Do PS4/Xbox One GPUs Feature Enough Compute Units to Utilize Full Potential of Tiled Streaming?

GamingBolt speaks to Granite SDK creator Aljosha Demeulemeester to find out.

Posted By | On 06th, Apr. 2014 Under News


One term that often comes up with the next generation of consoles, especially regarding the Xbox One, is tiled streaming. Though the PS4 has turned out to be more powerful the Xbox One, thanks in part to its unified architecture and 8 GB GDDR5 RAM, it’s often stated that the latter’s eSRAM can make up for it with its propensity for tiled streaming.

Speaking to Graphine CEO Aljosha Demeulemeester, whose company helped create the Granite SDK for HD texture streaming, and strong compression, GamingBolt asked if the PS4 and Xbox One GPUs had sufficient compute units to realize the full potential of tiled streaming that Granite allowed for. For information purpose, the PlayStation 4 has 18 compute units and on the other hand Xbox One has 12.

Aljosha stated that, “Tiled streaming does not really tax the GPU much, unless you’re decoding on the GPU. Support for this is not active in our current Granite SDK release. We are considering it because game developers love the flexibility of being able to choose between CPU-side decompression and GPU-side decompression. On the CPU side, decoding our tiles on the fly actually doesn’t take up much processing power. We spent a long time optimizing this!”

The technology was demonstrated last year [can be seen in the video above] during Microsoft’s //build keynote using DX 11.2. It is a representation of what is possible with eSRAM and tiled streaming if used efficiently.

In the past, Granite SDK has been efficiently used by Larian Studios in Divinity Dragon Commander. The problem that Larian Studios were facing was that textures looked alright when seen up-close but when they were seen from the sky, the landscape textures lacked natural details like cracks and erosion. However bigger the textures, the more memory and storage budget you require and as one must be aware, memory is scarce these days.

Using traditional methods of Layer Blending, the developers were not able to successfully compress data to a level that they wanted. This is where Granite SDK comes in. Using the SDK, Larian Studios were able to upscale the 16kX16k landscape textures to 32kX32k and in some cases even beyond. This resulted into a memory usage reduction to around 66% in most cases.

We will have more on Granite SDK in the coming week. Stay tuned for more news and updates. What are your thoughts on the same? Let us know in the comments.


Awesome Stuff that you might be interested in

  • HalfBlackCanuck

    so…yes?

  • rurk

    everyone has long been clear that xbone is a piece of shi… and will fail than a game than a new boss

  • cozomel

    Man, fanboys just dont get it. PS4 can do tiled resources too, thats a fact. tiled resources is not some new fandangled technology, its basically hardware megatextures, somthing thats been around for over 10 years now. The only thing that is new, is its new to DX (DX11.2 to be exact). You could already do this on OpenGL and if you really wanted to, you could do it on any capable hardware. DX is just like a program where you turn turn certain features on or off. But just because the program doesnt have the option doesnt mean the hardware cant do it. This is nothing more than MS trying to sale you on bullsh*t. This isnt some special feat. Tiled Resources has nothing to do with eSRAM. The reason MS went with eSRAM is because they used DDR3 and DDR3 doesnt have nearly enough bandwidth to feed the GPU, so the solution to up the bandwidth for them was to use embedded static RAM.

    • GHz

      Umm dude, Megatexturing is done in software as a CPU or CUDA implementation. This got nothing to do with that.

      What you’re referring to Partially resident textures. That’s what openGL uses. This was implemented with the introduction of AMD HD7970. PS4 don’t have a HD7970 as a GPU. So the notion is that PS4 support will come via software. Partially resident textures will be done by the MMU of the GPU.

      TR on the XB1 is supported via hardware, which have its own advantages. It takes the work load off CPU and GPU in a big way. The trouble is, for you to see the benefits of this, devs will need access to newer and more advanced graphic engines. Graphite SDK helps to introduce the hardware support elements to devs if they want a go at it.

      http://graphinesoftware.com/node/70

      In short, if I was a dev, I’d go the TR route. No heavy increases
      in overhead, won’t want to deal with manual filtering, complications of anisotropic filtering, and duplicating border regions. All because it’s done in hardware. Sony is hush on their PRT tech, and being that we all know it’s not an HD7970 in there, the assumption is that it’s software based, in other words oldschool.

