The GT300/Fermi Thread

[damha]

The Fermi Paradox



source: http://en.wikipedia.org/wiki/Fermi_paradox

lols

[Decami]

No.
All the modern accelerators even in multi-GPU setups struggle with Crysis with proper realistic resolution (4k x 3k).
So add a year for the tech to catch up at least.
Texture quality are also lacking. Need 2x-4x higher texture quality. 2-3 more years.
Geometry... is WAY behind. If you want photo realistic image quality in a 3D game you need 10-20x more polygons at least. That's 5-6 more years till it's actually possible to render that.
Current lighting is crap. We need raytracing. Realistic raytracing with huge resolutions won't be possible really soon, 5 years at minimum.
And now imagine creating this game. The content. Each model will take a TON of time to create. This game will take 5 games to develop for a HUGE studio.
So don't expect anything within the next 10 years for sure.

I love how you say "this game," as if these advancements would only be used in only one game before it anyone else used it. I also love how you seem to be adding the years together, as if one company and one will do this. Thats not the way it works.

Even if everything took as long as you said, its within a 10 year time frame.
People pull the Crysis theories out of your head. That game was no massive advancement, just a lot of semi recent methods put into play
and optimized visually. No one is doing, or has done this because in most cases this doesn't make them money.
I guarantee you in exactly 1 year hardware will have caught up with Crysis, and thats an awesome thing, thats technology moving faster than you think it is.

[zalbard]

I love how you say "this game," as if these advancements would only be used in only one game before it anyone else used it. I also love how you seem to be adding the years together, as if one company and one will do this. Thats not the way it works.

Grats, you have completely missed the point of my post while picking on my use of English...
I went ahead and heavily edited my post to reflect this.

Even if everything took as long as you said, its within a 10 year time frame.

No.
Hardware might be, yeah (barely, probably), but the developers will not have enough time to actually finish a serious game with such a level of gfx by then, the creation process itself will take a while as well.

People pull the Crysis theories out of your head. That game was no massive advancement, just a lot of semi recent methods put into play and optimized visually.

Yeah, sure, there are so many games that look a lot better these days! Oh wait...

[Chumbucket843]

You do realise that high polygon models will take 4-10x more time to create, right?
Look at the movies, that's not actually realistic gfx yet, cause you can't really come closer and look at everything carefully.
They take 5+ years to draw and render (Avatar for example), and this is very little content comparing to any interactive game, cause they only create the stuff that makes it into movie's frames... And for 2-3h of entertainment only...

well then lets ditch tessellation and go with svo. this is a very powerful wasy to increase texturing and geometric realism all in one.
http://www.youtube.com/watch?v=VpEpAFGplnI

[takamishanoku]

Guys as I would venture off topic aswell, I can't wait for HL 3. That should look incredible. Anyone else wondering why it's gone all quiet with epi 3?

Btw Crysis is definately photorealistic at times and def would be with a bump in res.

This is what happens to a thread when there is no real news to report. Give us something nvidia!

[Decami]

Grats, you have completely missed the point of my post while picking on my use of English...
I went ahead and heavily edited my post to reflect this.

No.
Hardware might be, yeah (barely, probably), but the developers will not have enough time to actually finish a serious game with such a level of gfx by then, the creation process itself will take a while as well.

Yeah, sure, there are so many games that look a lot better these days! Oh wait...

I did nothing to pick on your english. Didnt seem to be any mistakes in the previous post, at all.

I'm curious, you do know much about the industry, or are these guess assumptions out of thin air?

Yeah, sure, there are so many games that look a lot better these days! Oh wait...

No one is doing, or has done this because in most cases this doesn't make them money.

^ did you purposely completely ignore that line.

lets face it, Crysis was not a very good game, not terrible, but everyone will agree with less focus on its trying to be photo realistic and
more focus on gameplay it might could have been in a top 10 great games of all time.

Crysis is popular and sold what it did cause of its hype, no one else is going to be able to hype graphics like that and sell copies for a while.
This doesnt mean i dont like or dont respect he game for what its done, its pushed hardware and thats a great thing, but this explanation is the reason my above quoted statement is true.

