Predict: The Next Generation Console Tech

[liolio]

I didn't react much to the rumors Charlie unleashed, I happen to have some time

Spoiler

nice way to say I'm bored...

so let speculate (assuming there is truth to his talk that's it).
So the first chip A0 silicon is to come back to Redmond sometime during Q1 2012. It's a SOC.
Various thoughts:
1) I expect the chip the chip to be produce using IBM or GF 32nm lithography.
2) I believe that IBM lithography allowing for EDRAM is not ready @32nm, so if there is EDRAM on board MS would have to use their 45nm process. In addition I believe that this process may be costlier than their standard process => so see point one.
3) we can't expect a crazy high silicon budget, I believe that a +300mm² chip is a best case scenario. Somewhere between 250 and 300mm2 sound right (costs). I believe that next gen will be overall pretty costly to produce not only due the silicon/brain of the system, they have to pack kinect, a pad, +2GB of RAM, a unknown amount of Flash memory, a HDD and an optical drive and possibly a WiiU like controller. In this creepy economic climate I don't expect miracle in the upcoming two years, I don't see MS going for more than 499$ the the highest SKU, neither I see MS losing as much money as they did with the 360 at first (without taking in account RroD cost).
4) System will have to last and to be secured.

For the CPU.
*I believe that one implication for the CPU is that MS will want the max bang for buck out of the silicon they invest into the CPU (and the sytem overall). I believe that they will favor TLP (versus ILP) once again. So I believe that the chip will somehow resemble Xenon and Power A2/EN. It's the cheapest way to pack enough execution resources to grant the system some legs.
* I expect the integer pipeline and front end of the CPU to be reuse of the Power A2 design or close parents (In Order, 2 issues, 4 way SMT).
* I don't expect wider SIMD, I read (I can't find the source tho) reading that overall the amount of SIMD code in most workload (physic, AI, etc) is low, SMT is a good way to make most out of the SIMD units in those case.
* I also hope they will implement some support for scatter/gather, the same kind as described in some Intel white paper. It's supposedly cheap to implement (especially with pretty narrow SIMD 4-wide) and should grant a neat boost to SIMD units utility and utilization.

Spoiler

still I believe it won't happen...

* I expect them to go with 6 cores running at lower speed than xenon was (2.4GHz?).
* I expect 3 MB of shared L2, as Ms won't use EDRAM and won't get the benefit in power consumption and density, I believe 3 MB is the maximum amount to expect (same as a Core i3).
* I expect extensive clock gating as in Power A2.

For the GPU.
* I expect MS to have put its hand on AMD GCN architecture. I expect that the CPU and the GPU will share the same memory space, I also hope that it will be possible for the GPU to read and write from (or lock some parts of) the L2 as Intel gen6 is doing with the L3.
* I expect 3 " geometry engines" (not sure about how AMD is calling that) so the GPU can spit out an impressive amount of triangles. I expect triangles and tesselation to be the new GFLOPS in term of PR bullet point

Spoiler

be prepared for the next fans war...

.
* SIMD arrays in GCN seem to work by 4 (a scalar unit is shared by 4 SIMD arrays), I expect 12 of them running at conservative clock speed.
* I expect the system to be light on RBE.
* I expect the power consumption of the GPU to be capped.

Msic.
I wonder if MS could be interested to add some of the Power A2 networking and security engine, so Live security is really a first class citizen.

For the RAM.
I will rally Laa-Yosh professional POV, properly fill up to 2GB of RAM will be expansive enough. So I would expect 2GB of "not that fast" GDDR5 on a 128bits bus. (not sure DDR4 will have been considered if silicon is to get back as early as Q1 2012). So I assume +60GB/s of bandwidth.

For the optical drive, I wonder BRD player seems an obvious choice but x4 players are still not that cheap. Overall I would not be surprised if MS use some form of HD-DVD (so 15GB up to 30GB of storage). Ms is still pushing digital content distribution through Live and they may think it would help fighting piracy.

For storage, I don't believe that HDD will be standard this time again. On the other hand I expect the system to include 16 GB of flash memory (some reserved for the OS, caching, the rest available for DLC, save, etc.). I expect the system to support USB key for storage from scratch. I also expect MS to push proprietary BS for the HDD this time again...

Last point backward compatibility. I don't think Ms can pass on it, I'm not sure either it would be easy to emulate Edram with +60GB/s worse of bandwidth no matter the way more powerful RBEs, so I believe MS may simply pass on the headache and attached Xenos daughter die to the SoC (which would be not accessible to the devs so it can be removed later in system life).

