Predict: The Next Generation Console Tech

[Shifty Geezer]

you may think I am putting the cart before the horse but shouldn't we be discussing first what features the PS4 should have THEN the hardware to support those features. This is apparently what Sony did with the PS3.

I don't think that viewpoint sheds any light on the future. What console won't come with the latest HDMI or USB? And since when have chip considerations been dependent on IO devices?! Choice of CPU and RAM configuration and all the rest is independent of the pieces of the user experience you are talking about, as programmable hardware of a basic minimum spec, well within the limits of any budget for a next-gen machine, can turn its hand to any such tasks. This thread is more about the choice of CPUs, GPUs, RAM, buses, maybe specific hardware rendering modes, etc., and not the choice of IO ports.

 

[Danalys]

I don't think they'll even need that much better a CPU for PS4. A better graphics card can offload some of the slack it has to pick up and there's not much more that needs to be done in program control or physics or animation. All the things games are missing require massive amounts more processing power than is reasonable to expect any time soon.

I think GPUs will hit a wall within the next 5 years aswell, where going noticibly beyond what they can produce takes far more horse power than the next generation could provide. Once you've got super high res textures and displacement maps, a good enough GI and subsurface scattering approximation, where else is there to go that's the next step?

 

[corduroygt]

Some more general purpose CPU power in PS3 would be nice. Not only it makes better multiplatforms, but general purpose tasks like using the OS and browsing the web would be faster too. Or better, make the SPU's configurable as only SPU's (no cache coherency but blazing fast) or general purpose CPU's with cache coherency...If you can make such a block small enough, then it's just a matter of how many you can fit within your area budget.

 

[jeff_rigby]

I don't think that viewpoint sheds any light on the future. What console won't come with the latest HDMI or USB? And since when have chip considerations been dependent on IO devices?! Choice of CPU and RAM configuration and all the rest is independent of the pieces of the user experience you are talking about, as programmable hardware of a basic minimum spec, well within the limits of any budget for a next-gen machine, can turn its hand to any such tasks. This thread is more about the choice of CPUs, GPUs, RAM, buses, maybe specific hardware rendering modes, etc., and not the choice of IO ports.

Again, I can't seem to get this across. My first example; Realtime face, voice, Gesture and body recognition, will require massive CPU time and memory. Currently the schemes by both Kinect and Move cheat to reduce CPU and memory issues. IF you try to use one CPU (even with multiple CPU elements) and memory pool in a game console you will severely decrease the ability of the system to process anything else. With PCs and Game consoles we already see graphics memory being separate from main CPU memory. A logical extension is to have real world processing with it's own memory and CPU processor to reduce the load on main memory.

Graphics GPUs already multi thread/parallel process, extending that we might have more discrete/separated processes for graphics with separated memory like the 10 meg of memory in the Xbox GPU, except slightly larger with at least two pools to support 1080P 3-D and 4K resolution.

Using the Gigabit network port or wireless to allow Video to be displayed on any media player in the house is coming, Distributed processing can come with that.

Someone needs to do some research and try come back with what might be considered "standards" for CE devices in 5 years. Then a discussion of what it would take to support those "standards" is reasonable. Until then we are talking about what is the fastest cheapest CPU not how do we create a next generation Game Console. I.E. in my opinion the discussion so far is too limited.

 

[MarkoIt]

. Once you've got super high res textures and displacement maps, a good enough GI and subsurface scattering approximation, where else is there to go that's the next step?

There is always a next step, a way to improve things. The problem is the diminuished return you get. It will be always harder to get closer to reality, for a little advance in the look of the game. It will take few decades to get there, but i think the final step is virtual reality.

 

[Shifty Geezer]

Again, I can't seem to get this across. My first example; Realtime face, voice, Gesture and body recognition, will require massive CPU time and memory.
...
Until then we are talking about what is the fastest cheapest CPU not how do we create a next generation Game Console. I.E. in my opinion the discussion so far is too limited.

These consoles aren't research projects designed to enable a futuristic task. They are mass-consumer commodities. If the processing power required is as big as you seem to think, then the cost of including that processing hardware would be prohibitive. Everyone else is working here from the concept of a hardware budget, and not a feature wishlist, with good reason - that's ultimately the deciding factor. Unless you're Nintendo, where you pick a novel feature and build the cheapest possible system that could support it!

