Future console discussion and thoughts

[Cheezdoodles]

Since the PS3,Wii and Xbox 360 are all out thus making it current gen, i thought it would be fun with a discussion about the Nextgen consoles (PS4,Wii2?, Xbox 720?). It will be fun to drag up this old thread in 2011 or whenever than generation launches and compare our thoughts with reality.

Here are my 2 cents:

Technology

What can we expect from a next-gen console, in terms of hardware? Multiple- GPUs? CPU's with 40x cores?

Before we can answer that question, we need to ask when will the next-generation begin? How long will the consoles last? I would say it will begin somewhere between 2010 and 2011.

What do you think we can have in the hardware department in a $400 (or $600) console in 5 years?

I think it will look somewhat like this (numbers are based on the increase we have seen from the X360\PS3 compared to the Xbox) :

- Cell-ike CPU, with more cores. (definately alot of cores)
- 4 GB's of RAM
- 64mb of eDRAM (yes i know the xbox 1 didnt have eDRAM and if i compare the X360 to the PS2, the number shouldnt be this high, but i like that number )
- GPU (maybe dual cored?) with 1600+ million transistors (transistor count based on the 5,26x increase we seen in the PS3 vs the Xbox 1) based on whatever PC GPU's are pushing at the time.

Storage media: Bluray or whatever wins + standart HDD.

Resolution: 1080p (i doubt they will strive for higher resolutions, i certainly dont hope so)

Things that will vastly improve compared to current gen (other than graphics)

With exiting things like Procedural synthesis starting to hit the "streets" i think we will se vastly improved animations and destructibility in games.

Photo-realism?

After looking at screenshots from the Cryengine 2, im more confident than ever, that a dream that atleast i have dreamt about since i was a little kid, can be achieved, that small dream is photo-realistic games. Lets face it, sooner or later (probably later), the only thing that will prevent us from not being able to see the difference between real-life and games will be the fact that games will still be bottle-necked by the amount of pixels that our monitors\tv's are able to show.

Will it be achieved in the next-generation of consoles? Probably not, but it will look extremely close. The potential is certainly there, Crysis, which is only a first generation DX10 game, but the environments are slowly getting to the stage where i cannot really see (read: care) about the difference anymore.

Question: Will we get photo-realism during the next generation of consoles?

Personally, i think we can get close, very close. It wont be look as real life, but it will be close. Close enough, to the point were casuals and most normal people wont care anymore.

This of-course brings me to my next question:

What happens when we achieve photo-realism?

A friend of mine and i recently had a discussion about this, and he had some interesting view on the matter, he said that photo-realism is boring. When we achieve true-to-life graphics, and the initial buzz wears off, nobody will care about it anymore. We have true-to-life graphics in real life, its just a matter of opening our eyes, it will bore us. There is only so many amounts of polygons we can look at before we don't care about the difference anymore.

Art direction will probably be more important than ever before, because it will be the only real way to differentiate your game.


Other Questions:

When we achieve photo-realism at say so many frames per seconds that nobody cares anymore, what will happen to GPU evolution?

After we achieve photo-realism and the GPU's run photo-realistic games at super high framerates, will developers actually bother to develop new graphical engines?

Would we bother to ever again upgrade the hardware (not after next gen, but after we achieve the above mentioned situation)? Only situation i can think of is not really performance based, but stuff like VR or "materix" like consoles, surely from a performance standpoint, nobody will give a rats ass about how many supergigaflops the PS100 pushes.

So what are your thoughts on the future generation?

 

[cloudscapes]

I think we're going to have to deal with Uncanny Valley a lot in this generation and the next (2010-onwards). It will be interesting to see how devs cope with this. I expect the term to become more known and possibly even used as a buzzwword when marketing a game.

It will also be interesting how tools will evolve, since a greater amount of detail will require tools that will aid-in and automate some of the process. Simple example, spraying foliage onto a landscape based on color masks rather than the old-skool hand-placing/hand-painting way. It starts to get difficult when teams pass 125-150 devs, so just simply hireing more people do do all these extra details isn't the ultimate solution IMO.

 

[liolio]

My overall ideas
for the gpu it's a given it will be directX10 compliant.
Both MS and Sony system will aim at 1080p as a native reslolution. It's clear.
No that it's the best choice but none of this manufacturers will want to loss the next PR war...
I think 720p could be a better choice :
more shader power per pixel or save some power to do some other stuffs...
less edram or ram at the same level of AA
==> bigger transistors budget for the cpu.
Scaler technology will become better too.

