Geforce TitanX SSS material on GPU renders slow #44903

Closed
opened 2015-05-31 11:39:51 +02:00 by Martin Lindelöf · 21 comments

System Information
Win8 x64
Geforce TitanX

Blender Version
Broken: 2.74 2015-05-29 (buildbot build) hash:1a0c961
Worked: (optional)

Short description of error
SSS rendering on the TitanX is slow, for me it's slower than CPU.
I've got contacted on BA.org by a french user having a TitanX and a Titan Black. On his system the Titan Black is faster than the TitanX on SSS rendering.

I mentioned it because he is also eager to see if there's a solution, and if it could be of any assistance to mention that the Titan Black (Kepler) is faster rendering SSS
than the Titan X (Maxwell) card.

Correct me if I am wrong but it could be that they are using different cuda kernels. sm_50 and sm_52 are for maxwell cards, TitanBlack being kepler is it using sm_35?
anyway happy if someone could investigate if there is a solution.

Exact steps for others to reproduce the error
Just try and render a scene with a SSS material present, on a system with TitanX.
I added a simple .blend scene and two screenshots.
untitled.blend
render-cpu.png

render-gpu.png

**System Information** Win8 x64 Geforce TitanX **Blender Version** Broken: 2.74 2015-05-29 (buildbot build) hash:1a0c961 Worked: (optional) **Short description of error** SSS rendering on the TitanX is slow, for me it's slower than CPU. I've got contacted on BA.org by a french user having a TitanX and a Titan Black. On his system the Titan Black is faster than the TitanX on SSS rendering. I mentioned it because he is also eager to see if there's a solution, and if it could be of any assistance to mention that the Titan Black (Kepler) is faster rendering SSS than the Titan X (Maxwell) card. Correct me if I am wrong but it could be that they are using different cuda kernels. sm_50 and sm_52 are for maxwell cards, TitanBlack being kepler is it using sm_35? anyway happy if someone could **investigate** if there is a solution. **Exact steps for others to reproduce the error** Just try and render a scene with a SSS material present, on a system with TitanX. I added a simple .blend scene and two screenshots. [untitled.blend](https://archive.blender.org/developer/F182352/untitled.blend) ![render-cpu.png](https://archive.blender.org/developer/F182348/render-cpu.png) ![render-gpu.png](https://archive.blender.org/developer/F182349/render-gpu.png)
Author
Member

Changed status to: 'Open'

Changed status to: 'Open'
Martijn Berger was assigned by Martin Lindelöf 2015-05-31 11:39:51 +02:00
Author
Member

Added subscriber: @MartinLindelof

Added subscriber: @MartinLindelof
Member

I don't know if the maxwell card has to be faster here then the kepler card.

But I really think there is a high probability we could improve performance on maxwell in general.

I don't know if the maxwell card has to be faster here then the kepler card. But I really think there is a high probability we could improve performance on maxwell in general.

Added subscriber: @ThomasDinges

Added subscriber: @ThomasDinges

Which CPU do you have?

Which CPU do you have?
Author
Member

Intel i7-3820 3.6Ghz, Quad core stuff.
Usually my CPU doesn't outperform my GPUs (have two cards)
here's another test render comparison with my standard shader, regarding CPU vs GPU performance on my machine.

render-watermelon-cpu.png

CPU is 70s

render-watermelon-gpu.png

GPU is 9s

So it's weird to see my CPU outperform GPU on SSS, when usually it's so much faster. Gonna check with the french blender user if he can submit/contribute GPU renders Titan Black vs Titan X.

Intel i7-3820 3.6Ghz, Quad core stuff. Usually my CPU doesn't outperform my GPUs (have two cards) here's another test render comparison with my standard shader, regarding CPU vs GPU performance on my machine. ![render-watermelon-cpu.png](https://archive.blender.org/developer/F182729/render-watermelon-cpu.png) CPU is 70s ![render-watermelon-gpu.png](https://archive.blender.org/developer/F182730/render-watermelon-gpu.png) GPU is 9s So it's weird to see my CPU outperform GPU on SSS, when usually it's so much faster. Gonna check with the french blender user if he can submit/contribute GPU renders Titan Black vs Titan X.
Member

SSS is really really not nice on GPU's but that is to be expected. This is more about why the maxwell card performs less well then one would expect.

For that problem I expect some of the kepler specific hacks might actually improve performance for the SSS case. But I need tot run tests to see what is really going on.