And for the record, I think you're wrong about your estimation with processing power requirements. XB360 can already achieve realtime face, voice and body recognition. A conventional progression of performance will see higher-fidelity tracking next-gen without needing a greater than conventional increase in processing performance; those task you describe fit in with the sorts of system hardware we are expecting based on budgets.

 

[jeff_rigby]

These consoles aren't research projects designed to enable a futuristic task. They are mass-consumer commodities. <1> If the processing power required is as big as you seem to think, then the cost of including that processing hardware would be prohibitive. Everyone else is working here from the concept of a hardware budget, and not a feature wishlist, with good reason - that's ultimately the deciding factor. Unless you're Nintendo, where you pick a novel feature and build the cheapest possible system that could support it!

And for the record, I think you're wrong about your estimation with processing power requirements. XB360 can already achieve realtime face, voice and body recognition. A conventional progression of performance will see higher-fidelity tracking next-gen without needing a greater than conventional increase in processing performance; those task you describe fit in with the sorts of system hardware we are expecting based on budgets.

<1> CPU and memory TIME not processing power. That TIME necessary to process the video or Audio and compare to templates would bog down memory access for other functions. A simple cheap CPU or one SPU and PPE could do this provided it had it's own memory.

The XB360 just now with optimized code switched from 320 to 640 Horiz resolution. It can not use a 720P or 1080P camera for motion tracking just as the PS3 can't. Part of the issue will be corrected with USB 3, part would require more memory and processing time. I believe it would be more efficient to do the real world processing as a separate process with it's own memory.

We are not thinking outside the box, read console. Nintendo did this with the Wii controller showing that the INTERFACE is actually more important to consumers than the performance of the box.

The next generation box does not need much more CPU processing performance, it will need a better interface with the user and higher resolution. It will need a better set of features for use with the Digital ecosystem that is coming from all CE manufacturers.

As usual, everything you state is a valid concern....I'm speaking to the subtle differences that a step back from current designs and a look at what is coming to the CE industry means for game consoles.

Glassless 3-D would require the TV to shift (optically) the 2 images coming from the TV for each separate pair of eyes in the room knowing angle and distance from the screen. To do this would require input from at least a high res version of a Kinect and an optical scheme I can't even imagine. Yet it seems that writers are predicting such a system within 10 years.

LG is offering a remote for their 2011 TVs, Blu-ray and set top boxes that is essentially an Air Mouse that uses a subset of the hardware in the PS3 Move. They have a 10 foot UI and offer a Webkit browser and Apps; this in TVs NOW. This points to the future of the CE industry as does the Sony Snap program which provides for 3rd party applications which will include Home automation.

Samsung has device pairing for blu-tooth and WIfi made easier by allowing automatic connection if the devices are within 1 foot of each other.

Another article mentions CONFIRMED to developers from Sony, the PSP2 will have the power of the PS3. You as I are skeptical of this claim.

http://playstationlifestyle.net/2011/01/13/psp2-to-rival-ps3s-power-could-be-out-by-october/

Sony is set to aggressively increase their market share of the handheld market this year, rumored to be launching two portable devices in 2011. Along with the extensively leaked PlayStation Phone, the Japanese electronics company looks to release the successor to the PlayStation Portable later in the year. While many have predicted that the PSP2 would be a powerful machine, few would have guessed just how powerful the device will be.

Sources have told MVC that the PSP2 will be a true beast of a machine, with Sony apparently telling licensees that the device “is as powerful as the PlayStation 3″, however this may be in relation to the screen size, meaning that the actual power is less. Sony has been pitching the PSP2 as a portable equivalent of its next-gen home console, with similar games content planned – specifically asking for high-end games that differentiate its device from app-centric phones.

Contrary to some previous rumors, the PSP2 will support physical media, keeping retail in the loop – something that ensures that retailers will push the device, as well as supporting those without broadband. However, the device will also have plenty of downloadable content, including games and apps, with MVC even entertaining the notion that it might have phone functionalities.

Expected to launch in Q4, possibly even in October, Sony’s promotion of the handheld will be purposely timed to steal thunder from the rival 3DS. Not only is the platform holder thought to announce the device later this month, but “high-level sources” told MVC that Sony will outline its 2011 handheld strategy within a few days of Nintendo’s events covering their 3DS Easter rollout plans. More will be revealed at GDC and E3 later in the year.

