v2.3 Vs v2.4 performance

Discussion related to the LuxCore functionality, implementations and API.
User avatar
B.Y.O.B.
Developer
Posts: 3872
Joined: Mon Dec 04, 2017 10:08 pm
Location: Germany
Contact:

Re: v2.3 Vs v2.4 performance

Post by B.Y.O.B. » Thu Apr 30, 2020 2:34 pm

Dade wrote:
Thu Apr 30, 2020 2:30 pm
I assume I can regain some performance (but only in CUDA): for the textures alone, there is a "switch(textureType)" with 60+ cases, in CUDA it can be done with a jump table with function pointers.
Did you do any profiling yet to find the hotspots in the interpreter?

User avatar
Dade
Developer
Posts: 4817
Joined: Mon Dec 04, 2017 8:36 pm
Location: Italy

Re: v2.3 Vs v2.4 performance

Post by Dade » Thu Apr 30, 2020 3:00 pm

B.Y.O.B. wrote:
Thu Apr 30, 2020 2:34 pm
Dade wrote:
Thu Apr 30, 2020 2:30 pm
I assume I can regain some performance (but only in CUDA): for the textures alone, there is a "switch(textureType)" with 60+ cases, in CUDA it can be done with a jump table with function pointers.
Did you do any profiling yet to find the hotspots in the interpreter?
I have (micro-)kernel execution times, each kernel does a quite specific task. The main bottle neck is mostly the BSDF sampling.
Support LuxCoreRender project with salts and bounties

epilectrolytics
Donor
Posts: 708
Joined: Thu Oct 04, 2018 6:06 am

Re: v2.3 Vs v2.4 performance

Post by epilectrolytics » Fri May 01, 2020 9:06 am

Comparison with v2.2 is even scarier:
Screenshot.jpg

User avatar
Sharlybg
Donor
Posts: 2483
Joined: Mon Dec 04, 2017 10:11 pm
Location: Ivory Coast

Re: v2.3 Vs v2.4 performance

Post by Sharlybg » Fri May 01, 2020 9:30 am

epilectrolytics wrote:
Fri May 01, 2020 9:06 am
Comparison with v2.2 is even scarier:
Screenshot.jpg
this is a lot even between 2.2 and 2.3 there is half performance lost. :?
Support LuxCoreRender project with salts and bounties

Portfolio : https://www.behance.net/DRAVIA

epilectrolytics
Donor
Posts: 708
Joined: Thu Oct 04, 2018 6:06 am

Re: v2.3 Vs v2.4 performance

Post by epilectrolytics » Fri May 01, 2020 9:46 am

Sharlybg wrote:
Fri May 01, 2020 9:30 am
this is a lot even between 2.2 and 2.3 there is half performance lost. :?
Yep, I'm totally stunned.
Maybe we need Optix RTX support to regain the render speed we already had last year??

And AMD users are kind of screwed right now :(

acasta69
Developer
Posts: 361
Joined: Tue Jan 09, 2018 3:45 pm

Re: v2.3 Vs v2.4 performance

Post by acasta69 » Fri May 01, 2020 1:06 pm

Here are results with GTX970 (2.3 vs. 2.4 build 20200430.12):

LuxCore2.3.jpg
LuxCore2.4alpha0_OpenCL.jpg
LuxCore2.4alpha0_CUDA.jpg
On this card CUDA underperforms OpenCL quite a bit... :?
Anyone else with this or similar GPU series can confirm?

P.S. With the very latest build, LuxCoreUI exits without any error message while evaluating PhotonGI best radius:

Code: Select all

[LuxRays][1.828] [MBVHKernel] Compiler options: -D LUXRAYS_OPENCL_KERNEL -D PARAM_RAY_EPSILON_MIN=0.0001f -D PARAM_RAY_EPSILON_MAX=0.1f -D LUXRAYS_OPENCL_DEVICE -cl-fast-relaxed-math -cl-mad-enable
[LuxRays][1.828] [MBVHKernel] Compiling kernels
[LuxRays][1.828] [MBVHKernel] Program cached
[LuxCore][1.859] [PathOCLRenderEngine] OpenCL task count: 262144
[LuxCore][1.859] [PathOCLBaseRenderEngine] OpenCL max. page memory size: 1048576Kbytes
[LuxCore][1.859] PhotonGI evaluating best radius

C:\Users\Alessandro\Desktop\luxcorerender-latest-win64-cuda\scenes\LuxCore2.1Benchmark\LuxCoreScene>
Support LuxCoreRender project with salts and bounties

Windows 10 64 bits, i7-4770 3.4 GHz, RAM 16 GB, GTX 970 4GB v445.87

User avatar
Dade
Developer
Posts: 4817
Joined: Mon Dec 04, 2017 8:36 pm
Location: Italy

Re: v2.3 Vs v2.4 performance

Post by Dade » Fri May 01, 2020 1:12 pm

acasta69 wrote:
Fri May 01, 2020 1:06 pm
Anyone else with this or similar GPU series can confirm?
A 970 GTX can not be 7 times slower than a 2070 RTX :!:
Support LuxCoreRender project with salts and bounties

acasta69
Developer
Posts: 361
Joined: Tue Jan 09, 2018 3:45 pm

Re: v2.3 Vs v2.4 performance

Post by acasta69 » Fri May 01, 2020 1:17 pm

Dade wrote:
Fri May 01, 2020 1:12 pm
A 970 GTX can not be 7 times slower than a 2070 RTX :!:
Indeed! I dont' understand if it can be something broken on my system or not...
I tested both with Azure-built and self-built binaries, results are the same.
Support LuxCoreRender project with salts and bounties

Windows 10 64 bits, i7-4770 3.4 GHz, RAM 16 GB, GTX 970 4GB v445.87

User avatar
Dade
Developer
Posts: 4817
Joined: Mon Dec 04, 2017 8:36 pm
Location: Italy

Re: v2.3 Vs v2.4 performance

Post by Dade » Fri May 01, 2020 3:22 pm

acasta69 wrote:
Fri May 01, 2020 1:06 pm
P.S. With the very latest build, LuxCoreUI exits without any error message while evaluating PhotonGI best radius:
I should have fixed this problem.
Support LuxCoreRender project with salts and bounties

User avatar
Dade
Developer
Posts: 4817
Joined: Mon Dec 04, 2017 8:36 pm
Location: Italy

Re: v2.3 Vs v2.4 performance

Post by Dade » Fri May 01, 2020 3:36 pm

BTW, I'm doing the tests with PGI disabled (or the start up time may affect the stats).
Support LuxCoreRender project with salts and bounties

Post Reply