For me photo realism is not the goal for next gen.
Imagine astounding looking games with cheap physic, lifeless animation.
The more the image is acurate the more the laking in physics, animation are stricking.
( I don't konw that for sure but it maybe related to what scientist call "uncanny valley")

I hope next system will be balanced between better animations/interactions/physics/IA/improved graphics and will not spend all the transistor budget on the gpu.

For the cpu I would be surprised if the cpu is cell like for both systems.

Sony is likely to choose a improve version of the cell.
The memory controler is inside the cell, texturing throught flex IO induce huge latencie.
So Sony is likely to keep a NUMA architecture, or give the cell an huge rework but I can't see the cell works properly with an external memory controler.
So Sony is somewhat stuck with it's design for the better and for the worse.

For MS, it's more hard to guess. And so more interesting.
They could go with AmD/ati or stay with ATI(amd lol)/IBM.
I was thinking that Amd project are interesting and flexible and coukd map with MS goals but 3dilettante explained me.

AMD would probably go asymmetric before Intel (not that Intel won't do it eventually) because of their process disadvantage. Intel can afford to bloat up on x86 cores because they've always had at least a 6-month lead in process transitions.
Even on the same node, AMD's cache density has been measurably worse than Intel's, something symmetric x86 multicore would be more hurt by.

AMD needs to get more out of its more limited transistor budgets, and hope for two things:
1) Intel thinks the simpler process of adding more symmetric cores is a better bet
2) it turns out it isn't

So my guess given this insightfull response is that MS will stay with with IBM, backward compatibility can be an issue too.

From what i've read here the sweet spot for SMP is 4 cores, but valve next engine seems to do well with eight cores. performances don't ramp liearly but their room for improvement especially on a closed plateform.

I guess MS will choose 8 cores some kind of OoO and smtT or cmt implementation (or anything that help keeping execution units busy depending on transistors budget and power/heat requirement)with strong smid units.
Or maybe trade one core for dsp like devices, i've no clue.
It makes even more sense if you take in account that ms is a software company and will make choices that are compliant with availabe game engines and will consider vista/xbox portability
I think MS will keep an UMA architecture with some smart edram for framebuffer operation.

So I think on the hardware side both system will be different NUMA/UMA ans SMP/ASMP

interestingly, when both Ms and Sony will have to make theirs choice they will still be somewhat blind. They will have to make hardware decision for the system design before they know how much their actual architure yeld good result in regard to the challenge devs are facing ie write multithreaded games.

 

[pc999]

My takes:

Sony will have at the center of the console will be a (1?) cell based processor and very likely some Nvidia tech in it (probably inside the chip), again prepared to be a multimedia device with whatever will be at the time.

MS will have a console that will have a costum AMD fusion processor, meybe with a discret xPU if they aim for a higher price (400$ or more), with a standard big mass storage divice in it completely adjusted to episodic content and such.

Both will be perfectely able to display next (next) gen 1080p at 60FPS (probably not with a jump as big as PS2/XB--->PS3/360) but that will not be the main focus, instead physics/animation/AI/processural creation/... will be the focus.

For Nintendo I think it will also resemble AMD fusion (even if not by AMD), althought a low end version of it.

All the 3 wll do a good work in new interfaces, wouldnt suprisse me if next gen motion tech, speech recg. and/or standard camera (AR, Eyetoy 2) are standards by next gen.

 

[nonamer]

I think by next gen all consoles will be Cell-like, or some other asymmetric form of multi-core. It's just absurd to try otherwise at those levels of performance. Think about it: A 32 SPE + 2 PPE Cell, going at 7Ghz+ will get you ~1.9 TFLOPS (1 vector MADD/clock per core). With symmetric, like say 8 OoOE cores @ 7 Ghz will only get you 448 GFLOPS, and even thats unlikely since it would probably not hit 7Ghz. Plus it would be way more expensive to make. And you can't even claim ease of programming anymore, since at these numbers of cores, it's either embarrassingly parallel or it's simply not going to utilize 8 cores very well at all. Trying to do something like 16 or 32 way symmetric is also stupid, since while you become an equal in performance, you'll suck at single threaded operations since future Cell should (in theory at least) have OoOE-based main cores. It should be obvious, from both the consumer and developer standpoint that one method is way better than the other going into the future.