You sir, are the one missing the point, the point of your post was clear. But to an extent I do agree with you, exact photo realism of life in a game
I see not happening for some time, it will come close sooner than we think, but i could be wrong. But the advancements being worked on at the current moment are looking bright.

to add, what in the world does this mean.

All the modern accelerators even in multi-GPU setups struggle with Crysis with proper realistic resolution (4k x 3k).

4kx3k, where did you even get that number, I will tell you, this subject is a specialty of mine, and theres really no such thing as a proper realistic resolution.
that is based on so many things its ridiculous.

[BatteryOperated]

<annouancer> LOOK ITS FERMI!! I am not sure how many here know about tessellation, but fermi has it! Tessellation!! TESSLLATION!!!
<man holding an anti-sign> OMG TESSELLATION!!!
<announcer> Say tessellation 10 times real fast and get a free card!
<excited man> tessellation tessellation tessellation tessellation tessellation tessellation tessellation tessellation tessellation tessellation
<excited man 2> I can't believe this is happening! I hope they bring back Elvis!

[ElSel10]

Guys as I would venture off topic aswell, I can't wait for HL 3. That should look incredible. Anyone else wondering why it's gone all quiet with epi 3?

Btw Crysis is definately photorealistic at times and def would be with a bump in res.

This is what happens to a thread when there is no real news to report. Give us something nvidia!

Actually, I like Cherry Pie.


... This thread is a joke and it's ridiculous we aren't allowed to have another one.

[***Deimos***]

No.
For example, Crysis.
So, let's say it represents the current level of our gfx tech advancement. Let's try and get it to the realistic level.
All the modern accelerators even in multi-GPU setups struggle with Crysis with proper realistic resolution (4k x 3k).
So add a year for the tech to catch up at least (++ computing power requirements).
Texture quality is also lacking. Need 2x-4x higher texture quality. 2-3 more years (++ computing power requirements).
Geometry... is WAY behind. If you want photo realistic image quality in a 3D game you need 10-20x more polygons at least. .............

THE LIMITS
Why are we still powering games at 100fps+, where fps looks like a cardiogram?

What happens when #triangles is > #pixels? 19x10 .. is 2MP. A 100 mtri game would have 50 triangles/pixel... inefficient.. you think so?

What happens when #SP is > #pixels? Just a couple quick years and we already have 1600-3200SP. Once we reach millions of SP, we'll have enormous inefficiencies.

1. First of all, Crysis is over 2 years old (Nov 2007). I don't exactly see the likes of Batman Asylum pushing the envelope. The "saddest" part is that it took Carmack from id, 15-20 lines of code to implement mega-textures.. something even he was astonished nobody did yet.

2. Bigger texture and more triangles is an uphill battle of diminishing returns. ex. 3x bigger texture uses 9x more bandwidth/space. This brute force is wrong approach.

3. Every game, ESPECIALLY those like Crysis, use clever tricks to drastically reduce amount of work being done with minimal loss of detail. This is not a bad thing. This is why Radeon 9700 could do AA so easily.

4. Chumbucket843, thanks for sparse voxel example. THIS is what I want to see more of. More innovation. Not just more of the same. I think Fermi is VERY AGGRESSIVE step in the right direction.

PS: Gene/DNA computers are supposedly super at parallel tasks. So 3D gaming on "embroys" is a go?

[Florinmocanu]

Grats, you have completely missed the point of my post while picking on my use of English...
I went ahead and heavily edited my post to reflect this.

No.
Hardware might be, yeah (barely, probably), but the developers will not have enough time to actually finish a serious game with such a level of gfx by then, the creation process itself will take a while as well.

Yeah, sure, there are so many games that look a lot better these days! Oh wait...

Damn dude, you don't actually read what i said.
Your pulling stuff out of your imagination/personal assumption while i do 3d graphics for a living and still you actually think that high geometry will take a lot to implement.


Guess what, High poly models are created these days for every damn model in a game, coupled with a low poly model to be inserted in-game

You have no idea how a game pipeline works, so stop posting BS.

Don't take it as an offense, but you have no clue about this subject.


