3D Apps & Engines' APIs + GPU Archs

[gabmass]

Guys

going over the benchmarks for VGA cards getting benchmarks with different 3D Apps and Engine makes me so confused of what suits me the best.

 

so far I have realized the Architecture of GPU Cores is crucial to know to make the right choice and as far as I know there are 3 types:

Fermi, Kepler and Tesla.

I need some clarification on this one above all.

 

2ndly, for 3D Programs

3Ds Max

Maya

... and so forth

what APIs are used? any reference guide?

mainly for viewport (active shade or real time)

 

 

lastly, for render engines

Vray

Mental Ray

....

what types of GPU-Computing are used for real time? (other than CPU Computing)

 

 

I'm just so confused about all these terms of OPEN CL, OPEN GL, DirectX, Direct Computing and I bet this goes on

 

 

thanks in advance

[numerobis]

Guys

so far I have realized the Architecture of GPU Cores is crucial to know to make the right choice and as far as I know there are 3 types:

Fermi, Kepler and Tesla.

I need some clarification on this one above all.

I think there is something mixed up...

Fermi and Kepler are the last two Chip architectures/generations of nvidia cards. The newest is Kepler. The next one, Maxwell, is announced for february/march 2014.

Tesla is the nvidia product name for computing cards. They have Geforce (gaming), Quadro (workstation) and Tesla (GPGPU no display card). All cards of these 3 categories are always based on the same chip architecture (i.e. Fermi or Kepler) - with some features enabled in each category and different drivers, but based on the same chip.

 

2ndly, for 3D Programs

3Ds Max

Maya

... and so forth

what APIs are used? any reference guide?

mainly for viewport (active shade or real time)

 

 

lastly, for render engines

Vray

Mental Ray

....

what types of GPU-Computing are used for real time? (other than CPU Computing)

 

 

I'm just so confused about all these terms of OPEN CL, OPEN GL, DirectX, Direct Computing and I bet this goes on

OpenGL, DirectX or Nitrous are display engines/APIs. CUDA, OpenCL, DirectCompute are for computing: CUDA (Compute Unified Device Architecture) is the API of nvidia, OpenCL (Open Computing Language) is open source and supported by AMD, and DirectCompute developed by Microsoft and included in DirectX.

 

Currently most GPU renderers use CUDA exclusively or support CUDA and OpenCL but are faster on CUDA (Octane, v-ray, IRay, Arion, TheaRender, Indigo,...) But this might change in the future since apple is pushing opencl now.

[gabmass]

Oh Numberbis

Gut gemacht

this is like a piece of gold to me....veryyyyyyyyyyyyyyy useful. you made my day thank you

 

I just came up with few questions. could you tell me about it?

Tesla (GPGPU no display card) !?????? does it mean it's just used for GPU Computing and it doesn't have any ports for screen? so to be used on 2nd and 3rd slots (not in SLI/CFX though) ?

 

and about the Fermi and Kepler !? which one beats the other in 3D Real Time? and which one's great for Display Engine like the viewport in 3Ds Max?

 

and do all graphic cards basically support OPEN GL , DIRECT X and Nitrous ?

 

 

bis bald

 

 

I think there is something mixed up...

Fermi and Kepler are the last two Chip architectures/generations of nvidia cards. The newest is Kepler. The next one, Maxwell, is announced for february/march 2014.

Tesla is the nvidia product name for computing cards. They have Geforce (gaming), Quadro (workstation) and Tesla (GPGPU no display card). All cards of these 3 categories are always based on the same chip architecture (i.e. Fermi or Kepler) - with some features enabled in each category and different drivers, but based on the same chip.

 

 

OpenGL, DirectX or Nitrous are display engines/APIs. CUDA, OpenCL, DirectCompute are for computing: CUDA (Compute Unified Device Architecture) is the API of nvidia, OpenCL (Open Computing Language) is open source and supported by AMD, and DirectCompute developed by Microsoft and included in DirectX.

 

Currently most GPU renderers use CUDA exclusively or support CUDA and OpenCL but are faster on CUDA (Octane, v-ray, IRay, Arion, TheaRender, Indigo,...) But this might change in the future since apple is pushing opencl now.

[numerobis]

Tesla (GPGPU no display card) !?????? does it mean it's just used for GPU Computing and it doesn't have any ports for screen? so to be used on 2nd and 3rd slots (not in SLI/CFX though) ?

yes

 

and about the Fermi and Kepler !? which one beats the other in 3D Real Time? and which one's great for Display Engine like the viewport in 3Ds Max?

I can't give you a clear advice here at the moment, because i don't know it absolutely.

I'm still waiting for a new comparison (quadro/geforce) under max 2014.

It depends on the max version, the card and the driver. It seems that since openGL got obsolete for max viewport and the performance in directX raised on consumer cards, nvidia tries to block this evolution. I think the latest quadro cards are getting better again compared to even higher clocked geforce - but i have no clear values.

And it's even harder to compare since max 2014 is making use of some heavy Adaptive Degradation for the viewports... so to get comparable results i think we would have to disable the Adaptive Degradation ( http://docs.autodesk.com/3DSMAX/16/ENU/3ds-Max-Help/index.html?url=files/GUID-486136D3-3BCC-4A95-823B-3E7F9F75C2C4.htm,topicNumber=d30e11672 )

Maybe everyone who has done this "benchmark": http://forums.cgarchitect.com/73746-graphics-card-benchmark.html should rerun it without adaptive degradation turned on...

 

For CUDA:

If you compare the results of a benchmark in v-ray RT, it looks like Fermi was more effective, considering that Kepler has a much higher power looking at the rough numbers. And the first Kepler cards (GTX 6xx) were not that good, while the latest GTX7xx (GK110) are definitely good cards for CUDA especially because they support more RAM than ever before. (btw. CDA is only supported by nvidia cards, and no AMD card).

 

Some values for vray RT from this thread: http://forums.chaosgroup.com/showthr...GPU-benchmarks

 

1m 28s --- 2x GTX580 3GB ( Fermi GF110 )

1m 31s --- 1x GTX 690 4GB ( 2 GPUs Kepler GK104 )

1m 46s --- 1x GTX Titan 6GB ( Kepler GK110 )

1m 56s --- 1x GTX 780 Superclocked 3GB ( Kepler GK110 )

2m 5s --- 1x GTX 780 3GB ( Kepler GK110 )

2m 44s --- 1x GTX580 3GB ( Fermi GF110 )

3m 03s ---1x GTX680 2GB ( Kepler GK104 )

3m 07s --- 1x GTX670 4GB ( Kepler GK104 )

3min12s --- 1x GTX570 1.26GB ( Fermi GF110 )

3min57s ---1x GTX470 1.26GB ( Fermi GF110 )

4min50s --- 1x GTX560Ti 1GB ( Fermi GF114 )

6min40s --- 1x GTX460 1GB ( Fermi GF104 )

 

OpenCL is still no alternative option for vrayRT - it works, but much slower.

But this could change in the future since Apple is using AMD cards in the new Macs and is pushing the development of OpenCL together with AMD and Adobe.

If you compare the cards with renderers that are developed exclusively for OpenCL, AMD cards are MUCH stronger than nvidia cards ( http://www.tomshardware.com/charts/2013-vga-gpgpu/15-GPGPU-Luxmark,2971.html )

 

and do all graphic cards basically support OPEN GL , DIRECT X and Nitrous ?

yes

 

bis bald

Edited November 20, 2013 by numerobis