------Huge EDIT--------- Got a look at llano review and realize I may have overstate what can possibly fit in the design.

 

[Shifty Geezer]

I think the PS3/Xbox 360 sold pretty well at the $3-400 price point so I don't see why they couldn't target a viable entry level console around that level next time with good sales.

Though true, as mentioned much, much earlier in this epic thread, cost reduction quite possibly won'y be as pronounced. Release cost has to keep on eye on expected cost reductions. It may be safer to release at a lower price fearing limited cost reduction than go for a $400 console hoping it'll come down to $99 in 5 years. As I see it, traditional die shrinks aren't yielding the returns that the past couple of generations saw, but at the same time we're getting new fabrication methods, so it's hard to call. Designing your console has to be more about ongoing price for its entire life, rather than just launch price.

 

[liolio]

Though true, as mentioned much, much earlier in this epic thread, cost reduction quite possibly won'y be as pronounced. Release cost has to keep on eye on expected cost reductions. It may be safer to release at a lower price fearing limited cost reduction than go for a $400 console hoping it'll come down to $99 in 5 years. As I see it, traditional die shrinks aren't yielding the returns that the past couple of generations saw, but at the same time we're getting new fabrication methods, so it's hard to call. Designing your console has to be more about ongoing price for its entire life, rather than just launch price.

I kind of agree, I can't see MS launching at more than 499$ for the highest SKU in this creepy economy. In the same time I don't expect the system to reach 99$ in time soon (even five years after release), there is a rampant inflation worldwide, I believe that prices are set to remain higher than that (as for this gen).

 

[fehu]

With RAM currently at $6/GB, 4GB seems an absolute minimum. The price of RAM is going to roughly halve every other year. It will effectively be free by the time next gen console sales maximizes.

I expect a mininum of 4GB RAM (<$25 cost at launch) and 32GB flash (<$45 at launch), but 8GB (<$50) RAM and 48 GB Flash (<$65) to be much more likely. Launch at $400, hit $300 at year one, $250 at year two and $200 at year three.

Cheers

the real problem isn't the memory price but the bus size complexity to accomodate the amount of chips

 

[erick]

I would disagree. Food for thought: The average RAM in 2001 was about 1 Gigabyte, and now, 10 years later, its between 6-8GB, not quite 8 times, about 7. In 1991, your average 286 had 1 Megabyte in it. Ten years later, 2001, the average had increased a thousand-fold. People in 2001 probably were guessing we'd have nearly a Terabyte in 2011... Our ability to use RAM doesn't even come close to keeping up with our ability to make more/smaller RAM. I would be suprised to see more than 32GB average by 2021.

On the main topic, signs and industry analysts point to this gen being less of a huge upgrade than those before it, I would expect 2GB as likely, 4GB a possibility, 8GB probably has a snowballs chance in hell. Same with anything Cell related, I'm sure Sony learned that lesson this time.

Well thought out, but your facts are inaccurate.

Intel 286 was launched in the beginning of 1982, and had 1MB RAM supported from day one. In 1983 Apple Lisa launched with 1MB RAM as standard.

Also, if memory serves, 1GB RAM became available to masses (in 2x512MB DDR configurations) in late 2004, and became high-end standard in 2005 with Athlon 64. So in fact this 1000-fold increase took about 23 years.

As for 2005 and onwards?

2005 = 1GB
2006,5 = 2GB
2008 = 4GB (in late 2008 this rose to 6GB due to i7 900 series launch with its triple-channel memory)
2009,5 = 8GB
2011 = 16GB

The correct extrapolation of RAM sizes from 2011 onwards should look like this (giving 18 months for doubling):

2011 = 16GB RAM (gaining traction with Sandy Bridge already, built 1st 16GB one back in March)
2012,5 = 32GB
2014 = 64GB (this is where the next gen consoles should hit the market, with 4GB?)
2015,5 = 128GB RAM
2017 = 256 GB
2018,5 = 512GB
2020 = 1024GB

So ... in this light 4GB for RAM in the next gen consoles is woefully obsolete, and even at 8GB it would be very hard for the console makers to justify the raging 56GB difference in RAM capacity by claiming "Windows overhead".