About high res textures, that is not such a big deal actually. Right now, they use 512pixels and 1024pixels textures. a 2k-3k texture usage will need video cards with 4-6GB of ram per gpu, but that is not actually such a big deal.

Main issue for the future is tesselation and displacement in particular and actually, we just need more render power so GPU can calculate displacement faster, a lot faster so we can use it everywhere in-game, but properly, not Stalker or Dirt2 style.

[KingOfsorroW]

Maknig a game that has Avatar level of graphics will take a lot of time. It's not only about the technology, but also and mainly about hell of the amount of time to spend on the project and work. And of course money. Zalbard is right here it would take a decade to create a game that would look like Avatar at least. If it were otherwise we would be having Avatars every two months in pc games and in the movies too )) And it would definitely be making them money, if a project like that wouldn't have required that much of efforts and time thus bearing a load of risks to never pay off...Right?
It looks like you guys are saying the same thing but don't understand each other for some reason.

[Florinmocanu]

Maknig a game that has Avatar level of graphics will take a lot of time. It's not only about the technology, but also and mainly about hell of the amount of time to spend on the project and work. And of course money. Zalbard is right here it would take a decade to create a game that would look like Avatar at least. If it were otherwise we would be having Avatars every two months in pc games and in the movies too )) And it would definitely be making them money, if a project like that wouldn't have required that much of efforts and time thus bearing a load of risks to never pay off...Right?
It looks like you guys are saying the same thing but don't understand each other for some reason.

No, we are not saying the same thing.

He argues that if you want a lot of complex geometry in a game, than surely it will take a huge amount of time to create it.

I'm saying that in today's games they are already creating high poly models along with the normal, low poly in-game model. They use the high poly model to create normal maps, bump maps, ambient occlusion maps etc.. to enhance the shading on the low poly one. They are already creating high poly models for each box and each character in the game.

How do you guys think they create a normal map? with the additional detail? You cannot manually paint it, you extract it from the high poly model and apply it to the low poly model.

So, in the future, if GPU power will be sufficient, it will be faster to create a game since you can skip creating the low poly version of all models. More geometrical power for GPUs in the future will mean simplifying the workflow and get things done faster.


And also, VFX in movies are done by having huge render farms behind, with 500-1000 CPUs or even more, that's why you don't see that kind of work done in games, because it's damn impossible to do it, not because of time constraints but because we don't have the processing power to render such a game in real-time.

[saaya]

4. Chumbucket843, thanks for sparse voxel example. THIS is what I want to see more of. More innovation. Not just more of the same. I think Fermi is VERY AGGRESSIVE step in the right direction.

hmmm why? whats the agressive step nvidia took with fermi?
the ability to handle more geometry?

Maknig a game that has Avatar level of graphics will take a lot of time. It's not only about the technology, but also and mainly about hell of the amount of time to spend on the project and work. And of course money. Zalbard is right here it would take a decade to create a game that would look like Avatar at least.

i dont think so... right now there are many games using many game engines which use different lightning models... the market of game engines is consolidating and this trend will continue... its less and less about the engine, and more and more about a set of objects and textures that come bundled with the engine... so you dont actually have to build the same thing over and over and over... there will be more and more objects that get recycled in various games... and as such, artwork wont require that much more time and money i think...

plus afaik a lot of work right now on artwork is spent on lightning things... manipulating the code to make the scene look the way you want... once you have raytracing for lighnting, you wont have to do that... unless you want abstract lighnting, and even then it should be very easy to do compared to ligghtning now...

[jmke]

there will be more and more objects that get recycled in various games.

that would be a good thing imho, we don't need 1001 barrel variations

[KingOfsorroW]

No, we are not saying the same thing.

He argues that if you want a lot of complex geometry in a game, than surely it will take a huge amount of time to create it.

I'm saying that in today's games they are already creating high poly models along with the normal, low poly in-game model. They use the high poly model to create normal maps, bump maps, ambient occlusion maps etc.. to enhance the shading on the low poly one. They are already creating high poly models for each box and each character in the game.

How do you guys think they create a normal map? with the additional detail? You cannot manually paint it, you extract it from the high poly model and apply it to the low poly model.

