ihamoitc2005]

But my friend your highlighting superior stuff in one game while neglecting superior things in the other game to drive home your point.

I have no point my friend, only to respond directly to Joker454's earlier comments for which I had disagreement.

Primary statements for which I had disagreement: He claims low-res buffer (smoke/dust) and very little transparency in GT5 Prologue compared with Forza 3. He used the term predictable over-draw to imply Forza 3 has high-res buffer and many transparency due to hardware difference between 360 and PS3. I disagree because regardless of theoretical ideas, actual games shows opposite. GT5P has much better smoke/dust (Forza 3 has almost none) and much more transparency. Digital Foundry also agrees with me on this.

I provided videos for the evidence. If you watch the videos you can also see that GT5P has great transparency and also great smoke/dust.

Secondary subjects for which i had discussion (but not exactly disagreement): Vertex vs CPU bottleneck for missing detail in Forza 3, studio vs studio for shader efficiency, etc...

If you read from the beginning of joker454/Ihamoitc2005 it will all make sense.

 

[Alucardx23]

But my friend your highlighting superior stuff in one game while neglecting superior things in the other game to drive home your point. How about Forza 3 having clearly more detailed vegetation and tracks or how cars around you are reflected on the car body. How about 'soft flesh' rendering for the tires to have them flex. Forza 3 has tiremarks that stack up on the track how does it go for GT5p?

That’s one of the things that surprise me the most about Forza 3, some of the best reflections that I have seen this generation, GT5 prologue reflections are low res, have a lot of pop up and don’t reflect other cars. Can Forza 3 decision of reflecting the other cars affect the number of cars on screen?

Edit: Another thing is that on GT5 prologue the rear view mirrors show the cars at a lower level of detail, it not as bad on the center mirror but real bad on the side mirror. This does not happen on Forza 3, the only difference I could find is that the rear view mirror uses a lower resolution, but even that is not as low as GT5.

 

[betan]

I
Pedestrians are easy, been there done that.

Are you comparing a racing game pedestrians near road to MLB 2K7 spectators?
I'm almost sure you are not.

 

[joker454]

He claims low-res buffer (smoke/dust) and very little transparency in GT5 Prologue compared with Forza 3.

I'm not saying GT5 has less transparency than Forza 3, I'm saying they both have very little of it. I've only played the Forza demo though.

Are you comparing a racing game pedestrians near road to MLB 2K7 spectators?

No, we had to draw 30k+ of them at 60fps. But I know enough about drawing spectators to know that it's total cake to do.

When you say CPU what are you referring? Physics calculations? There are other games that have more than 8 cars on 360, Need for Speed Shift for example supports up to 16 cars all of them starting the race at the same time, but Need for Speed is 30fps so I guess is easier to have more cars, but my question is what does that have to do with the CPU.

Well I figure they must have targeted 16 cars at some point since it's not like it was a secret that GT was gonna have it. But they ended up with 8. So what would 8 extra cars have added? It's not pixel load. The cars are mostly opaque, so while they add pixel load, they also remove pixel load of whatever is behind them. The do add vertex load. However the game already tiles with 2xmsaa, which means it's already capable of handling a potentially very high vert load. If it was just a vertex issue, then why not just switch to variable msaa, like perhaps dropping msaa/tiling when many high lod cars are close by? And as you said other games like NFS already draw 16 cars.

So what about cpu? Forza is more sim, so presumably it's physics are more involved than Grid or NFS. Plus they run their simulation at a high rate, 360fps was it? On top of that presumably the AI and collision cpu load would be increased with 8 more cars. So I figure maybe it was a cpu thing. Just a guess though.

 

[Crossbar]

Can Forza 3 decision of reflecting the other cars affect the number of cars on screen?

Hmm, would love to hear the discussions leading up to that decision.
-Should we have car reflections or allow more cars in races?
-I think car reflections are cool.
-Car reflections it is then.

On a serious note @Joker. Do you think PD or other developers have used the "small tiles" technique suggested by nAo?

