• Hey Guest. Check out your NeoGAF Wrapped 2025 results here!

GPGPU Computing & why you should be more excited about Kinect 2 & the Next PS-Eye.,

onQ123

Member
with the new generation of consoles the GPU's will more than likely be GPGPU that can be used to do some amazing things with Image Processing , Augmented Reality , Gesture Recognition , Face Recognition & more.

add to that, the fact that the cameras should be higher resolution at higher frame rates next gen because they won't be held back by the limitations of USB 2.0 & the consoles will also have better CPU's, so most of the shortcomings will be fixed this time around, it's hard to imaging the PlayStation Move tracking being much better than it is now, but with a HD PlayStation Eye that's made around tracking the controller things could only get better.


side note: I'm typing this as I fall to sleep so 1/2 of this might sound crazy when I wake up & read it tomorrow so hopefully it makes sense to everyone else.


Edit: I think we should just make this the GPGPU for Consoles thread
___________________________________________________________________________________

Bumped for the Wii U situation:


The Wii U has a weaker CPU than expected, but that's not as big of a problem as most people might think. because with the GPGPU taking over some of the jobs that used to take a lot out of the CPU it's no longer required to be as powerful as CPU's once was in Consoles.


http://www.neogaf.com/forum/showpost.php?p=42142347&postcount=2837

Froblins - AI simulation, tesselation, & rendering, all entirely on the GPU - SIGGRAPH 2009

ATI - March Of The Froblins (Radeon HD 4800 Series)


could Next Gen see 1 of the biggest improvements between console gens do to GPGPUs?
 
hteng said:
does that mean i can live in the future now?


well depending on the Camera Resolution & how good the CPU/GPCPU is Kinect 2 might be able to scan a 3D model of you into the game as the avatar.


shagg_187 said:
You're not Jeff Rigby.

& you're not Wario64!
 
onQ123 said:
well depending on the Camera Resolution & how good the CPU/GPCPU is Kinect 2 might be able to scan a 3D model of you into the game as the avatar.
I think most people would settle for it differentiating between them and the lampshade on the coffee table.
 
what are they going to do to solve the latency problem though? it seems like the fact that there is image processing that there will always be an input lag? I suppose they can make it better by having a very tight loop for games where it matters but I would think with the move controller that it is a dead end for hardcore games until they create something that doesn't rely on a camera. i can see where the kinect would improve as there is more interpretation involved and throwing more at it would lead to less errors in recognition but games with aiming are still going to feel sluggish
 
hteng said:
latency is a big issue in kinect type controls.

That is why Op is saying Higher framerates.
But if im right kinect right now input should be about 1 frame(33ms) behind games not using it.

DeaconKnowledge said:
Microsoft is already out of ideas with current technology. I doubt that advancement of the core tech's power will change hat.

There was that really cool Touchscreen everywhere tech research video. The video was Released a couple of weeks ago.
http://www.youtube.com/watch?v=Pz17lbjOFn8
 
Shai-Tan said:
what are they going to do to solve the latency problem though? it seems like the fact that there is image processing that there will always be an input lag? I suppose they can make it better by having a very tight loop for games where it matters but I would think with the move controller that it is a dead end for hardcore games until they create something that doesn't rely on a camera. i can see where the kinect would improve as there is more interpretation involved and throwing more at it would lead to less errors in recognition but games with aiming are still going to feel sluggish


that's the point of having a faster connection to the Console USB 3.0 or Thunderbolt or whatever , higher FPS Camera & a GPGPU that will process the image data faster.
 
I think the skeleton tracking in kinect software already uses GPGPU. Or at least the publications about their pattern recognition method mentions performances when running on a GPU.

Shai-Tan said:
what are they going to do to solve the latency problem though? it seems like the fact that there is image processing that there will always be an input lag? I suppose they can make it better by having a very tight loop for games where it matters but I would think with the move controller that it is a dead end for hardcore games until they create something that doesn't rely on a camera. i can see where the kinect would improve as there is more interpretation involved and throwing more at it would lead to less errors in recognition but games with aiming are still going to feel sluggish

There are several causes of latency, many of them have ways of improvement :

- data transfer : this is the most limiting right now, and the easiest to correct. Right now kinect has to send several images through USB2. A simple upgrade to a faster protocol would reduce the latency.