I'm also interested to see some ZRAM being used. With twice the density of conventional eDRAM, you can easily fit an enormous amount of eZRAM on the CPU die. I've did some quick math and something like 128MB of ZRAM would only take ~100mm^2 at the 45nm process. The only catch is that it need SOI wafers, which may limit this to only Sony based consoles. No matter, Sony does seem like the company that would use it. If they are willing to either build the PS4 at the 32nm process, or go 45nm and let the CPU reach 300+ mm^2, they can easily fit a gigantic amount of ZRAM. With a proper interface with the GPU, the framebuffer problem will be solved, as well as make for an awesome piece of cache/scratchpad-RAM. How impressive would Quad HD (2560x1440) with 4xAA for free sound? Or only touching main memory when fetching textures or streaming content, with main code running exclusively in on-die RAM? That's something to be about.

And GPUs should be DX11 unless graphics technology slow down signficantly. DX9 to DX10 hardware was only 4 years, so unless some console comes out in 2010, it should definitely be DX11.

 

[Rangers]

I predict they'll have more stuff.

I hope they aim for 720P rather than 1080P, but I dont know how that will shake out. We really just need displays that dont force you into a choice..

4 GB of RAM would be the expected..8X prev gen. Xbox 64 MB, Xbox 360=8X=512. Though I guess PS3 actually had a 16X jump (32>512). Then again PS2 was an extra year older.

I'm actually wondering if Moore's law is slowing down and 4GB wont happen though. It does't seem to me RAM is increasing in the desktop space at near the breakneck pace it used too. 1 GB of RAM has been acceptable for a desktop for ages now.

I also dont think a cell type architecture is by ANY means a given, at least for microsoft's next foray. To me it is still uproven whether Cell will be utilized effectively this generation. A lot hinges on that answer.

I would def hop MS switches back to an AMD CPU. Just cus. Something like a beefy 8 core (maybe 16 by then?) AMD CPU. Forget this in order junk, and go with something easy to program (they would have been MUCH better off with an AMD dual core in 360 imo).

It's clear there will still be a massive leap in technology. Something like a kentsfield and 7900GTX in quad SLI would basically already provide a "next gen" theoretical power increase, and it is available today. So there's no question of hitting any walls.

I'm basically tremendously excited for it. It's my belief that this is the first gen that allows us to do "pretty" 3D. Aka you look back at PSX or PS2 games and they are ugly, jaggy..ugly. You look back at 2D SNES games and they are still pretty. This is the first gen that I see pretty 3D, due to the resolution, power, and AA increases enabling clean images. That in itself is a huge step. Next gen, I hope for the Killzone demo, the motorstorm demo, if not better I suppose, in real time. And that will be awesome! And should be achievable. At least I think so...

 

[liolio]

Woa!! you're enthousiastic (nonamer).

My first point : we still have NO PROOF of witch cpu (ie xenon or cell) will perform better in regard of game code dev will manage to push.
If the cell performs slightly better we have to take in account that cell is bigger chip for the same transistor budget xenon could ~4 cores + 2MB of l2 cache.

My seconf GFLOP is not the good mesure to mesure a system performance ( i know it has always be like this for console folk mips, flop, etc.. but that prooves nothing).

I've make my estimation based on some topic i've more than one time one this board, and It seems pretty clear that far far more knowledgeable members than me don't think cell is a perfect implementation of ASMP. Nor that xenon is a perfect implementation of SMP (far from it from what I read lol). And at some point it's clear that is useless to compare differents architectures without taking in account witch workload it's supposed to handle. Shortly nobody agree on that matter and for astoundingly complex reasons (at least to me).

Anyway it's a given that sony will stay with a cell++. the ibm roadmap sys 2 ppe+32spe ~2010.
It will still be a complex ship, but it's the road map for the DP version
So sony could chose such a cpu in SP version.

That don't mean a SMP cpu can't do better for a given workload ie games.

More if Sony and MS will have slightly the same transistors budget, so if the cell++ is 1.9 GFLOP, this number is likely to be irrevelant in regard to the number of flop the gpu will push.
Ms could chose to have a slightly tinnier SMP CPU and a bigger GPU that could run some embarassingly parallel problems (GPGPU will have some root in 2011).