 

[Squilliam]

And for the record, I think you're wrong about your estimation with processing power requirements. XB360 can already achieve realtime face, voice and body recognition. A conventional progression of performance will see higher-fidelity tracking next-gen without needing a greater than conventional increase in processing performance; those task you describe fit in with the sorts of system hardware we are expecting based on budgets.

Similarly you could argue the same applies to the OS functions of the Xbox 360 and PS3 relative to the performance of their previous generation consoles. Personally I don't think it is anything worth worrying about. Both the Xbox 360 and PS3 coped with a radical increase in processing requirements for non game functions. Heck even the Gamecube *2 did pretty well with a next generation interface.

 

[jeff_rigby]

Similarly you could argue the same applies to the OS functions of the Xbox 360 and PS3 relative to the performance of their previous generation consoles. Personally I don't think it is anything worth worrying about. Both the Xbox 360 and PS3 coped with a radical increase in processing requirements for non game functions. Heck even the Gamecube *2 did pretty well with a next generation interface.

We do not yet know how the PS3 and Xbox 360 will cope with non-game functions as most are yet to come. We have been speculating that One PPU and lack of memory may affect how the PS3 performs with a web browser. This may be a feature where the Xbox would shine with three PPUs and twice the system memory.

CE and cell phones have been driving optimized code and the PS3 can benefit from that.

 

[Squilliam]

We do not yet know how the PS3 and Xbox 360 will cope with non-game functions as most are yet to come. We have been speculating that One PPU and lack of memory may affect how the PS3 performs with a web browser. This may be a feature where the Xbox would shine with three PPUs and twice the system memory.

CE and cell phones have been driving optimized code and the PS3 can benefit from that.

Arguably they are rather unimportant OS tasks. Also Sony would hardly have much of an incentive to improve it since none of the competition bother to offer much in the way of a similar feature themselves.

 

[Blazkowicz]

Again, I can't seem to get this across. My first example; Realtime face, voice, Gesture and body recognition, will require massive CPU time and memory. Currently the schemes by both Kinect and Move cheat to reduce CPU and memory issues. IF you try to use one CPU (even with multiple CPU elements) and memory pool in a game console you will severely decrease the ability of the system to process anything else. With PCs and Game consoles we already see graphics memory being separate from main CPU memory. A logical extension is to have real world processing with it's own memory and CPU processor to reduce the load on main memory.

<1> CPU and memory TIME not processing power. That TIME necessary to process the video or Audio and compare to templates would bog down memory access for other functions. A simple cheap CPU or one SPU and PPE could do this provided it had it's own memory.

I really don't follow you there, I thought you would use, you know, an Operating System for that, complete with a thread scheduler and memory management.
You should be glad that someone found tasks to run on the multiple cores (and multiple thread contexts of those cores)

 

[jeff_rigby]

I really don't follow you there, I thought you would use, you know, an Operating System for that, complete with a thread scheduler and memory management.
You should be glad that someone found tasks to run on the multiple cores (and multiple thread contexts of those cores)

Higher resolution video processing requires MUCH faster GPUs because of the size of the video. We have in the past hit walls, for instance the PS2 GPU used a 128 bit register for video rather than 16 or 32 because silicon, at that time, could not run fast enough for 1080i video processing. The same issues are found in inputing high resolution video.

So silicon is now fast enough but the same issues that require separate memory for video buffers in GPUs (back buffer, front buffer) apply to inputing high resolution video and running comparison templates to determine movement (old video frame compared to new frame) and what was moved (frame compared to template of human body, face or voice).

You compound this with the real world happening in real time. When video is inputing from a camera to memory, it happens at it's own speed which requires a buffer memory at that speed and then a write at main memory speed from the buffer.

The cheat I mentioned before would have the color depth changed from 24 bits to 2 or 3 like a scanner converts from Jpeg (picture greyscale) to text mode which is either black or white with no graduations. This reduces the size of the image and the time spent in comparisons it can also introduce errors depending on clothing color. The "move" uses several cheats, it has input from gyroscopes as that is faster and more accurate than video recognition and uses a glowing colored orb so comparisons frame to frame is a simple color key comparison. All these cheats are because video processing is time expensive.