- image processing : that one depends directly on the complexity of the process, and the power of the hardware (and the way it's used, like with GPGPU). The current software already does a good job tracking the user in real time, better hardware will make it even faster. And the base software can improve too, as the latest update of kinect SDK said it (they announced a 20% performance gain).

- gesture analysis : that's the most tricky one, since it depends on the way you design your software. If you create a rule like "make a throwing gesture to launch a grenade" and implement it the regular way, you will have to first capture the gesture, then recognize the throwing motion, then trigger the on-screen throwing animation. So your character will start throwing after you finished the move, hence the latency. To avoid this problem, gesture analysis should be done "on the fly", and character animation be less event-based if you want to match the user motions.
 
Goddammit man. Of course I'm excited about next gen interface technology.

So why isn't there an article on cool next gen interface technology in this thread??


Seriously though, camera based motion control will really come into its own when paired with head tracking HMDs and cameras mounted on said HMDs.

The whole UI and computer interface game will change - screen size is no longer the bounding box for user interface - you can have elements of a virtual desktop placed around your person in an even more natural fashion than actual objects (free floating, weightless, suspends where you left it last).

And of course, kinect style motion tracking will facilitate the users interaction within such a true 3D virtual space.
 
It doesn't matter if the hardware gets better because most developers will simply just ignore them (at best, you'll get some throwaway unwanted shovelware from them)
 
Nuclear Muffin said:
It doesn't matter if the hardware gets better because most developers will simply just ignore them (at best, you'll get some throwaway unwanted shovelware from them)

well none of this will make a difference to the devs that's going to ignore it but for the ones that put it to use it will matter.
 
Bumped for the Wii U situation:


The Wii U has a weaker CPU than expected, but that's not as big of a problem as most people might think. because with the GPGPU taking over some of the jobs that used to take a lot out of the CPU it's no longer required to be as powerful as CPU's once was in Consoles.


http://www.neogaf.com/forum/showpost.php?p=42142347&postcount=2837

Froblins - AI simulation, tesselation, & rendering, all entirely on the GPU - SIGGRAPH 2009

ATI - March Of The Froblins (Radeon HD 4800 Series)
 
No, I'm not excited about Kinect2/Move2 at all.

IMO, the only useful feature from kinect is voice recognition, I hate navigating menu's through gestures or jumping up and down like an idiot.

I do have interest in VR, but imo, it will be ready for primetime when they can deliver it for under $199 in the form factor of the Google Glass prototypes.
 
I can't believe some people still haven't realized that "controller free" only works for a small subset of game-types.

The problem isn't latency, or sensor resolution, (if that was the case then Move would have represented a huge step over Wii-mote), its the lack of sensory feedback on action combined with the "always-on" nature of the interface that limits it.

How long is it going to take before the penny drops?
 
The problem with the reliance on GPGPU is that it's not cost-free, and any performance of the GPU spent on general processing is performance lost to graphics processing.

It makes some sense for some types of task to make this trade-off though as GPU's fly at particular jobs, but it does have to be remembered that their is a trade off, it's not magical pixie dust that just adds to the overall performance, just balances it and trades one type of task for another.

but it might be a good trade-off because from what I'm seeing in the rumors PS4 & Xbox Next will also be taking this trade-off.
 
Bumped for the Wii U situation:


The Wii U has a weaker CPU than expected, but that's not as big of a problem as most people might think. because with the GPGPU taking over some of the jobs that used to take a lot out of the CPU it's no longer required to be as powerful as CPU's once was in Consoles.

No, a weak CPU is still a weak CPU. You are with a GPGPU solution more flexible for different situations but it hurts the perfomance of your system.
 
No, a weak CPU is still a weak CPU. You are with a GPGPU solution more flexible for different situations but it hurts the perfomance of your system.

yeah but if the work is done better with a GPGPU & it take a lot of the load off of the CPU a console with a weak CPU can still survive with that weaker CPU thanks to the GPGPU.
 
I'll have to see more about wiiu's SDK before I can say anything more definitive but honestly, I don't think skimping on the CPU in favor of a GPGPU is a particularly great idea. There are still alot of issues that still really haven't been solved.

It probably doesn't matter much to Nintendo because their performance targets are probably alot more forgiving but still.
 