EDIT:

There is one perspective which would make even the 4GB make sense, sort of. That would be if consoles used ultra-fast GDDR5 (graphics RAM). Then, of course, all the RAM would be unified like in X360. At the moment we have 2GB graphics cards available (refraining from bringing in ultra-ultra high end). The extrapolation would then go like this, giving 2 years for doubling (since GDDR does it slower lately):

2011 = 2GB
2013 = 4GB (this is probably where the specs are laid down for the next gen)
2015 = 8GB (this is where it would launch)
2017 = 16GB (this is where mainstream gaming PCs would approach the point where they supercede console hardware at a reasonable price point, but still more expensive than consoles, while the high-end will be untouchable)

 

[liolio]

You're estimation are obscene
Most computer ship with 4 GB of RAM that's it, plenty of gamers don't go further than 4GB (stock).
What the point really to have 16 GB, for games? Do you expect MS to up RAM requirements for Windows 8?
Come on quit geek land and come back to earth, integrated graphic will become more and more relevant to PC gaming as they will grow the potential user base, do you think editors will waste this possible godsend by asking people to have 16 GB of RAM to run their games properly?
If you want a big PC stick to big PC not to mention how feasible it is, you can only stuck a given amount of memory chip to a bus of a given size.

EDIT By the way you could be right for the 4GB, it doable and kind of affordable but assuming the whole package companies will try to save every % of 1 penny they can. On top of it, if flash storage is standard and used for caching it could really help swapping the content of the RAM fast, look how shitty hardware in mobile device looks reactive thank to the solid state storage

 

[erick]

I never said that I was basing my extrapolation on what was "mainstream", just on when what amounts of RAM become possible (and financially feasible) in a gaming PC. Becoming available and becoming mainstream are two completely different things.

However, when a console launches with the idea of sticking around for the next 10 years, you do assume it will have the best of what is available at that time (my original thought when seeing the X360 spec sheet in 2005 - only 512MB RAM? But that's like cutting your own leg off!).

RAM is always the first thing you run out of, so smart people make it stick when the price is right.


Also, Windows 7 x64 is completely up to the task of using 12GB+ for system cache (who needs SSDs??)... or alternatively you could run your own MineCraft server with 80+ simultaneous users

 

[function]

I built an Athlon XP with 2 x 512MB in the middle of 2003. "High end" but hardly bleeding edge. 2 x 2GB is now the norm with 8GB being "high end" for serious things (and also for people who like larger numbers). Task manager is showing 8GB is doing nothing for me in games - or anything else.

Just because nerds can put large amounts of memory in a big box doesn't mean it's doing anything other diluting the cost effectiveness of the machine they're building. Console vendors don't have the luxury of ignoring this and stuffing their machines with specs that represents a poor investment.

It's not really relevant what PCs or phones or anything else in a different market is doing wrt memory, and won't be unless consoles are basing their success on ports from these systems with no allowances for memory models.

 

[liolio]

I never said that I was basing my extrapolation on what was "mainstream", just on when what amounts of RAM become possible (and financially feasible) in a gaming PC. Becoming available and becoming mainstream are two completely different things.

Indeed what possible is irrelevant and assuming a 128bits bus you won't be able to put more than 4 GB (or 8 I'm too lazy to dig the value they are available some pages ago and it varies with the type of RAM).

However, when a console launches with the idea of sticking around for the next 10 years, you do assume it will have the best of what is available at that time (my original thought when seeing the X360 spec sheet in 2005 - only 512MB RAM? But that's like cutting your own leg off!).

Sorry but the 360 will last 10 years and Ms have made money out it while becoming the leading on the US market for core gamers.

RAM is always the first thing you run out of, so smart people make it stick when the price is right.
Also, Windows 7 x64 is completely up to the task of using 12GB+ for system cache (who needs SSDs??)... or alternatively you could run your own MineCraft server with 80+ simultaneous users

The first thing manufacturers run off is money as they subsidize for hardware (sell at loss) and I expect them to continue to do the same (just to lose less). You should check the average conf for a gamers on stream while taking in account the Os overhead as well as how convenient extra RAM is if you run task in the background (when a game run good enough to allow that). On a console there will be possibly a 64MB OS and no multitasking. On top of it frame buffer will be lighter than on PC as I expect alternative top Msaa to gain traction.
Anyway point is console are supposed to be cheap to produce.

 

[Rangers]

On the main topic, signs and industry analysts point to this gen being less of a huge upgrade than those before it, I would expect 2GB as likely, 4GB a possibility, 8GB probably has a snowballs chance in hell. Same with anything Cell related, I'm sure Sony learned that lesson this time.

I think 4GB is possible, as we drag on 8GB becomes more and more likely. 2GB I doubt.