It's not that a main CPU can't do this it's that memory time is being hogged by these processes and conflicts with memory scheduling can result. The choice is to increase memory speed even more (which would be expensive) or preprocess video input with it's own processor and memory. A slower less expensive processor can do the video preprocessing. Since the camera has to input the video unmodified for video chat or overlays as in "Start the party" the preprocessing can't be in the camera.

I would expect that we will have 3-D 1080P or possibly 720P video cameras during the life of the PS4 and provisions to process two streams of video which when combined with going from 480P to 1080P will massively increase input video processing.

So even if you don't use a separate CPU and memory pool for video preprocessing you would have to make provisions for 1080P 3-D and gesture, facial and voice recognition video input and that would affect CPU and memory choice. You must determine the features you will support before deciding what hardware and CPU you will use.

 

[jeff_rigby]

Arguably they are rather unimportant OS tasks. Also Sony would hardly have much of an incentive to improve it since none of the competition bother to offer much in the way of a similar feature themselves.

If you limit the comparisons to game platforms then yes.....If you look at the CE industry then TVs now have features not in some game consoles (Xbox), do web browsing better than the PS3 and Wii and have apps - games and app stores. In the next two or three years before a new generation game console is released what will be the competitive driven CE industry (TVs and blu-ray players) features set?

Oh and both the LG and Sony 2011 TVs and blu-ray players feature 720P video conferencing in addition to web browsers and apps. The Samsung 2010 CE electronics feature Air 2.5 and apps.

"rather unimportant OS tasks" Perhaps this is why the Sony flagship PS3 blu-ray player does not yet have a modern browser and Sony -LG 2011 blu-ray players do. Shortly I will not be able to recommend the PS3 as the best blu-ray player/DLNA/picture and video viewer for consumers. It's falling behind.......and the other game consoles aren't even trying, there is no competition. Just in case you are thinking they are game consoles not CE platforms, CE platforms now have games.

 

[Shifty Geezer]

Higher resolution video processing requires MUCH faster GPUs because of the size of the video.

4x the resolution requires 4x the processing power, unless you're using something like an O(n^2) algorithm.

Silicon is now fast enough but the same issues that require separate memory for video buffers in GPUs (back buffer, front buffer) apply to inputing high resolution video and running comparison templates to determine movement (old video frame compared to new frame) and what was moved (frame compared to template of human body, face or voice).

You're kinda talking twaddle now. A 1080p frame is ~2 million pixels. At 24 bits RGS, that's 6 megabytes a frame. 60 FPS would be 360 MBps bandwidth consumed for shifting through every pixel. A comparison between two 1080p60 framebuffers would be 720 MBps. PS3 has >45 GBs. Next gen will have more. A couple of GB/s from the total RAM pool is no great loss, and certainly nothing needing a whole extra memory system as you are suggesting.

Multitasking OSes for decades have been able to run concurrent, independent task on the same RAM and processors by shifting tasks on the fly. Video processing is no different. The current Kinect PC demos are doing exactly that, using the same processor to evaluate the Kinect data, and then run whatever tasks are happening concurrently.

The cheat I mentioned before would have the color depth changed from 24 bits to 2 or 3 like a scanner converts from Jpeg (picture greyscale) to text mode which is either black or white with no graduations.

Again, this isn't at all accurate. You'd either have a posterized image losing all the information that denotes objects, or you'd have to dither it, making it nigh impossible to do optical processing on. And it'd still look like crap. If you're using the video feed in game, you'll need the full colour image. Good optical recognition wants as little noise and as much information as possible. JPEG compressing a video stream is bad enough, let alone throwing away most of the image information! And it's uneccessary. Future ports will be able to cope with higher camera resolutions. Maybe they'll be limited to 720p. Regardless, that's all covered by the general IO choices of the console and don't need any special attention, unless you feel a brand new port needs to be designed specifically for high-speed cameras because the likes of USB3 aren't up to it.

It's not that a main CPU can't do this it's that memory time is being hogged by these processes and conflicts with memory scheduling can result.

No different to every other system out there. We don't break PCs up into a processing component with CPU and RAM for audio, another for video, another for physics, another for browser, etc. We take one pool of resources and use it dynamically.

The choice is to increase memory speed even more (which would be expensive)...

Again, you're looking at a few GB/s maximum. In systems with likely well in excess of 50GB/s, that's not a problem that needs special attention.

I would expect that we will have 3-D 1080P or possibly 720P video cameras during the life of the PS4 and provisions to process two streams of video which when combined with going from 480P to 1080P will massively increase input video processing.

