Windows Build FAILED

Discussion related to the LuxCore functionality, implementations and API.
User avatar
Dade
Developer
Developer
Posts: 5672
Joined: Mon Dec 04, 2017 8:36 pm
Location: Italy

Re: Windows Build FAILED

Post by Dade »

epilectrolytics wrote: Fri Apr 24, 2020 6:46 pm
Dade wrote: Fri Apr 24, 2020 6:02 pmhave you tried to select only one ?
Because of the auto-start render in LuxCoreUI I cannot select a device there.
You can define any property on the command line (including the one to select the devices).
Support LuxCoreRender project with salts and bounties
User avatar
Dade
Developer
Developer
Posts: 5672
Joined: Mon Dec 04, 2017 8:36 pm
Location: Italy

Re: Windows Build FAILED

Post by Dade »

kintuX wrote: Fri Apr 24, 2020 9:44 pm Right now I'm getting nowhere fast :D
You need to install the CUDA SDK to compile the CUDA code (the 2.6GB stuff).
Support LuxCoreRender project with salts and bounties
acasta69
Developer
Developer
Posts: 472
Joined: Tue Jan 09, 2018 3:45 pm
Location: Italy

Re: Windows Build FAILED

Post by acasta69 »

kintuX wrote: Fri Apr 24, 2020 9:44 pm Is there a way to have bare minimal files, since I really don't want to DL the whole 2.6 GB while needing maybe 10% of that... ?
To me every Byte matters. ;)
This should be close to the bare minimum, also without any documentation - 58 MB:

Code: Select all

cuda_10.1.243_win10_network.exe -s nvcc_10.1 cudart_10.1 nvrtc_10.1 nvrtc_dev_10.1
Edit: It's a silent install, so you won't see anything, but you can check the "setup.exe" process in task manager. For me it takes a couple of minutes to complete.
Support LuxCoreRender project with salts and bounties

Windows 10 64 bits, i7-4770 3.4 GHz, RAM 16 GB, GTX 970 4GB v445.87
johannes.wilde
Posts: 68
Joined: Fri Sep 21, 2018 7:57 am

Re: Windows Build FAILED

Post by johannes.wilde »

Hello!
Do one have to install this Cuda Files as well for the BlendLuxCoreCuda Version? I downloaded it yesterday for Windows and get the following message:

Hardware device selection string has the wrong length, must be 2 instead of 1

Sorry, but i don't understand anything from this coding magic...

Thanks for your great work on LuxCoreRender!
acasta69
Developer
Developer
Posts: 472
Joined: Tue Jan 09, 2018 3:45 pm
Location: Italy

Re: Windows Build FAILED

Post by acasta69 »

johannes.wilde wrote: Sat Apr 25, 2020 10:26 am Hardware device selection string has the wrong length, must be 2 instead of 1
I am getting the same error, which is related to this property in the scene cfg file:

Code: Select all

opencl.devices.select
As Dade explained in another post, with Cuda each Nvidia card is listed twice, both as OpenCL device and as Cuda device.
I think this is probably not yet supported in the BlendLuxCore addon, and unfortunately I'm not able to suggest a workaround. (*)
Let's wait for B.Y.O.B.'s suggestions about this.


(*) If this is suitable for you, you might try the following:
  1. export the scene from Blender in text format
  2. edit the above mentioned row in the "render.cfg" file, adding a digit to the right-hand-side
  3. render using LuxcoreUI with the standalone version
At step 2, you need to choose the Cuda device, not the openCL one.
Devices are listed in the log like this:

