• Advertisement

3D Rough idea on maximum 3D texture size

Recommended Posts

Now I know this is dependent on the PC to a big extent but I've been out of the loop with modern hardware for a few years. I have a project where a large 3D texture might solve things, we're talking maybe 512^3 but probably 8-bit, single channel. Memory wise this is not a big deal (it's not a game project and there are not lots of models or textures in use) but I really don't know if it would be supported on a typical PC (not a games rig). Or if it would slow things to a crawl... or once it's in the GPU memory it would not be unduly performance-limiting?

 

Share this post


Link to post
Share on other sites
Advertisement

the maximum size of a 3D texture in D3D11 hardware must  is 2048 (on all axis, so total maximum resolution can be 2048*2048*2048). so yes, it is supported on virtually every hardware out there. Performance wise there shouldn't be a difference that a 2D texture

Edited by Yourself

Share this post


Link to post
Share on other sites

Under OpenGL, this limit is defined by querying GL_MAX_3D_TEXTURE_SIZE. Also, as stated here, try to stick with POT texture sizes as much as possible.

To my opinion, the API will never give a size that, when used, will make the texture unpracticable (ie, too slow to be rendered or to be probed in a shader).

Edited by _Silence_

Share this post


Link to post
Share on other sites

512^3 is 128MB with 8-bit texels, so it reasonably fits into GPU memory on modern hardware.

2048^3 is 8GB which does not fit into on-board memory of common gaming GPUs. It does fit into professional (as in Quadro) cards, but even in professional applications it is rare that such large individual buffers would be used.

Note that the memory access is relatively much slower than shader arithmetic, if compared to 5-10 years ago. This is because memory bus width cannot be easily scaled up, but the number of shader cores has increased quite a bit. Whereas in the old days something would run better from a 3d lookup table, the same logic could now run much faster by just running the math on the shader. Depends on the program, of course.

Edited by Nik02

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now


  • Advertisement
  • Advertisement
  • Popular Now

  • Advertisement
  • Similar Content

    • By KarimIO
      Hey guys,
      Are lightmaps still the best way to handle static diffuse irradiance, or is SH used for both diffuse and specular irradiance now?
      Also, do any modern games use direct light in lightmaps, or are all direct lighting handled by shadow maps now?
      Finally, how is SH usually baked?
      Thanks!
    • By KarimIO
      Hey guys
      So I was wondering how modern terrain and water geometry works both with and without tesselation. Essentially:
      1) Is Geoclipmapping still the best CPU tesselation technique?
      2) Is Geoclipmapping still used with tesselation?
      3) Is non-tesselated water just flat? Is there any other (reasonable) ways to simulate it? Do people use Geoclipmapping for that too?
      Thanks!
    • By Atwo Studios
       
      Hey guys,

      Anthony here from Atwo Studios bringing you some new updates for the new year!
      In this video I go over our game ROY, the new games and some general updates to the company!

      If you have not checked out ROY feel free to give it a try! Many people have said they enjoyed the game thus far!
      ROY: https://goo.gl/o6JJ5P
       
    • By evelyn4you
      hi,
      until now i use typical vertexshader approach for skinning with a Constantbuffer containing the transform matrix for the bones and an the vertexbuffer containing bone index and bone weight.
      Now i have implemented realtime environment  probe cubemaping so i have to render my scene from many point of views and the time for skinning takes too long because it is recalculated for every side of the cubemap.
      For Info i am working on Win7 an therefore use one Shadermodel 5.0 not 5.x that have more options, or is there a way to use 5.x in Win 7
      My Graphic Card is Directx 12 compatible NVidia GTX 960
      the member turanszkij has posted a good for me understandable compute shader. ( for Info: in his engine he uses an optimized version of it )
      https://turanszkij.wordpress.com/2017/09/09/skinning-in-compute-shader/
      Now my questions
       is it possible to feed the compute shader with my orignial vertexbuffer or do i have to copy it in several ByteAdressBuffers as implemented in the following code ?
        the same question is about the constant buffer of the matrixes
       my more urgent question is how do i feed my normal pipeline with the result of the compute Shader which are 2 RWByteAddressBuffers that contain position an normal
      for example i could use 2 vertexbuffer bindings
      1 containing only the uv coordinates
      2.containing position and normal
      How do i copy from the RWByteAddressBuffers to the vertexbuffer ?
       
