Graphical effects rarely seen this gen that you expect/hope become standard next-gen

*sigh*

I'd rather take the fluid dynamics and distortion from R2's water over reflections any day of the week. Reflections don't mean jack if the water is static or moves in awkward ways (ala Uncharted 2). The water in Uncharted 2 is probably the games least impressive feat.

 

I'm thinking the main goal was to deal with the blending, fill rate, and AA problems the Xbox1 had. I'm sure when they asked the development community those problems cropped up more often than most. I wonder if the development community asked for a scratch pad though.

 

I don't think it is totally obvious as there are a number of options, the question is what compromises.

1. eDRAM "like Xenos." Size undetermined, but essentially a framebuffer.
2. eDRAM "like the PS2." Size undertermined, but surely bigger than option 1.
3. No eDRAM "like the PS3." Split memory pools.
4. No eDRAM "like the Xbox." Unified memory pools.

It may be, although I doubt it, that option 1 is the best compromise for performance and features. #2 may be far too expensive. Maybe there are no good options other than Xenos style tiling and enough eDRAM to texture and do other effects for HD resolutions and textures is too high for a complex multiport memory that also has a huge bus to the main GPU (or, worse, on chip). #3 may not be desirable due to complexity and busses. #4 could be a killer in terms of bandwidth (GPU eats up the bandwidth and leaves none to share).

So it may not be an issue like shaders where incrimental increases in performance and features is a given. The cost/benefit of doing a PS2 style eDRAM may just be too expensive to get the benefits desired within the budgets. If it was cheap, why did Xenos reduce features from the PS2? The cost was probably not seen to be worth the benefits.

The cost of making an smarter tiling eDRAM with more read/write abilities and maybe even general purpose use may be so expensive that an alternative model (e.g. 3 or 4) is better (I think nAo leans this way, even this gen to a degree) or that 1 is the best compromise and let devs work out the finer details of the compromise.

But I don't see devs happy with 1 again, but it isn't obvious that 1 will be totally dismissed. Cost wise maybe a slightly bigger pool (e.g. 28MB for 720p 4xMSAA with tiling mechanisms ala Xenos for higher resolutions/color) will be the end product and devs needs to adjust accordingly?

 

I would also add, "What is the GPU vendor trying to sell MS, and how do they propose their silicon deals with the problems MS raises." e.g. Even if developers say, "We want a scratch pad!" AMD could say, "That would be 4x bigger with a 2x larger bus. Note our new designs deal quite well with memory, so this isn't needed. What we propose is [---]" where [---] is their current design features they are upsell.

 

A proper TBDR GPU would probably work well with a smaller amount of eDRAM. I'd have thought TBDR would work well with shader-heavy pixels too. A nice chunk of eDRAM could be very fast and pretty flexible.

I think these come down to cost more than anything. #4 is ideal, and the tech should exist, but wide busses are too costly. I'm still holding out for XDR to hit their targets! If you can't afford a big blob of fast RAM, split pools could offer a fair compromise of performance and cost. eg. A Flash RAM cache could add several gigabytes of low-latency virtual RAM.

Although I see this is going OT!

So ignoring what hardware is available, I'd like to see convincing dynamic GI and much better IQ. I want more variery from procedural content and better animation. I think we're so far from photoreaslitic graphics that they're not a reasonable target next gen, so I'd rather have cleaner visuals and better games, than chase just graphics.

 

Well, it'd need a fast bus, and that costs, which I guess is the reason for the limited solution.

 

It could also have been an engineering afterthought.

 

Uncharted 2 water reflects and refracts properly, flows (in Uncharted 1 the player rides up river against the current and in Uncharted 2 the player can fight the current in a channel he's supposed to cross), and ripples when the player(s) runs through it, in SP as well as MP, can they improve upon it? Of course they can, but it does what the game needs it to do.

But you would rather have R2 water that looks more like baby oil, doesn't really behave like normal water, and worse yet, it visually represent water poorly with no reflections and the distortion is wrong and more like that of baby oil, and the so-called "fluid dynamics" doesn't factor into the gameplay at all, they never really took advantage of the buoyancy in any meaningful way, if you want to complain about static water, R2 water is the definition of static water, there were basically static pools of water. The player is enjoying none of the benefits of their implementation, yet you would prefer a feature that does nothing to help the game and hurts the game's overall look when so many other games have reflective water.