An 8x increase in needs will match a natural 8x increase in performance that comes with the next generation of console hardware. The impact will be no more than the current requirements are on this gen. XB360 wasn't designed with a memory and processing subsystem for a future 3D camera. Instead the Kinect works by using a fraction of the system's available resource pool, with no song-or-dance complications about it crippling the running of other applications because it's getting in the way of their memory accesses.

So even if you don't use a separate CPU and memory pool for video preprocessing you would have to make provisions for 1080P 3-D and gesture, facial and voice recognition video input and that would affect CPU and memory choice. You must determine the features you will support before deciding what hardware and CPU you will use.

Only if the features are specialst. Everything you say can fit into the possible processing choices we've outlined before in this discussion. Only if you are doing something extraordinary that conventional processors can't cope with (in the same way a 2005 tri-core PPC and GPU can't cope with 2010 cutting-edge 3D vision tracking) would you need to consider extraordinary CPU soltuions, but that would be cost prohibitive meaning you'd drop that feature and go for a lesser one that works within budget-constrained hardware choices. Wii didn't include a gyro when it launched, even though that'd have provided the full features Nintendo wanted, because the cost was too high. They reduced features to match a price target. Next-gen consoles will have a CPU and GPU made to a price, and the interface options will be built around those knowing that they'll consume a small fraction of resources.

 

[Tahir2]

Question:

If you had only one choice what would you prefer the extra processing power was spent on in the next gen consoles?

1. Higher graphics fidelity (more effects, polygons and schwizz bang wow fx) @ 1080p

or

2. Same level of graphics fidelity but higher resolution (4k screens etc) @ 2160p (ish).

I know what I would prefer...

 

[Blazkowicz]

I would like very much a 720p output with perfect anti-aliasing, perfect AF and perfect framerate then working out the better effects within such constraints

 

[Tahir2]

I would like very much a 720p output with perfect anti-aliasing, perfect AF and perfect framerate then working out the better effects within such constraints

You would think and hope that by 2011/12/13* the collective might (money, experience, partners, advancement in manufacturing process etc, etc) and desire to +1 each other, MS and Sony would be able to figure out a way to make a console that can render consistently at 1080p60 with gobs of AA and AF.
I thought that was a given but then again maybe I am being naive.

* delete as appropriate, chances of a 2011 launch are very slim though and with Kinect doing wonders for Xbox360 maybe 2013 for MS.

 

[Blazkowicz]

well it will depend on games and developer choices.
I believe a 2013 console will be less powerful than current graphics high end. that's still an enormous amount of power with the advantage of a single target of course.

 

[Nesh]

you may think I am putting the cart before the horse but shouldn't we be discussing first what features the PS4 should have THEN the hardware to support those features. This is apparently what Sony did with the PS3.

It probably should have the following:

Realtime face, voice, Gesture and body recognition.

Video 4K resolution with 1080P 3-D games.

Audio, 7.1

HDMI 1.4 (Full implementation)

Gigabit Network port

USB 3.0 ports

I believe they are going to aim for multitasking abilities (run two or more applications simultaneously) since the future is to provide a media and entertainment device for all needs. Sony must have been aiming for that from the beginning since the PS3 was supposed to be able to display on two different displays (2 HDMI's) and do different tasks on each originally but the resources werent there. Since they want to satisfy all kinds of digital needs, you wouldnt want one user to prevent someone else from using another function when he wants to play a game watch a movie etc etc

Once you've got super high res textures and displacement maps, a good enough GI and subsurface scattering approximation, where else is there to go that's the next step?

Extremely detailed physics, macro and micro scale physics, animations etc. After you get all the detail you d want the right physics and animations to accompany it. You want materials to behave like their real ounterparts, you want different mass for each material, you want material fillings, you d want one material to affect another naturally etc etc.

Currently even the most detailed game visually like Crysis loses from its immersion when some part of the environment feels dead and unafected by any kind of force or interaction.

 

[Danalys]

I think the two HDMI ports were there initially for 3D. HDMI 1.3 made that unnecessary. Rendering on two screen was just something else they could have done with 2 but wouldn't have been used much.

We've seen things like digital molecular matter already. So getting most physics down without too much more processing power required is possible. Things wont be quite right but getting them exactly right will take far more power and completely different ways of modelling.