So, in the future, if GPU power will be sufficient, it will be faster to create a game since you can skip creating the low poly version of all models. More geometrical power for GPUs in the future will mean simplifying the workflow and get things done faster.


And also, VFX in movies are done by having huge render farms behind, with 500-1000 CPUs or even more, that's why you don't see that kind of work done in games, because it's damn impossible to do it, not because of time constraints but because we don't have the processing power to render such a game in real-time.

That's exactly the point he was trying to make I guess but somehow went into the wilds of crysis. lol I understand that the main point of discssion was the impossibility for a long time to come to make a game that has some photo-realistic graphics (not just high poly models) or graphics like in the Avatar movie - well, everyone agrees on this one I guess, so thats why I said there is like some kind of misunderstanding going on )

[Florinmocanu]

yup, a lot of time takes right now to fake GI by using multiple light sources to give the impression of light bouncing through the scene.

With raytracing you have you normal light source, lightbulb, sun etc.. and the rest is calculated by the engine.

[kgk]

About high res textures, that is not such a big deal actually. Right now, they use 512pixels and 1024pixels textures. a 2k-3k texture usage will need video cards with 4-6GB of ram per gpu, but that is not actually such a big deal.

Then why do 4096x4096 texture packs for stuff like Oblivion run just fine maxed out on 1GB video cards?

Honest question

[Florinmocanu]

Because they do not affect everything. Nobody does a 4k texture for a rock, planks, leaf or some rat/cat in a game, they are only for enviroments, walls, backgrounds , some characters etc....

If you would have film quality texturing (4k textures for almost everything), then you would need a lot of RAM.

[neliz]

If you would have film quality texturing (4k textures for almost everything), then you would need a lot of RAM.

Or a different way to access a pool of system memory, fast.

Even though memory throughput to the GPU is not a bottleneck, the limited amounts are. I'd like to see AMD and NV invest in a new memory transfer protocol so that graphics chips can enjoy fast transfers from their VRAM but have a faster way to access system RAM than todays normal Memory caching methods. Since this is evident with yesteryears architecture (GT200 chokes when it runs out of memory) I can only hope that NV (and ati) look further than wait for PCIe 3.0 where basically doubling bandwidth to 32GB/s only puts it around 15% of the VRAM bandwidth.

Solving that bottleneck would allow for massive amounts of textures in game.

[saaya]

Or a different way to access a pool of system memory, fast.

Even though memory throughput to the GPU is not a bottleneck, the limited amounts are. I'd like to see AMD and NV invest in a new memory transfer protocol so that graphics chips can enjoy fast transfers from their VRAM but have a faster way to access system RAM than todays normal Memory caching methods. Since this is evident with yesteryears architecture (GT200 chokes when it runs out of memory) I can only hope that NV (and ati) look further than wait for PCIe 3.0 where basically doubling bandwidth to 32GB/s only puts it around 15% of the VRAM bandwidth.

Solving that bottleneck would allow for massive amounts of textures in game.

not only bw, i heard the worst problem with ram caching is actually latency... pciE supposedly has a very high latency... high enough to be a problem even for gpus

[SKYMTL]

Even though memory throughput to the GPU is not a bottleneck, the limited amounts are. I'd like to see AMD and NV invest in a new memory transfer protocol so that graphics chips can enjoy fast transfers from their VRAM but have a faster way to access system RAM than todays normal Memory caching methods.

Better watch out, Rambux probably has a patent for that.

[neliz]

Better watch out, Rambux probably has a patent for that.

QPI/DMI is teh slows.. we need a HT slot for graphics!

Maybe NV can buy RamBucks? they'll earn a lot of cash.

[BenchZowner]

Actually Intel has a good part of Rambus's shares, and I doubt they'd cash them out anytime soon or at all.

[WeeMaan]

This thread is sliding more and more away from original topic.

[saaya]

QPI/DMI is teh slows.. we need a HT slot for graphics!

Maybe NV can buy RamBucks? they'll earn a lot of cash.

qpi has bad latency?
or your saying HT is the way to go cause its an open standard?