As a GRAPHICAL feature R2 water is a failure, the game doesn't look better because of it, it didn't do anything as far as the gameplay is concerned and processing power could have been better spent elsewhere.

 

Also, do keep in mind MRTs...

 

from what I remember Uncharted1 was supposed to have better water like R2 that moves a lot better.

All Uncharted has better is reflections and imho that effect is NOTHING if the water just gives a weak splash.

 

no, not at all. I personally really like R2 water, looks like the DiRT DX11 tessellation water effect in the trailer.

 

May be we should move the Edram talk in the "predict Next gen thread", would make more sense, no?

In regard to Edram my feel (after trying to have a better understanding of GPUs, still by far a work in progress tho...) is that to become more flexible GPU memory model has to be reworked. GPUs need more generic access to data (throught texture most of the time now), they need more generic caches supporting read and write. Both Microsoft Halo team and Epic highlight the need of super high bandwidth, looking at the aggregate bandwidth they are speaking about, from my understanding L1 caches seems in the best position to deliver while offering other much wanted features (read/write operation, coherent memory model).
Sweeney thinks it should be possible to render micro-polygons with next generation system whether this holds true or not I think that his idea on tiled based rendering is still relevant especcially the use of tiles that are optimized for best cache usage. He hints of 8x8 pixels for the bins size, that 64 pixels ie 2048bits. That's 512bits per "component" (not sure it's the appropriate naming I mean R,G,B and A could be X,Y,Z and W), or it can be 16 pixels per cache line. I don't know the size of a cache line but Sweeney knows hence his choice can't be innocent, 512bits as a value (for cache line size) seems consistent with 16 wide vector unit. I don't which arragement would be better if you consider scatter/read.

Still different hardware vendors may come with different solutions. I'm wondering about what will be MS attitude in regard to directx12. I feel like there are still need for standardization. Could they push toward TBDR with defined tile size? Could they enforce 16 wide as the maximal/most wanted granularity/width? Shortly a bunch of questions and depending on the answer EDRAM may not end useful.

 

FEAR had muzzle flash shadows. Did Doom 3?

And the effect I'd like to see used more often would be shadows created by rising smoke. I don't think I've ever seen that.

 

More sparks and bloom.

Subtle use of DoF. I dont want 70% of the screen blurred out.
In reality i want good self shadows on models. Not hat flickering mess that they call shadows in Mass Effect 1.
Persistent smoke. I hate that fires in games have smoke trail of ~15-20m, and that grenade smoke is gone in ~5sec.
Clothing that is not part of the model [sorry for my butchered engrish]. How to explain this easiest... im playing DA:O right now and i notice all the time when NPC's are moving their arms, their clothes STRECH [aroung heir shoulder area] in the most unnatural ways. Is there any game that has separate model for clothing, that react nicely when character moves in them?

 

World in Conflict did that.

That's a biggie for. Could anyone comment on why this is so hard to do?

 

Well, to do it "right" you'd need soft body (cloth) simulation -- including rather fine-grained collision detection -- for each piece of clothing on every single character.
I also really want to see that but given current hardware limitations it's not particularly surprising that body armor is so popular in games going for realism.

(And beyond the computational requirements I'm not sure how well the content creation pipeline is equipped to handle real clothes at this point)

 

Water

I have to agree my friend. Resistance 2 has the best water I have seen for wave and disturbance simulation. Most water only has surface disturbance simulation but R2 also looks how underwater movement can affect a water surface.

If you have been in the ocean and moved hands quickly under water you can see similar waves (soft, slow).

http://www.youtube.com/watch?v=JyuY3-SLuDA&feature=related

This video has good critique of the water (and also some other things):

http://www.youtube.com/watch?v=Y0Nm_v5DanM&feature=related

 

What do you do when someone or several people start spamming grenades? That is part of the reason the smoke is gone in about 5seconds.

DOAXBV did it on the Xbox and so did DOAX 2. I think the main problem people have with it is the extra time that has to be put into creating such models.

 

And water is another thing I'd agree with. I haven't played R2 so I'd have to say Bioshock DX10 is the best I've seen. I can't count how many times in that game where I just stopped and starred at the water as I waded through it. Overall more advanced and realistic water movement.

 

Does the water react with your movements in Bioshock? I played it on my PC and consoles but none depicted water reaction as far as I remember