It's not enough of a leap, for one. With generations lengthening, we will need a big leap otherwise people wont bother upgrading.

Task manager is showing 8GB is doing nothing for me in games - or anything else.

Well I noticed the Crysis 2 hi res texture pack "recommends" 8GB RAM, which is kind of scary. First time the 4GB in my machine isn't enough.

Also as always I know DDR3 isn't necessarily relevant to consoles, but it's financially pretty easy to built a PC with 16GB. 16GB of DDR3 can be had for about $100 now. Hardly the most expensive part of a decent PC or even close.

 

[function]

However, when a console launches with the idea of sticking around for the next 10 years, you do assume it will have the best of what is available at that time (my original thought when seeing the X360 spec sheet in 2005 - only 512MB RAM? But that's like cutting your own leg off!).

The x1800 xt and 7800 gtx were both bleeding edge cards and had 256 MB of GDDR 3. Then Nvidia went and released the super duper mega expensive limited release 7800 gtx 512 ("previewed" around the time of the 360's launch).

The 360 was in mass production with 512 MB of GDDR 3 in the months running up to November 2005!

RAM is always the first thing you run out of, so smart people make it stick when the price is right.

Other things you can run out of: money, shareholder patience, hardware to sell to customers, time!

(Damn, liolio beat me to that!)

 

[function]

I think 4GB is possible, as we drag on 8GB becomes more and more likely. 2GB I doubt.

It's not enough of a leap, for one. With generations lengthening, we will need a big leap otherwise people wont bother upgrading.

I'd rather have a large area of memory, such as 8/16GB of 100MB/s SSD space, that could be used for buffering optical disk (or 2.5" HDD) or storing uncompressed data or whatever else developers want it for. A 6x Bluray drive is going to be a pretty slow way to put GBs of data into main ram, and streaming from such a slow device can only do so much!

What might be good is if the instant you put a disk in, even before selecting play in the dashboard, the console started to build up a virtual disk image at full speed (the data already layout optimised when it's burnt to disk). This way you could be ready to start using the virtual image quickly and benefit from reduced loading times, and in a worst case would only need to transfer the data directly from optical disk / 2.5 HDD once if the user broke from the predicted usage path or got ahead of the process.

Well I noticed the Crysis 2 hi res texture pack "recommends" 8GB RAM, which is kind of scary. First time the 4GB in my machine isn't enough.

There's always some crazy fool making hi-res textures somewhere!

On console Crysis these would be streamed into memory at a required point iirc, rather than loaded in one big, painful gulp.

 

[Shifty Geezer]

The correct extrapolation of RAM sizes from 2011 onwards should look like this (giving 18 months for doubling):

2011 = 16GB RAM (gaining traction with Sandy Bridge already, built 1st 16GB one back in March)
2012,5 = 32GB
2014 = 64GB (this is where the next gen consoles should hit the market, with 4GB?)
2015,5 = 128GB RAM
2017 = 256 GB
2018,5 = 512GB
2020 = 1024GB

Ha ha ha ha ha! Show me any ordinary person using even half their 4GB PCs for substantial amounts of the time! Just because RAM is that cheap, doesn't mean it serves a purpose. You'll notice rather than most people using 16GB systems now, most are prefering to use smaller netbooks and tablets. Form and functionality take preference, because all that RAM is mostly useless unless you ware working in superHD productivity, such as massive multilayer image processing or editing loads of concurrent HD streams. Actually, that latter doesn't really work, because if those streams exceed your RAM capacity, they'll have to be streamed anyhow.

I look forward to 2020 and the first game to require 1 TB of RAM for fitting all its content! I hope Laa-Yosh gets the job of managing the asset creation - he'll relish working with teams of a 1000 artists!

 

[erick]

I built an Athlon XP with 2 x 512MB in the middle of 2003. "High end" but hardly bleeding edge. 2 x 2GB is now the norm with 8GB being "high end" for serious things (and also for people who like larger numbers). Task manager is showing 8GB is doing nothing for me in games - or anything else.

Obviously my memory did not serve in this instance. Indeed, 1GB was possible in the summer of 2003: http://www.anandtech.com/show/1138/4

Just because nerds can put large amounts of memory in a big box doesn't mean it's doing anything other diluting the cost effectiveness of the machine they're building. Console vendors don't have the luxury of ignoring this and stuffing their machines with specs that represents a poor investment.

It's not really relevant what PCs or phones or anything else in a different market is doing wrt memory, and won't be unless consoles are basing their success on ports from these systems with no allowances for memory models.