I believe edram makes less sense this generation compared to the previous one.
Most of the time we need to render slow opaque pixels and fast relatively simple transparent pixels.
The first case can be managed by a standard GPU with an external memory pool, while the second case can be addressed with a small (ie 64x64) on chip tile cache, if we use our powerful CPUs to tile all our transparent geometry and to submit geometry in tile order.
It would make perfectly sense on a console..

further clarified here:

If they didn't have EDRAM they'd need a 256bit bus or two bus' to make any sort of high def rendering practical.

I'm not sure I agree with this..
Opaque pixels are likely to be processed with long shaders, hence most of the available per pixel bandwidth would be used to fetch textures, not color + z.

Tiling is interesting, but I've yet to see real evidence that it actually reduces transistor counts at a given performance level.

Just to make it clear, I was not advocating any kind of tiled rendering GPU, I was thinking that a small on chip tile cache would use much less transistors than n MBytes of edram.
I'd like to use those transistors for more ALUs..
I also believe that next gen consoles CPUs might be very good at tiling geometry
Basicly one would render all the opaque pixels as in current GPUs (that tile cache I'm talking about would not be even used in this case) and all the transparent pixels in tile order.
A relatively cheap way to achieve near-to-theoretical fillrate without devoting tons of transistors to edram.

I think it's way to early to be declaring MS's or Sony's graphics chip choices bad or good, we have to wait and see.

I'm not advocating Xenos nor RSX, I was just expressing an opinion on a hypothetical GPU

Marco

Even further clarified here:

Emh..I'm a late to this discussion but I'll try to re-state my point anyway.
I was not dissing Xenos, nor pimping RSX. I was just expressing an opinion regarding edram uselfuness this generation.
Main idea is that we need to render slow complex pixels (long vertex and pixel shaders) and very fast transparent pixels (particles..trees, etc..)
In the first case edram is not needed cause we are going to ouput 1 or even less pixels per clock cycle anyway, a 'slow' externa bus (even a 128 bit bus) is ok.
In the second case edram is a big win, but it takes a lof of on die space.. so I thought that would be nice to have a small (64x64?) on chip tile cache for frame buffer and zbuffer and use the CPU to tile geometry and then submit transparent geometry in tile order.
It would way cheaper than edram and it will give us a close to thoeretical fill rate.

Do you think the technique he suggests is viable in reality (using the built-in cache of the RSX) and have it been implemented in any games yet. In that case would GT5 be a game that would benefit from it?

As I understand the technique could also help the MSAA pass by easing the bandwidth utilisation of external memory during the render pass and thereby speeding up the render pass, is that right?

 

[ihamoitc2005]

I'm not saying GT5 has less transparency than Forza 3, I'm saying they both have very little of it. I've only played the Forza demo though.

You said that both are predictable over-draw and that unlike PS3, Xbox360 won't be over-draw limited which is the same as saying Forza 3 has more over-draw.

I'm saying the opposite. GT5 Prologue has a lot more over-draw than Forza 3. In fact, you can fill the screen with clouds of smoke and dust at any time and also windows have great clarity. You can even see the interior rendered with lighting and shadow of rearwards cars in the rear view mirror. It is quite amazing.

Forza 3 you cannot see these things. That is why I am saying exactly opposite of your original comment. You say they have predictable over-draw (low in GT5, high in 360), I am saying they have opposite .... "unpredictable" overdraw (high in GT5, low in 360).

No, we had to draw 30k+ of them at 60fps. But I know enough about drawing spectators to know that it's total cake to do.

I am quite sure that if Forza 3 could have more spectators it would. Instead they cut back spectator count from Forza 2. A detail reduction. Maybe for your game it is easy because of slow and also predictable LOD transitions but for fast moving racing games where camera is 100% player control I am sure this is very different.

Well I figure they must have targeted 16 cars at some point since it's not like it was a secret that GT was gonna have it. But they ended up with 8. So what would 8 extra cars have added? It's not pixel load. The cars are mostly opaque, so while they add pixel load, they also remove pixel load of whatever is behind them. The do add vertex load. However the game already tiles with 2xmsaa, which means it's already capable of handling a potentially very high vert load. .