    • Xtreme Derp

      Completely wrong. All AMD GPUs based on their GCN architecture have hardware support for PRT. That includes both consoles.

      http://www.anandtech.com/show/4455/amds-graphics-core-next-preview-amd-architects-for-compute/5

      http://www.amd.com/en-us/innovations/software-technologies/gcn

    • GHz

      You’re absolutely right about PRT GCN connection, and that PRT is supported in both consoles, but you’re wrong in believing that TR and PRT are exactly the same. They share similar philosophies but TR is more advanced @ the moment. PRT is only limited to textures while TR can do other things besides textures, hence the reason they refer to it as “resources” and not “textures”. Implementing shadows through shadow mapping is one example.

      But openGL in due time can probably add extensions to bring it up to speed to TR. In the meantime, from what graphine software and MS had to say, TR is still better deal for now.

      Thanks for the correction though! Much appreciated! 🙂

    • Xtreme Derp

      I don’t think I said TR and PRT are the same thing, they’re different implementations between DirectX and OpenGL. Didn’t mean to give that impression.

      There’s a bit of confusion in that AMD calls its hardware function “PRT” but PRT is also sometimes referred to OpenGL’s implementation. I think the “official” name in OpenGL is the “sparse texture” function or something like that.

      Tiled Resources could be more advanced than PRT, I’m not exactly sure. PS4’s API is a custom derivative of OpenGL so Sony engineers could have written their own way of accessing the GCN feature set.

    • Johnny

      Alot of people don’t understand what you are saying sadly enough. To many keyboard devs these days.

  • BrendaSBarrientez

    You could already do this on OpenGL and if you really wanted to, you could do it on any capable hardware. DX is just like a program where you turn turn certain features on or off. But just because the program doesnt have the option doesnt mean the hardware cant do it. http://qr.net/sBx9

  • Xtreme Derp

    Both console GPUs are based on AMD GCN (graphics core next) technology, which includes hardware level support for PRT (partially resident textures). That hardware can be exposed to developers through APIs like OpenGL or DirectX. Both consoles are able to use texture tiling to save on memory size and bandwidth.

    The Granite SDK supports PC, mac, and mobile phones and basically does the heavy lifting for devs including hardware support if available (according to them).

    “First off, we have added hardware tiling support. On DirectX this is called “Tiled Resources”, but there is also an OpenGL equivalent under the name of “Partially Resident Textures”. Hardware support means that using our technology you will not only have to run less instructions in the shader, but there will also be a reduction in cache sizes (video memory), because no extra pixel borders are needed for cross-tile hardware texture filtering.”

    http://graphinesoftware.com/our-technology/how-it-works
    http://graphinesoftware.com/Announcing-Granite-2.0
    http://graphinesoftware.com/node/70

  • gamesepcs.com.br

    Xbox one won streaming quality HD 720 crystal com 3 mega upload.

  • Cigi

    @cozomel:disqus sorry buddy but you don’t get it. And by the way every one of your post starts with ohh fanboys.. man fanboys ….

    The issue here is that everybody can do tiled resources, but as shown with Rage software tiling has a lot of bottlenecks and problems. And with tiled resources you need to move blocks of textures a lot… not like streaming the whole scene. And is this case you relay on latency.

    See it like you can move by using the express lanes with lots of bandwidth – like with the GDDR5. But you can also take another road which i a lot shorter but has a lot of turns. Here your max speed does not help you. As you need to change course a lot of times. The eSRAM os ON THE DIE of the GPU and has very very low latency. THIS makes all the difference. The xbox was made with TR and RT in mind. Once again you show yourself as a poor fanboy with no insights to what you write…

    • Xtreme Derp

      Latency is a non-issue, PS4 has hardware level support for PRT like all AMD GCN GPUs.

      This is just typical ignorant fanboys grasping at straws on things they don’t understand and will have little to no impact on the graphics performance gap.

    • justerthought

      What you fail to understand is that a tiled ESRAM stream is singularly linear. Any change in direction needs a restarted stream. Half built textures are scrapped to build the new one, or new ones cannot start until the previous one is complete. Processing is wasted resulting in latency and pop-in.

      When a game requires multiple textures at the same time in response to where the player goes in an open world, it is clear that an ESRAM linear tiled stream is not going to be powerful enough, no matter how fast it runs. It might be OK for a linear tiled graphical tunnel racer like Forza where it can be predicted what textures are need in advance around the next bend, but not in an open world where the game cannot predict where the player will look next.