Now you're applying PC-building logic to console building. Cost-effectiveness to the extreme, regarding key system components, is a nice thing to have if you can easily upgrade, but it's the wrong way to go if you want your box to last 10 years (and put out nice graphics while doing so).

This kind of thinking is why there are practically no native Full HD AAA games. X360 would be perfectly capable of proper Full HD (it's called using the 10MB daughter die in tiling mode I think, which helps to get over the 720p 2xAA limitation most heavy 3D games face), but what's the point if the textures will look simply awful due to RAM restrictions?

The decision to put 512MB in X360 instead of 1GB in the hopes of saving something around $15 per unit was essentially what cost MS the ability to do Full HD AAA games, in my opinion (arguments welcome). One would hope they learned from it.

EDIT:

Ha ha ha ha ha! Show me any ordinary person using even half their 4GB PCs for substantial amounts of the time! Just because RAM is that cheap, doesn't mean it serves a purpose. You'll notice rather than most people using 16GB systems now, most are prefering to use smaller netbooks and tablets. Form and functionality take preference, because all that RAM is mostly useless unless you ware working in superHD productivity, such as massive multilayer image processing or editing loads of concurrent HD streams. Actually, that latter doesn't really work, because if those streams exceed your RAM capacity, they'll have to be streamed anyhow.

I look forward to 2020 and the first game to require 1 TB of RAM for fitting all its content! I hope Laa-Yosh gets the job of managing the asset creation - he'll relish working with teams of a 1000 artists!

1. Like it was already mentioned, there are mods out in the wild that can and will use more than 4GB. Or maybe even 8GB.
2. It is easier to make high-fidelity graphics from the start than to spend countless hours shoehorning your nice original hi-res textures into a small memory footprint while making sure they don't suck too obviously.

 

[liolio]

15$ x60millions units is 900.000.000 Dollars. Assuming sales will continue it will be a billion and more, even without the Rrod that alone may have suck a super consistent amount of the profit the 360 will have regenerated for MS.

 

[ShadowRunner]

After listening to a recent interview with Carmack i really am in agreement that the next big leap will be through display technology, specifically head mounted displays. Looking at the environments in games like say Skyrim or Rage the way we are veiwing games at the moment is the limiting factor, not the graphics themselves in many cases. People still scoff at VR as a pipe dream but i really do think that for next gen graphics and environments will be at such a level that i really think its time for someone to take another proper crack at it. Sony seem to be doing some work in this area, lets hope that they can come up with something good, the advancement in OLED display tech is promising for HMD tech.

 

[wco81]

People got dizzy and disoriented with those VR sets.

Just imagine people getting into car accidents after a VR session.

 

[erick]

15$ x60millions units is 900.000.000 Dollars. Assuming sales will continue it will be a billion and more, even without the Rrod that alone may have suck a super consistent amount of the profit the 360 will have regenerated for MS.

No no no It was $15 in 2005!

Remember, RAM costs roughly halve every 2 years. By now it should be something like 23 cents...

 

[ShadowRunner]

People got dizzy and disoriented with those VR sets.

Just imagine people getting into car accidents after a VR session.

It wont be for everyone, but then what is? The great thing about the display device is that it can be totally optional, but is trivial to implement support for if you already have 3D support.

 

[Shifty Geezer]

1. Like it was already mentioned, there are mods out in the wild that can and will use more than 4GB. Or maybe even 8GB.

Sure. If people want to spend their free time making mods, good for them. No company save a rare few (CryTek) is seriously going to invest in making masses of assets for a tiny market. Look at the RAM requirements of most actual games and it's nothing like 8 GBs, let alone the 16 GBs your talking about being a target in 2011.

2. It is easier to make high-fidelity graphics from the start than to spend countless hours shoehorning your nice original hi-res textures into a small memory footprint while making sure they don't suck too obviously.

The cost in creating assets doesn't come from simplifying them for low spec. It comes from building them. 100x the content is going to take 100x the effort. Sure, you can easily get a 4x increase in RAM requirement just by using 2k textures instead of 1k textures, but texture res will only get you so far. 1 TB of content is astronomical. It'll require at least an order of magnitude more expense on assets. I'll point you to the 50GB potential of BRD, and the fact that since the beginning of this gen, game sizes have remained in the DVD capacity range. What exactly are you going to distribute your 1 TB RAM consuming games on?? We'll spend two hours loading the game?? By 2020 we'll even be looking at streaming game like OnLive.