I would not make that conclusion my friend.

1) What is behind the cars is the road (mostly). Road has much less pixel load than the car, especially if there is transparent surfaces for windows, lights, etc.... More cars means less road which means net pixel load is much more.

2) Since Forza 3 has 2xMSAA then it has 2 tiles. For reasons you explained earlier (cars straddling tiles) once you have tiling, polygon increase becomes very expensive. When you have more cars and cars are always moving around on the screen, the chance of too many cars being in both tiles is a very high risk. If those cars also have too many polygons then it is a very big danger to performance.

If it was just a vertex issue, then why not just switch to variable msaa, like perhaps dropping msaa/tiling when many high lod cars are close by?

Variable MSAA for this game means high chance of no AA because with 16 cars, often there will be many cars nearby, especially if you have slow cars that move like herds. 720P with 2xMSAA is "ok" but 720P with no AA is bad for a racing game.

And as you said other games like NFS already draw 16 cars

NFS is 30fps. 2 x CPU & GPU cycles to process everything. Easy.

So what about cpu? Forza is more sim, so presumably it's physics are more involved than Grid or NFS. Plus they run their simulation at a high rate, 360fps was it? On top of that presumably the AI and collision cpu load would be increased with 8 more cars. So I figure maybe it was a cpu thing. Just a guess though.

Given all the cutbacks to visuals (trackside objects, over-draw), performance is more likely GPU limited.

As for NFS, it has, possibly, a better physics engine than Forza 3. It uses an improved sim physics model from PC company Simbin. The calculation rate is meaningless if you have bad algorithms, no? Forza 1 and 2 had high calculation rate and had bad physics. Forza 3, I hear, has many good improvements so I am eager to drive it well and see.

 

[ihamoitc2005]

Do you think the technique he suggests is viable in reality and have it been implemented in any games yet.

Sacred 2 uses similar technique.

 

[grandmaster]

As for NFS, it has, possibly, a better physics engine than Forza 3.

I'm rather taken aback that you're suggesting the handling and physics model is "possibly better" when you've not even played one of the titles you're comparing! Do you actually have any technical data to back up this assertion? Quantifiably better how, exactly?

With regards NFS, in terms of areas of technical comparison that can be measured, it comes up some way short of what Turn Ten has achieved with Forza. Lower frame rate, far more frame drops, screen tearing, what seem to be lower poly vehicles, less challenging to render tracks - the list goes on.

I'd be genuinely interested to know in what way we can objectively measure physics, because at the end of the day it usually comes down how "good" the handling "feels" and in this respect, based on my playthrough of SHIFT, it feels "OK" but nowhere near as responsive, precise or "realistic" as either Forza 3 or GT5P.

Which is all great but personal opinion is meaningless in a technical discussion thread, hence the query about actual metrics.

 

[MazingerDUDE]

Here's a couple of shots regarding alpha use in GT5P

http://www3.telus.net/public/dhwag/GT5P_Alpha_01.jpg

The wire fences are well alpha blended in this shot looking very smooth.

http://www3.telus.net/public/dhwag/GT5P_Alpha_02.jpg

Now there's no alpha here nor they are properly loded, creating some fugly artifacts.

 

[Alucardx23]

Originally Posted by Crossbar
Hmm, would love to hear the discussions leading up to that decision.
-Should we have car reflections or allow more cars in races?
-I think car reflections are cool.
-Car reflections it is then.

Just making the question because I don’t have it clear what is the cost of having reflections in a game, and why it’s so rare to see it in a quality as high as Forza 3 manages to have, so does it have any kind of effect in the quantity of polygons you can display, or is something that has no effect? And why would GT5 prologue choose not to have car reflections then? I know that is impossible to think that they choose to have better reflections instead of more cars, just want to know if it makes things harder when you try to have more cars together.