SSS is really really not nice on GPU's but that is to be expected. This is more about why the maxwell card performs less well then one would expect. For that problem I expect some of the kepler specific hacks might actually improve performance for the SSS case. But I need tot run tests to see what is really going on.
Author
Member

Maxwell has performed good for me, except SSS. I will try and get some test renders Titan black vs Titan X from the french user.

And if you need me to run tests, I will help out!

Maxwell has performed good for me, except SSS. I will try and get some test renders Titan black vs Titan X from the french user. And if you need me to run tests, I will help out!

Added subscriber: @P2design

Added subscriber: @P2design

Hi all,

Here are my specs :
AMD FMX 9370 - 8 cores - 4.41Ghz
32 GB DDR 3

ASUS GEFORCE TITAN BLACK - 6GB
ASUS GEFORCE TITAN X - 12 GB
Blender 2.74 official release
Nvidia Drivers : 352.86

Here are the result based on Martin Blend.File.
Titan X run 5 time slower than Titan Black.

TITAN X
Titan-X.png

TITAN BLACK
Titan-black.png
This performance gap append on EVERY SHADERS, as soon as Experimental mode is enabled.
And the performance loss is more important on TITAN X than BLACK. Test made with same file, glass shader with samples bumped to 500.

TITAN X
Titan-X-glass.png

Titan-X-glass-experimental.png

TITAN BLACK
Titan-black-glass-experimental.png

Titan-black-glass.png

I hope it will help and you can improve Titan X performances.
I'm really not a good technician about these technologies... I just can say that Titan X is supposed to be faster as it has more CUDA cores and is not -_-°

Hi all, Here are my specs : AMD FMX 9370 - 8 cores - 4.41Ghz 32 GB DDR 3 ASUS GEFORCE TITAN BLACK - 6GB ASUS GEFORCE TITAN X - 12 GB Blender 2.74 official release Nvidia Drivers : 352.86 Here are the result based on Martin Blend.File. Titan X run 5 time slower than Titan Black. TITAN X ![Titan-X.png](https://archive.blender.org/developer/F182965/Titan-X.png) TITAN BLACK ![Titan-black.png](https://archive.blender.org/developer/F182967/Titan-black.png) This performance gap append on EVERY SHADERS, as soon as Experimental mode is enabled. And the performance loss is more important on TITAN X than BLACK. Test made with same file, glass shader with samples bumped to 500. TITAN X ![Titan-X-glass.png](https://archive.blender.org/developer/F182969/Titan-X-glass.png) ![Titan-X-glass-experimental.png](https://archive.blender.org/developer/F182970/Titan-X-glass-experimental.png) TITAN BLACK ![Titan-black-glass-experimental.png](https://archive.blender.org/developer/F182973/Titan-black-glass-experimental.png) ![Titan-black-glass.png](https://archive.blender.org/developer/F182974/Titan-black-glass.png) I hope it will help and you can improve Titan X performances. I'm really not a good technician about these technologies... I just can say that Titan X is supposed to be faster as it has more CUDA cores and is not -_-°

BTW....

If you need any other further tests, do not hesitate I'll be glad to perform them.

BTW.... If you need any other further tests, do not hesitate I'll be glad to perform them.
Author
Member

Ah yeah, can confirm that, from 8.5s to 28.5s to render a test scene when going from supported to experimental.

Ah yeah, can confirm that, from 8.5s to 28.5s to render a test scene when going from supported to experimental.

Added subscriber: @mont29

Added subscriber: @mont29

Changed status from 'Open' to: 'Archived'

Changed status from 'Open' to: 'Archived'

The Experimental kernel was added because of the increased memory usage and performance drop with enabled SSS. We are aware of that and hope to improve that in the future (CUDA kernel split should fix these issues).

Closing as Archived, as this is a ToDo.

The Experimental kernel was added *because* of the increased memory usage and performance drop with enabled SSS. We are aware of that and hope to improve that in the future (CUDA kernel split should fix these issues). Closing as Archived, as this is a ToDo.

Hi Everyone,

Did you have any other information about this fix?
Do you have any idea when it could be fixed?

Thanks for your interest and B.REGARDS

Hi Everyone, Did you have any other information about this fix? Do you have any idea when it could be fixed? Thanks for your interest and B.REGARDS

Added subscriber: @albron

Added subscriber: @albron

Hello everyone,

Discovering this thread as I just got a Titan X and discovering this problem with "experimental" render enable (using SSS).

I tried on 2.75 and 2.76 and the problem is the same...

My specs:
Processor I7 4770
MB: Asus Z87 WS
GC : Gigabyte GTX titan X

WIndows 8.1

Very frustrating as I'm rendering characters so using SSS all the time...
Arghh...