Code: Select all

[LuxRays][24.218] OpenCL Platform 0: NVIDIA Corporation
[LuxRays][24.218] OpenCL Platform 1: Advanced Micro Devices, Inc.
[LuxRays][24.218] CUDA driver version: 10.20
[LuxRays][24.218] CUDA device count: 1
[LuxRays][24.218] Device 0 name: Native
[LuxRays][24.218] Device 0 type: NATIVE_THREAD
[LuxRays][24.218] Device 0 compute units: 1
[LuxRays][24.218] Device 0 preferred float vector width: 4
[LuxRays][24.218] Device 0 max allocable memory: 17592186044415MBytes
[LuxRays][24.218] Device 0 max allocable memory block size: 17592186044415MBytes
[LuxRays][24.218] Device 0 has out of core memory support: 0
[LuxRays][24.218] Device 1 name: GeForce GTX 970
[LuxRays][24.218] Device 1 type: OPENCL_GPU
[LuxRays][24.218] Device 1 compute units: 13
[LuxRays][24.218] Device 1 preferred float vector width: 1
[LuxRays][24.218] Device 1 max allocable memory: 4096MBytes
[LuxRays][24.218] Device 1 max allocable memory block size: 1024MBytes
[LuxRays][24.218] Device 1 has out of core memory support: 0
[LuxRays][24.218] Device 2 name: Intel(R) Core(TM) i7-4770 CPU @ 3.40GHz
[LuxRays][24.218] Device 2 type: OPENCL_CPU
[LuxRays][24.218] Device 2 compute units: 8
[LuxRays][24.218] Device 2 preferred float vector width: 8
[LuxRays][24.218] Device 2 max allocable memory: 16323MBytes
[LuxRays][24.218] Device 2 max allocable memory block size: 4080MBytes
[LuxRays][24.218] Device 2 has out of core memory support: 0
[LuxRays][24.218] Device 3 name: GeForce GTX 970
[LuxRays][24.234] Device 3 type: CUDA_GPU
[LuxRays][24.234] Device 3 compute units: 128
[LuxRays][24.234] Device 3 preferred float vector width: 1
[LuxRays][24.234] Device 3 max allocable memory: 4096MBytes
[LuxRays][24.234] Device 3 max allocable memory block size: 17592186044415MBytes
[LuxRays][24.234] Device 3 has out of core memory support: 1
Traceback (most recent call last):
  File "C:\Users\Alessandro\AppData\Roaming\Blender Foundation\Blender\2.82\scripts\addons\BlendLuxCore\engine\base.py", line 83, in render_final
    final.render(self, depsgraph)
  File "C:\Users\Alessandro\AppData\Roaming\Blender Foundation\Blender\2.82\scripts\addons\BlendLuxCore\engine\final.py", line 49, in render
    _render_layer(engine, depsgraph, statistics, layer)
  File "C:\Users\Alessandro\AppData\Roaming\Blender Foundation\Blender\2.82\scripts\addons\BlendLuxCore\engine\final.py", line 61, in _render_layer
    engine.session = engine.exporter.create_session(depsgraph, engine=engine, view_layer=view_layer)
  File "C:\Users\Alessandro\AppData\Roaming\Blender Foundation\Blender\2.82\scripts\addons\BlendLuxCore\export\__init__.py", line 251, in create_session
    return pyluxcore.RenderSession(renderconfig)
RuntimeError: Hardware device selection string has the wrong length, must be 3 instead of 2
ERROR: Hardware device selection string has the wrong length, must be 3 instead of 2
Info: Compiling OpenCL kernels (just once, takes a few minutes)
Error: Hardware device selection string has the wrong length, must be 3 instead of 2
In this case, my string would have length 3, and the CUDA_GPU device is the last listed, so I would use:

Code: Select all

opencl.devices.select = 001
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
Developer
Posts: 5672
Joined: Mon Dec 04, 2017 8:36 pm
Location: Italy

Re: Windows Build FAILED

Post by Dade »

neo2068 wrote: Fri Apr 24, 2020 6:43 pm Yes, it is related to multi GPU config. It compiles the kernels and exit without error.
I was able to replicate a problem by using 2 times the same device (to simulate multiple GPUs) and I should have fixed the issue.
neo2068 wrote: Fri Apr 24, 2020 6:43 pm The same happens if you change some render config with only one device selected, e. g. sampler or render size.
I'm not able to replicate this. Are you running LuxCoreUI and just selecting a different sampler ?
Support LuxCoreRender project with salts and bounties
epilectrolytics
Donor
Donor
Posts: 790
Joined: Thu Oct 04, 2018 6:06 am

Re: Windows Build FAILED

Post by epilectrolytics »

Dade wrote: Sat Apr 25, 2020 1:03 pm I was able to replicate a problem by using 2 times the same device (to simulate multiple GPUs) and I should have fixed the issue.
Yep, now it works in LuxCoreUI, thanks Dade and acasta69!
First impressions: 30% speed gain with CUDA, kernel compilation faster but the UI gets unresponsive compared with OCL version.

Edit: While in the OCL version I can now deselect one of my GPUs and render with the other one, the CUDA version throws this error that hardware string length is wrong and should be 4 instead of 2.
neo2068
Developer
Developer
Posts: 260
Joined: Tue Dec 05, 2017 6:06 pm
Location: Germany

Re: Windows Build FAILED

Post by neo2068 »

Dade wrote: Sat Apr 25, 2020 1:03 pm
neo2068 wrote: Fri Apr 24, 2020 6:43 pm The same happens if you change some render config with only one device selected, e. g. sampler or render size.
I'm not able to replicate this. Are you running LuxCoreUI and just selecting a different sampler ?
Yes, I selected only one device and change the sampler in LuxCoreUI or resized the LuxCoreUI window. I will test if it still happens with the new build.
i7 5820K, 32 GB RAM, NVIDIA Geforce RTX 2080 SUPER + GTX 1080, Windows 10 64bit, Blender 2.83.5
Support LuxCoreRender project with salts and bounties
neo2068
Developer
Developer
Posts: 260
Joined: Tue Dec 05, 2017 6:06 pm
Location: Germany