Originally Posted byihamoitc2005

I am quite sure that if Forza 3 could have more spectators it would. Instead they cut back spectator count from Forza 2. A detail reduction. Maybe for your game it is easy because of slow and also predictable LOD transitions but for fast moving racing games where camera is 100% player control I am sure this is very different.

Take note that in the track that GT5 prologue and Forza 3 have in common almost all of the spectator are 2d sprites for GT5, while in Forza 3 most of them are 3D, and that also includes the overall detail on the track, it’s better on Forza 3.

Originally Posted byihamoitc2005
Variable MSAA for this game means high chance of no AA because with 16 cars, often there will be many cars nearby, especially if you have slow cars that move like herds. 720P with 2xMSAA is "ok" but 720P with no AA is bad for a racing game.

Please take into account that the cars in GT5 prologue start the race running and always try to keep a set distance between one another, I have said this several times, when you force the cars to stay together the framerate goes down even when you only get eight cars close, If only PS360 would do a framerate test to confirm by how much it goes down.

 

[AlNom]

This is not a FM3/GT5 COMPARISON THREAD.

He talking of transparencies differences however.

He claims low-res buffer (smoke/dust) and very little transparency in GT5 Prologue compared with Forza 3.

It was a general statement regarding PS3.


Stop dragging the GT5 vs discussion back onto the forums folks. The degeneration into neverending arguments is getting old.

 

[Neb]

Just making the question because I don’t have it clear what is the cost of having reflections in a game, and why it’s so rare to see it in a quality as high as Forza 3 manages to have, so does it have any kind of effect in the quantity of polygons you can display, or is something that has no effect?

I asked myself the same question regarding Forza 3 and Shift car reflections on car body. For example in Shift the cars reflected also have transparency for windows as you can clearly see through their slightly tinted windows in the reflection and see cars ahead and trough their windows to. Most likely Forza 3 has the same for it's reflection. I also assume mirrors reflecting cars in high detail with transparencies would have an effect.

And would it up polygon count when cars reflected are 1:1 with reference cars?

 

[ihamoitc2005]

I'm rather taken aback that you're suggesting the handling and physics model is "possibly better" when you've not even played one of the titles you're comparing! Do you actually have any technical data to back up this assertion? Quantifiably better how, exactly?

With regards NFS, in terms of areas of technical comparison that can be measured, it comes up some way short of what Turn Ten has achieved with Forza. Lower frame rate, far more frame drops, screen tearing, what seem to be lower poly vehicles, less challenging to render tracks - the list goes on.

I'd be genuinely interested to know in what way we can objectively measure physics, because at the end of the day it usually comes down how "good" the handling "feels" and in this respect, based on my playthrough of SHIFT, it feels "OK" but nowhere near as responsive, precise or "realistic" as either Forza 3 or GT5P.

Which is all great but personal opinion is meaningless in a technical discussion thread, hence the query about actual metrics.

Dear Grandmaster,

I cannot respond in detail to what you say because that would be "comparisons" and I made a promise to not to continue comparison discussions (no matter who started it) to Alstrong but I can say game technology comments on only about one game:

1) If standard for proof is technical data and quantifiable proof then we must all have that standard, no? When I cannot give data for my statements or at least share images/videos for demonstration (picture is worth 1000 words) I always say "possibly."

2) Some people who have looked at NFS Shift (PC) data files say it is a lot like GTR2 for PC. They claim NFS Shift Pro-mode is true sim quality.

Information is from here:
http://www.nogripracing.com/forum/showthread.php?t=164378

 

[joker454]

On a serious note @Joker. Do you think PD or other developers have used the "small tiles" technique suggested by nAo?

I personally don't know of anyone doing that. That software tile idea came up in another thread recently, maybe it was in this thread? Anyways I replied to it then so I'll just repeat what I had said there. I had considered doing something like that way back on PS3, where I'd break the screen up into small spu-able chunks and do a transparency pass on the spu's at full res, but in small pieces. The problem was two fold. First, it was a ton of work for something that ultimately could be designed away and/or resolved with low res buffers. Second, at the time the spu's weren't used much so it made more sense then. But the idea was axed because we figured the spu's would get very busy in the next rev of the game, and maxed on the rev after that. In other words we decided it was better to spend the spu time elsewhere and stick with traditional methods to deal with the issue. Someone else may have gone with that approach, I just personally don't know of any.