Sony should make and market a high end 3D 1080p 120hz VR headset (with the features of the Oculus Rift) alongside the PS4.

The Wii got a shit ton of hype from newspeople everywhere for offering something that had never been done on a mass market device before. Sony could get that same kind of glowing coverage for the PS4 at launch.


A Tablet isn't innovative, but VR is...

http://www.youtube.com/watch?v=61G07Z8aqlw

Sony does what Nintendont

Both gaming and video watching would be immersive as hell.
 
Regards the Wii U. Gimped CPU, lowish end GPU and dual screens may very well just cancel each other out in the end. I do think given what Nintendo seem to have done with the CPU and the dual screen nature of the system, they should have put more of a GPU in their or jut not done what they seem (we don't know yet) to have done with the CPU.
 
yeah but if the work is done better with a GPGPU & it take a lot of the load off of the CPU a console with a weak CPU can still survive with that weaker CPU thanks to the GPGPU.
GPUs are useful for embarassingly parallelizable problems that can be solved through stream processing, like physics and certain scientific calculations. And graphics rendering, obviously. They don't do so well with serialized code, or code with lots of branches and random memory access, which is typical in video game software. I think what we're seeing with PhysX is basically the ideal situation for offloading calculations to the GPU, but the CPU is still going to be the bottleneck if it's the weakest component in the system.
 
Sony should make and market a high end 3D 1080p 120hz VR headset (with the features of the Oculus Rift) alongside the PS4.

The Wii got a shit ton of hype from newspeople everywhere for offering something that had never been done on a mass market device before. Sony could get that same kind of glowing coverage for the PS4 at launch.


A Tablet isn't innovative, but VR is...

http://www.youtube.com/watch?v=61G07Z8aqlw

Sony does what Nintendont

Both gaming and video watching would be immersive as hell.

999 US DOLLAR incoming!
 
GPUs are useful for embarassingly parallelizable problems that can be solved through stream processing, like physics and certain scientific calculations. And graphics rendering, obviously. They don't do so well with serialized code, or code with lots of branches and random memory access, which is typical in video game software. I think what we're seeing with PhysX is basically the ideal situation for offloading calculations to the GPU, but the CPU is still going to be the bottleneck if it's the weakest component in the system.

The point that I'm making is that if a lot of the work is taking off of the CPU because it can be done better with the GPGPU it will free up the CPU to handle the work that it's good at by it's self.
 
The point that I'm making is that if a lot of the work is taking off of the CPU because it can be done better with the GPGPU it will free up the CPU to handle the work that it's good at by it's self.
If a game can't keep a steady framerate due to the complexity of the game's logic the GPU might not be helpful there. That's not something that can be offloaded to the GPU easily and efficiently. So it's not something that can be relied on in a general situation.

You kind of see this in Borderlands 2, where Nvidia owners can offload the particle effects to the GPU because they don't affect the gameplay, but even then, those with weak CPUs will still struggle to maintain a high framerate.
 
If a game can't keep a steady framerate due to the complexity of the game's logic the GPU might not be helpful there. That's not something that can be offloaded to the GPU easily and efficiently. So it's not something that can be relied on in a general situation.

You kind of see this in Borderlands 2, where Nvidia owners can offload the particle effects to the GPU because they don't affect the gameplay, but even then, those with weak CPUs will still struggle to maintain a high framerate.



you don't have to offload everything to the GPU but the fact that you can offload some things that would have been slowing the CPU down means that the CPU will be able to handle the other stuff that the GPGPU couldn't help with a lot better because it has less work to do.

why is that so hard to get?
 
GPGPU is just hype right now. How is it really much different from the SPEs in PS3?
With the use of OpenCL (or DirectCompute in Microsoft's case) I'm guessing it's quite easier to use rather than the DMA approach of SPU's (also kinda cool with the impressive power & sea of gpu cores these new consoles will offer).
 
With the use of OpenCL (or DirectCompute in Microsoft's case) I'm guessing it's quite easier to use rather than the DMA approach of SPU's (also kinda cool with the impressive power & sea of gpu cores these new consoles will offer).

Yep & GPGPUs are kinda the reason that Cell is pretty much dead because they are achieving the same goal.
 
Top Bottom