Re: Windows Build FAILED

Post by neo2068 »

With the latest fixes it works fine for me.
i7 5820K, 32 GB RAM, NVIDIA Geforce RTX 2080 SUPER + GTX 1080, Windows 10 64bit, Blender 2.83.5
Support LuxCoreRender project with salts and bounties
kintuX
Posts: 810
Joined: Wed Jan 10, 2018 2:37 am

Re: Windows Build FAILED

Post by kintuX »

Dade wrote: Fri Apr 24, 2020 11:36 pm
kintuX wrote: Fri Apr 24, 2020 9:44 pm Right now I'm getting nowhere fast :D
You need to install the CUDA SDK to compile the CUDA code (the 2.6GB stuff).
acasta69 wrote: Sat Apr 25, 2020 9:20 am
kintuX wrote: Fri Apr 24, 2020 9:44 pm Is there a way to have bare minimal files, since I really don't want to DL the whole 2.6 GB while needing maybe 10% of that... ?
To me every Byte matters. ;)
This should be close to the bare minimum, also without any documentation - 58 MB:

Code: Select all

cuda_10.1.243_win10_network.exe -s nvcc_10.1 cudart_10.1 nvrtc_10.1 nvrtc_dev_10.1
Edit: It's a silent install, so you won't see anything, but you can check the "setup.exe" process in task manager. For me it takes a couple of minutes to complete.
Right...
- So I looked FOR a bit and found http://www.cngwireless.net/apps/Drivers ... etwork.exe (19.3MB) :D
- Ran it, installed CUDA stuff only.
- Restarted.
- Compiled latest (1d8818b34) /cuda:

Code: Select all

Build succeeded.
    0 Warning(s)
    0 Error(s)
- Ran Blender 2.81a.
//// CUDA is not available yet.
//// OCL (as noted) gives an error:

Code: Select all

Traceback (most recent call last):
  File "C:\Users\309\AppData\Roaming\Blender Foundation\Blender\2.81\scripts\addons\BlendLuxCore\engine\base.py", line 83, in render_final
    final.render(self, depsgraph)
  File "C:\Users\309\AppData\Roaming\Blender Foundation\Blender\2.81\scripts\addons\BlendLuxCore\engine\final.py", line 49, in render
    _render_layer(engine, depsgraph, statistics, layer)
  File "C:\Users\309\AppData\Roaming\Blender Foundation\Blender\2.81\scripts\addons\BlendLuxCore\engine\final.py", line 61, in _render_layer
    engine.session = engine.exporter.create_session(depsgraph, engine=engine, view_layer=view_layer)
  File "C:\Users\309\AppData\Roaming\Blender Foundation\Blender\2.81\scripts\addons\BlendLuxCore\export\__init__.py", line 251, in create_session
    return pyluxcore.RenderSession(renderconfig)
RuntimeError: Hardware device selection string has the wrong length, must be 4 instead of 2
ERROR: Hardware device selection string has the wrong length, must be 4 instead of 2
Total render time: 0:00:00.435060
Estimated completion: 0:01:48.329887
Traceback (most recent call last):
  File "C:\Users\309\AppData\Roaming\Blender Foundation\Blender\2.81\scripts\addons\Scatter_S1.py", line 5498, in post
    proxy_scene_on()
  File "C:\Users\309\AppData\Roaming\Blender Foundation\Blender\2.81\scripts\addons\Scatter_S1.py", line 5464, in proxy_scene_on
    refresh_particles_show_hide()
  File "C:\Users\309\AppData\Roaming\Blender Foundation\Blender\2.81\scripts\addons\Scatter_S1.py", line 5568, in refresh_particles_show_hide
    for m in Terrain.modifiers:
AttributeError: 'NoneType' object has no attribute 'modifiers'
Info: Compiling OpenCL kernels (just once, takes a few minutes)
Error: Hardware device selection string has the wrong length, must be 4 instead of 2
- So I exported the scene, set .cfg to

Code: Select all

opencl.devices.select = 0011
ran luxcoreui.exe and loaded the scene...
and "Compiling" took forever = Kernels compilation time: 155375ms but when it showed up was already @ 2412 samples - :lol: great!!!
LC_CUDA.jpg
PS
Also note: OCL too, takes foooreeveeer to compile now :shock: :P
Post Reply