      The PS4 runs all its data in parallel almost as fast as the 32MB ESRAM linear stream maxed out. And don’t forget, all these textures you’re building up with linear streams, where are they going to go while your rendering the frame. They cannot go back into the ESRAM. It’s only a short term cache. On a PC they go into the GDDR5 RAM. They cannot float in thin air, so on XB1 they have to go into the slow DDR3 RAM with the GPU still waiting for data while drawing frames into the frame buffer.

      ESRAM on the die as you put it is a major flaw. In order to fit it on, the GPU has had to be cut in size. PS4 GPU has 50% more compute units than XB1, 18 vs 12. That performance drop can never be reclaimed. Latency is not an issue for graphics. It deals with large data sizes, not lots of small data sizes in quick succession like you get with general computing on a PC where the DDR3 is put to best use. For gaming, speed and parallel processing is more important than latency, so that is why every gaming PC worth any merit has at least 2GB of GDDR5 RAM attached to the GPU.

      Dream on buddy. You’ll eventually get the message. I’m going to call you the fanboy because your sensitivity to the term is very telling.

    • Cigi

      Surry but you are very wrong. You have to take the Move engines into account.

      Also as mentioned in this post there is a difference between PRT and TR. yes they are both hardware implementations of it, but the Xbox one has gone a step further and created the ENTIRE system to be low latency and capable of ‘MOVE’ing textures in a very low latency fashion.

      You see, if anyone can go have a look at a very OLD game called RAGE, rage had one MASSIVE texture which was the entire Game, they used ‘megatextures’ to ‘STREAM’ in the textures as they were needed.

      It was slow and also due to 2 factors was very limiting.

      1) all textures were Extremely Compressed as having one large image for the entire game instead of using the same image over and over and over and over again (like normal games, which have boring sceneries) all the texture data was Unique, every single part of the game was not the same as anything else. so what this means was 100’s of GB’s of Texture data and they had to compress the ‘crap’ out of it, so much so that it needed 3 DVD’s to install on the xbox 360 version of the game.

      *We have Blueray discs now which allow for much higher quality textures.

      2)With higher quality textures would then mean longer ‘pop in’ times for images on the screen because with megatextures it was all done with software.

      *This is where the xbox one’s move engines, low latency DDR3 ram and entire low latency system comes into consideration, as this then means very high quality textures can be streamed in without taxing the GPU (which on the PS4 it will be using the GPU to stream data in)

      *secondly, because they are now only streaming the data that they need (as what you can see, instead of the entire textures (which is what is done currently) this also means that you are no longer needing bandwidth like GDDR5 and instead will only need a tiny fraction of it) which means that Not only is DDR3 better suited for CPU, it will also be better for Tiled Resources as it’s latency is much better than GDDR5.

      The other thing to note is that ‘like with rage’ you could have 1080p graphics at 60 fps ‘constant’ and have massive unique worlds and it used next to nothing of the GPU.

      This is just one of the many reasons why I also said that when MS get the devs to use TR things are going to shine (and guess what, we are now seeing that sony will be using the likes of Granite… funny that, because they know it is the future, and the xbox one is already built to exploit this.

      http://gamingbolt.com/granite-sdk-interview-delivering-next-gen-texture-streaming-and-compression-middleware#sF6TsVGHWtbtzM9Q.99

      Perhaps the biggest benefits of using the Granite SDK is less taxation on the GPU side of things. This is especially relevant in case of PS4 and Xbox One who have far less compute units available [18 and 12 respectively] compared to a high end PC GPU. So the question of whether those GPUs have enough compute units to fully utilize the potential of Granite SDK is irrelevant according to Aljosha.

      So there goes you point of 18 vs 12 CU.

  • Rick Rambo

    Shut up…..everyone. I have never read and heard so much bullish!t in my life. My advice, live and let live. If you have PS4, love it. If you have XB1, love it. Bunch of dumba$$ haters on here!!!

  • You are flat out wrong

    This told us absolutely nothing. Thanks!

  • Guest

    Microsoft has bet the farm on DX12 and to ensure a win they need to make it a HUGE install base. X1 sales alone are not going to achieve that, we can all agree, so by also releasing Windows Threshold and making D3D12 universal they exponentially increased the potential install base and relevance of DX to a point where the optimizations (e.g. tiled resources) will be something devs will take time to program. X1 was designed specifically for DX12 and those optimizations and we’ll see that hopefully at E3 this year with announced games.

    PS4 has the brute force and it will remain in the lead for the current foreseeable future, but I do believe parity is on the horizon…so long as MS can prove to dev’s that optimizing through DX12 is worth their time.


 

Copyright © 2009-2015 GamingBolt.com. All Rights Reserved.