Hope you guys will be able to fix this soon....

Thanks a lot for your good work and dedication to this great software.

Alex

Hello everyone, Discovering this thread as I just got a Titan X and discovering this problem with "experimental" render enable (using SSS). I tried on 2.75 and 2.76 and the problem is the same... My specs: Processor I7 4770 MB: Asus Z87 WS GC : Gigabyte GTX titan X WIndows 8.1 Very frustrating as I'm rendering characters so using SSS all the time... Arghh... Hope you guys will be able to fix this soon.... Thanks a lot for your good work and dedication to this great software. Alex

Added subscriber: @MohamedSakr

Added subscriber: @MohamedSakr

I got 2 cards, GTX 970 and GTX 780, testing with experimental mode with a SSS material

GTX 780 = 8 seconds
GTX 970 = 30 seconds

the same scene without SSS "just diffuse"
GTX 780 = 4 seconds
GTX 970 = 6 seconds

something is wrong with Maxwell kernel settings "probably registers"

I got 2 cards, GTX 970 and GTX 780, testing with experimental mode with a SSS material GTX 780 = 8 seconds GTX 970 = 30 seconds the same scene without SSS "just diffuse" GTX 780 = 4 seconds GTX 970 = 6 seconds something is wrong with Maxwell kernel settings "probably registers"

Hello,

I wanted to have some news about this issue...
Can I help in any way?

Testing scenes or things like this?

Thanks for your work and your feedbacks.

Alex

Hello, I wanted to have some news about this issue... Can I help in any way? Testing scenes or things like this? Thanks for your work and your feedbacks. Alex
Sign in to join this conversation.
No Label
Interest
Alembic
Interest
Animation & Rigging
Interest
Asset Browser
Interest
Asset Browser Project Overview
Interest
Audio
Interest
Automated Testing
Interest
Blender Asset Bundle
Interest
BlendFile
Interest
Collada
Interest
Compatibility
Interest
Compositing
Interest
Core
Interest
Cycles
Interest
Dependency Graph
Interest
Development Management
Interest
EEVEE
Interest
EEVEE & Viewport
Interest
Freestyle
Interest
Geometry Nodes
Interest
Grease Pencil
Interest
ID Management
Interest
Images & Movies
Interest
Import Export
Interest
Line Art
Interest
Masking
Interest
Metal
Interest
Modeling
Interest
Modifiers
Interest
Motion Tracking
Interest
Nodes & Physics
Interest
OpenGL
Interest
Overlay
Interest
Overrides
Interest
Performance
Interest
Physics
Interest
Pipeline, Assets & IO
Interest
Platforms, Builds & Tests
Interest
Python API
Interest
Render & Cycles
Interest
Render Pipeline
Interest
Sculpt, Paint & Texture
Interest
Text Editor
Interest
Translations
Interest
Triaging
Interest
Undo
Interest
USD
Interest
User Interface
Interest
UV Editing
Interest
VFX & Video
Interest
Video Sequencer
Interest
Virtual Reality
Interest
Vulkan
Interest
Wayland
Interest
Workbench
Interest: X11
Legacy
Blender 2.8 Project
Legacy
Milestone 1: Basic, Local Asset Browser
Legacy
OpenGL Error
Meta
Good First Issue
Meta
Papercut
Meta
Retrospective
Meta
Security
Module
Animation & Rigging
Module
Core
Module
Development Management
Module
EEVEE & Viewport
Module
Grease Pencil
Module
Modeling
Module
Nodes & Physics
Module
Pipeline, Assets & IO
Module
Platforms, Builds & Tests
Module
Python API
Module
Render & Cycles
Module
Sculpt, Paint & Texture
Module
Triaging
Module
User Interface
Module
VFX & Video
Platform
FreeBSD
Platform
Linux
Platform
macOS
Platform
Windows
Priority
High
Priority
Low
Priority
Normal
Priority
Unbreak Now!
Status
Archived
Status
Confirmed
Status
Duplicate
Status
Needs Info from Developers
Status
Needs Information from User
Status
Needs Triage
Status
Resolved
Type
Bug
Type
Design
Type
Known Issue
Type
Patch
Type
Report
Type
To Do
No Milestone
No project
No Assignees
7 Participants
Notifications
Due Date
The due date is invalid or out of range. Please use the format 'yyyy-mm-dd'.

No due date set.

Dependencies

No dependencies set.

Reference: blender/blender#44903
No description provided.