For ZRAM i don't know. It could be interesting instead of edram.
In the cpu I'm not sure, some cpu have huge L3 cache but no cpu has edram or alike so I guess latenci is more an issue than size. (NB i haven't found latencies figure on your link )


Anyway I hope this thread won't be derailed by fan-boy because a lot of senor members have interesting thing to share. ( most of the cpu side has been already discussed )

 

[groper]

For the next xgpu i agree with the DX11/SM6.0 estimations on the presupposition of a late 2010 release date.
If we keep getting a new shader model every 2 years and a new DX api every 4 years & if xbox3 will come out close to Novemeber 2010 then the situation will be shaped rather like this :

S.M 2.0 = DECEMBER 2002 (DX 9)
S.M 3.0 = AUGOUST 2004

S.M 4.0 = DECEMBER 2006 (DX10)
S.M 5.0 = AUGOUST 2008

S.M 6.0 = DECEMBER 2010 (DX11) XBOX3

Of course there is always the probability the next xbox to come out late 2009 instead of late 2010. In this case i think that it will have a full SM 5.0 gpu with some SM 6.0 extensions.
For the next playstasion i dont know what to think/suppose because Sony tend to be always unpredictable :smile:

Now speaking for photorealism or what the casual gamers believe as p/r.
The only sure think is that If with an early SM4.0 game we can come that close to p/r like those screens demostrate then i 'm surelly pretty excited for the level of realism that we can expect from the next consoles with their sm6.0 capabilities.


For ps3/xbox360 CPU's i will rather give them the next year to show me what they can achieve in their current form before i start praying for a next omg 50 core Cell cpu.

 

[rounin]

For ps3/xbox360 CPU's i will rather give them the next year to show me what they can achieve in their current form before i start praying for a next omg 50 core Cell cpu.

Agreed.

 

[Acert93]

http://www.beyond3d.com/forum/showthread.php?t=31379&highlight=console

We discussed this a couple months back. All my old predictions in there... already seems to need an update!

 

[Graham]

Ok here are a few things I believe may well happen.

A shift to higher level libraries + managed code, and a better form of data storage.

I see a library that sits well atop directX 10.2 (or whatever is out by then), which is optimised to absolutely scream on the hardware, but still be easy to use. Doing things like motion blur, material combinations, tone mapping, even a simply blur filter is still really not that much easier than it was 5 years ago. And because of that, there is a lot of room for unintentional inefficiency. A graphics lib that does this out of the box would be great. Other areas like audio, networking and input have all slowly become higher level, yet graphics not so much. XNA is an example, but I still believe they could have done a much better job hiding the internals...

I've said this one before, but I'd expect the next xbox have 'first class' support for something like the Polyphonic C# (now Cω) language (or whatever it turns into).

DVDs need to be thrown out for good. The combination of poor read speed, hardware reliability and seek times really doesn't help.
What would be the point of a machine with 4gb of memory if the drive only reads at 15mb/s?

 

[Rangers]

DVDs need to be thrown out for good. The combination of poor read speed, hardware reliability and seek times really doesn't help.
What would be the point of a machine with 4gb of memory if the drive only reads at 15mb/s?

What do you propose takes it's place? Mind you 5400 RPM HDD's (what is used in consoles) are not much faster.

 

[Acert93]

Digital distribution. Also, and I think this is a shoe in, a tiered memory system. Something like

System Memory (~ 2-4GB)
Flash memory (~ 20GB)
HDD (~ 200GB)
Optical Game Disk

You buy a game, and much of the common content gets cached, like on Xbox1. When a game is in use, the Optical Disk could have audio plus any information that is not yet in use and has a high tolerance for loading (e.g. things outside your visual distance) and the Flash memory is a "faster" pool of assets that cannot be streamed from the optical format but are not in immediate use by system memory. Kind of like L1/L2/System memory configuration, but for content. This is the general idea behind virtual memory, but with the addition of a faster intermediate step between the HDD/Optical storage and the system memory.

 

[nonamer]

Flash RAM is slower than a HDD at streaming data.

 

[Graham]

Digital distribution. Also, and I think this is a shoe in, a tiered memory system. Something like

System Memory (~ 2-4GB)
Flash memory (~ 20GB)
HDD (~ 200GB)
Optical Game Disk

You buy a game, and much of the common content gets cached, like on Xbox1. When a game is in use, the Optical Disk could have audio plus any information that is not yet in use and has a high tolerance for loading (e.g. things outside your visual distance) and the Flash memory is a "faster" pool of assets that cannot be streamed from the optical format but are not in immediate use by system memory. Kind of like L1/L2/System memory configuration, but for content. This is the general idea behind virtual memory, but with the addition of a faster intermediate step between the HDD/Optical storage and the system memory.