Aside from that, I'm not in huge agreement with the idea of a tile rendered approach for transparencies on new hardware compared to having a chunk of edram, for two main reasons. First is from this quote:

"Most of the time we need to render slow opaque pixels and fast relatively simple transparent pixels."

I think the reason games are like that is not by choice, but by necessity, because they have to work on edram-less hardware so transparent pixels get removed or designed away to accommodate that wherever possible. In other words it's not that people don't want to go nuts with transparencies, but they can't due to hardware constraints. So they work around it with new design, tricks like a2c, or whatever.

Secondly, the 'money shot' as it were with edram would be to have the ability to sample from it. The 360 missed that opportunity this gen by having edram which was good, but not letting us sample from it in shaders. If we had that ability, it would open up doors to do all kinds of cool effects. For example we could do a monstrous amount of post processing with that much bandwidth. Most post process data buffers are reduced 1/4 size anyways, so we could shove a bunch of them in edram and just go nuts.

Anyways, cost reasons may require that edram be ditched and a hardware supported tiling approach be taken on next gen as he suggests. But I secretly hope we get a pile of fast edram that we can read/write from at will. We'll see.

As I understand the technique could also help the MSAA pass by easing the bandwidth utilisation of external memory during the render pass and thereby speeding up the render pass, is that right?

Nopers, msaa would be slow on PS3 irregardless.

 

[Rangers]

Never mind, read later posts rendering my reply irrelevant.

 

[joker454]

You said that both are predictable over-draw and that unlike PS3, Xbox360 won't be over-draw limited which is the same as saying Forza 3 has more over-draw.

No I'm saying racing games by their nature are probably the most overdraw predictable games out there. Aside from smoke/dust, nothing ever randomly overlaps like it can in other styles of games.

What is behind the cars is the road (mostly). Road has much less pixel load than the car, especially if there is transparent surfaces for windows, lights, etc.... More cars means less road which means net pixel load is much more.

The road is usually at a very skewed angle to the camera meaning to get it to look nice you have to either use a high level of af or tweak the lod bias, both of which mean the road might have a higher texture gpu load than the cars do. Plus the road has to use a similar lighting model as the cars if they are to all match. The road might not have stuff like vehicle reflections, but it does have other stuff unique to it like tire marks which need to be maintained far into the distance, unlike reflections that can be lod dropped fairly quickly.

Variable MSAA for this game means high chance of no AA because with 16 cars, often there will be many cars nearby, especially if you have slow cars that move like herds. 720P with 2xMSAA is "ok" but 720P with no AA is bad for a racing game.

I don't believe the "no msaa is bad" argument given that most PS3 games ship without it yet few people seem to mind. You can design the issue away like GT does by not letting cars herd together. Or just drop msaa when there are many cars nearby. That would get them at least 12 car support if it was purely a vertex limitation.

 

[Dr. Nick]

snip about tiling.

In a case where you got stuck porting a 360 lead game,where whoever was in charge decided to make use of edram bandwidth, to the PS3 you had to preserve most of the original content would you give that idea a try?

Anyways, cost reasons may require that edram be ditched and a hardware supported tiling approach be taken on next gen as he suggests. But I secretly hope we get a pile of fast edram that we can read/write from at will. We'll see.

I get the feeling that is something that may happen. I hope devs let it be known that it's something they want in the next systems and the best possible way for it to work.

 

[Acert93]

imo Post processing and singificant transperancies (everything to smoke and clouds to grass and leaves) opens up the doors for probably the biggest visual leap in titles from this gen to the next (animation is right up there, too) so lets hope that becomes a hardware design influencer. But I think typically the best "bang for buck" hardware dictates software more often.