      (Code from turanszkij )
      Here is my shader implementation for skinning a mesh in a compute shader:
      1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 struct Bone { float4x4 pose; }; StructuredBuffer<Bone> boneBuffer;   ByteAddressBuffer vertexBuffer_POS; // T-Pose pos ByteAddressBuffer vertexBuffer_NOR; // T-Pose normal ByteAddressBuffer vertexBuffer_WEI; // bone weights ByteAddressBuffer vertexBuffer_BON; // bone indices   RWByteAddressBuffer streamoutBuffer_POS; // skinned pos RWByteAddressBuffer streamoutBuffer_NOR; // skinned normal RWByteAddressBuffer streamoutBuffer_PRE; // previous frame skinned pos   inline void Skinning(inout float4 pos, inout float4 nor, in float4 inBon, in float4 inWei) {  float4 p = 0, pp = 0;  float3 n = 0;  float4x4 m;  float3x3 m3;  float weisum = 0;   // force loop to reduce register pressure  // though this way we can not interleave TEX - ALU operations  [loop]  for (uint i = 0; ((i &lt; 4) &amp;&amp; (weisum&lt;1.0f)); ++i)  {  m = boneBuffer[(uint)inBon].pose;  m3 = (float3x3)m;   p += mul(float4(pos.xyz, 1), m)*inWei;  n += mul(nor.xyz, m3)*inWei;   weisum += inWei;  }   bool w = any(inWei);  pos.xyz = w ? p.xyz : pos.xyz;  nor.xyz = w ? n : nor.xyz; }   [numthreads(1024, 1, 1)] void main( uint3 DTid : SV_DispatchThreadID ) {  const uint fetchAddress = DTid.x * 16; // stride is 16 bytes for each vertex buffer now...   uint4 pos_u = vertexBuffer_POS.Load4(fetchAddress);  uint4 nor_u = vertexBuffer_NOR.Load4(fetchAddress);  uint4 wei_u = vertexBuffer_WEI.Load4(fetchAddress);  uint4 bon_u = vertexBuffer_BON.Load4(fetchAddress);   float4 pos = asfloat(pos_u);  float4 nor = asfloat(nor_u);  float4 wei = asfloat(wei_u);  float4 bon = asfloat(bon_u);   Skinning(pos, nor, bon, wei);   pos_u = asuint(pos);  nor_u = asuint(nor);   // copy prev frame current pos to current frame prev pos streamoutBuffer_PRE.Store4(fetchAddress, streamoutBuffer_POS.Load4(fetchAddress)); // write out skinned props:  streamoutBuffer_POS.Store4(fetchAddress, pos_u);  streamoutBuffer_NOR.Store4(fetchAddress, nor_u); }  
    • By mister345
      Hi, can someone please explain why this is giving an assertion EyePosition!=0 exception?
       
      _lightBufferVS->viewMatrix = DirectX::XMMatrixLookAtLH(XMLoadFloat3(&_lightBufferVS->position), XMLoadFloat3(&_lookAt), XMLoadFloat3(&up));
      It looks like DirectX doesnt want the 2nd parameter to be a zero vector in the assertion, but I passed in a zero vector with this exact same code in another program and it ran just fine. (Here is the version of the code that worked - note XMLoadFloat3(&m_lookAt) parameter value is (0,0,0) at runtime - I debugged it - but it throws no exceptions.
          m_viewMatrix = DirectX::XMMatrixLookAtLH(XMLoadFloat3(&m_position), XMLoadFloat3(&m_lookAt), XMLoadFloat3(&up)); Here is the repo for the broken code (See LightClass) https://github.com/mister51213/DirectX11Engine/blob/master/DirectX11Engine/LightClass.cpp
      and here is the repo with the alternative version of the code that is working with a value of (0,0,0) for the second parameter.
      https://github.com/mister51213/DX11Port_SoftShadows/blob/master/Engine/lightclass.cpp
  • Advertisement