I think I posted exactly that in the last thread of this nature Just not so detailed

But I'm not going to predict what will be fast and economical enough in 5 years. Just look at DS games on SD like memory cards (upto 128mb afaik?).

There definitely will be digital game distribution I'd expect. Then maybe installing games via optical disks or some other media. Still I don't think running games directly off a disk is going to be workable in 5-10 years.

But hey, my opinion of course

 

[Fran]

After looking at screenshots from the Cryengine 2, im more confident than ever, that a dream that atleast i have dreamt about since i was a little kid, can be achieved, that small dream is photo-realistic games.

Do you really want photorealistic games? I mean, I can see photorealism 24/7, I'd like to play and feel something different, something I can't see in real life, an environment which is not real, merely believable, but which gives emotions that can't be felt in real life.

Photorealism is boring.

 

[DuckThor Evil]

Do you really want photorealistic games? I mean, I can see photorealism 24/7, I'd like to play and feel something different, something I can't see in real life, an environment which is not real, merely believable, but which gives emotions that can't be felt in real life.

Photorealism is boring.

Yep, but you can't do everything in real life even though you can see it. I think what he meant by photorealism is that the quality of craphics is as good as it will get basically, instead of that he want's to play accounting
I think by the time our games are something like the Matrix, the overall productivity of human species will go down a bit, since nobody gives a shit anymore :smile: Let's hope that by then we have already enslaved those robots to do our work for us...

 

[Shifty Geezer]

Do you really want photorealistic games? I mean, I can see photorealism 24/7, I'd like to play and feel something different, something I can't see in real life, an environment which is not real, merely believable, but which gives emotions that can't be felt in real life.

Photorealism is boring.

I think rendering needs to be able to be photorealistic to enable the fantastic to appear more realistic. Then artistic merit will reign supreme and without limits. A racer like GT or PGR wants to be photorealistic. An RPG might well not want to, especially one with characture type characters. Taking A Bard's Tale, there were some ugly charactures in that game which worked because they were obviously art. A realistically modelled and rendered person with those faces and proportions would be an eyesore! For something like Fable, photorealism in the scenery would help create a more immersive environment and make it more believable. Then fantastic vegetation won't look like an artist's idea, but a real plant.

I guess photorealism as desired really applies to lighting in the main. What's really wanted is realtime GI. When we have realtime GI, everything will look solid and properly composed. The shading style and textures etc. will then just be art choices. Take the CG renders for the latest Pokemon and Motostorm. Both have very different art styles, and both used copious amounts of advanced lighting. Both are the better for it then the real games. That photorealistic lighting applied to a non-photorealistic world like Pokemon is far more desirable then sticking with the limited current options.

 

[doob]

Software aside, regarding hardware manufacturing processes, more exactly in the micro-electronics area(cpu/gpu's), current technology is reaching its limits, im not a electronics engineer or anything near that field, but from what i'v been reading 32nm is pretty much the end of the line, and at this pace, look at Intel, they are already at 65nm, 45nm in end 2008, so in 5 years time 32nm products will much likely be on the market, so where i want to reach with this is, in 8-10years time what will companies like Intel, IBM, AMD etc will turn to? Quantum manufactoring techniques? Doesn't seem very likely from what i'v read, unless some breakthrough happens, one thing is almost guaranteed, we can pretty much expect cpus and gpu's being manufactured at around 30nm or slightly less. So with that in mind, i believe more educated people can extrapolate what can be achieved with that manufacturing process.
Anyway, i wish i am wrong about this, and it wouldn't be non-sense, and quantum electronics would be commercially avaiable at that time.

 

[TheChefO]

From what I've read on multithreaded game code it seems there is a point in which it becomes cost ineffective to add more "cores". My question would be to the devs: More ram or processor power?

Obviously in 5 years both will improve but would it be more effective to have 2 cells at 5ghz and 6gb ram or 6 cells at 5ghz and 2gb ram for a 2010-2015 console?

I would think devs would want more ram as it enables larger more varied environments and while it would come at a cost to processor power I would think the trade-off would be worth-while.

Am I wrong?