LuxCoreRender Forums

Hi to all, this is my first post here.
I have a question regarding LuxCorerender performance on xeon dual processor servers.
The test escene is LuxCore 2.1 Benchmark downloaded from the download area.
OS is debian 64 buster
Render engine type Pathcpu
Sampler Sobol

Dual Xeon 5520 ,3 ghz (8 cores)= 745 sec
Dual Xeon 2683 V3, 2 ghz (24 cores) = 235 sec
Dual Xeon 2697 V4, 2.2 ghz (36 cores) = 227 SEC

Anyone has an idea why there is almost no performance increase on the 2697 V4 xeon compared to the old 2683 V3 Xeon?
If you ask why i'm interested for CPU rendering it is because i'm looking only for bidirectional+metropolis.
Thanks for any advise.
David

Dual Xeon 2697 V4, 2.2 ghz (36 cores) should be quite faster than the 24 cores Xeon, at least according to geekbench benchmarks.

Did you check if all cores are supported/listed by the LuxCoreRender 2.1 benchmark?

Anyone has an idea why there is almost no performance increase on the 2697 V4 xeon compared to the old 2683 V3 Xeon?
If you ask why i'm interested for CPU rendering it is because i'm looking only for bidirectional+metropolis.
Thanks for any advise.
David

Clear that the V4 should do alot better i think.

So why not threadripper 3970X/32 cores / 64 threads/ 3.7GHz ?

because i already have the hardware. I just want to use it with Luxrender and i want to understand Luxrender's behaviour. Thanks!

thanks FarbigeWelt, i normally use the Spec 2017 fpRates which are quiet precise too.
Wen rendering with luxcore the 2697 uses 100% of the cpus almost all the time. that's why i'm a bit lost. Should be at least 30% faster.

following are the render settings i used:

# For a description of the CFG properties, visit:
# https://wiki.luxcorerender.org/LuxCore_ ... anual_v2.1

# Do not use OpenCL CPU devices (native C++ rendering is faster)
opencl.cpu.use = 0
# Use all GPU devices we can find
opencl.gpu.use = 1
# You can use this setting to specify exactly which OpenCL devices to use
#opencl.devices.select = "1"

path.pathdepth.total = 13
path.pathdepth.diffuse = 7
path.pathdepth.glossy = 7
path.pathdepth.specular = 12
sampler.sobol.adaptive.strength = 0.94999998807907104
sampler.random.adaptive.strength = 0.94999998807907104
sampler.metropolis.largesteprate = 0.40000000000000002
sampler.metropolis.maxconsecutivereject = 512
sampler.metropolis.imagemutationrate = 0.10000000000000001
sampler.type = "SOBOL"
film.opencl.enable = 1
film.width = 733
film.height = 994
film.filter.width = 1.5
film.filter.type = "BLACKMANHARRIS"
lightstrategy.type = "UNIFORM"
scene.epsilon.min = 9.9999997473787516e-06
scene.epsilon.max = 0.10000000149011612
renderengine.type = "PATHCPU"
path.clamping.variance.maxvalue = 10
path.forceblackbackground.enable = 0
renderengine.seed = 1
batch.haltspp = 500
batch.haltthreshold.stoprendering.enable = 0
batch.haltthreshold = 0.0001
batch.halttime = 0
film.imagepipelines.1.0.type = "BCD_DENOISER"
film.imagepipelines.1.0.scales = 3
film.imagepipelines.1.0.histdistthresh = 0.7
film.imagepipelines.1.0.patchradius = 1
film.imagepipelines.1.0.searchwindowradius = 6
film.imagepipelines.1.0.filterspikes = 0
film.imagepipelines.1.1.type = "NOP"
film.imagepipelines.1.2.type = "TONEMAP_LINEAR"
film.imagepipelines.1.2.scale = 1
film.imagepipelines.1.3.type = "COLOR_ABERRATION"
film.imagepipelines.1.3.amount = 0.0020000000298023225
film.imagepipelines.1.4.type = "VIGNETTING"
film.imagepipelines.1.4.scale = 0.20000000000000001
film.imagepipelines.1.5.type = "GAMMA_CORRECTION"
film.imagepipelines.1.5.value = 2.2000000000000002
film.imagepipelines.0.0.type = "NOP"
film.imagepipelines.0.1.type = "TONEMAP_LINEAR"
film.imagepipelines.0.1.scale = 1
film.imagepipelines.0.2.type = "COLOR_ABERRATION"
film.imagepipelines.0.2.amount = 0.0020000000298023225
film.imagepipelines.0.3.type = "VIGNETTING"
film.imagepipelines.0.3.scale = 0.20000000000000001
film.imagepipelines.0.4.type = "GAMMA_CORRECTION"
film.imagepipelines.0.4.value = 2.2000000000000002
film.imagepipelines.1.radiancescales.0.enabled = 1
film.imagepipelines.1.radiancescales.0.globalscale = 1
film.imagepipelines.1.radiancescales.0.rgbscale = 1 1 1
film.imagepipelines.0.radiancescales.0.enabled = 1
film.imagepipelines.0.radiancescales.0.globalscale = 1
film.imagepipelines.0.radiancescales.0.rgbscale = 1 1 1
film.outputs.0.type = "RGB_IMAGEPIPELINE"
film.outputs.0.index = 0
film.outputs.0.filename = "RGB_IMAGEPIPELINE_0.png"
# Denoiser imagepipeline
film.outputs.1.type = "RGB_IMAGEPIPELINE"
film.outputs.1.index = 1
film.outputs.1.filename = "RGB_IMAGEPIPELINE_1.png"
scene.file = "scene.scn"

daros wrote: ↑Thu Dec 12, 2019 5:53 pm following are the render settings i used:

# For a description of the CFG properties, visit:
# https://wiki.luxcorerender.org/LuxCore_ ... anual_v2.1

# Do not use OpenCL CPU devices (native C++ rendering is faster)
opencl.cpu.use = 1
# Use all GPU devices we can find
opencl.gpu.use = 0
# You can use this setting to specify exactly which OpenCL devices to use

Your config file was wrong for the use of CPU only change accordingly to bold font above.

FarbigeWelt wrote: ↑Fri Dec 13, 2019 3:16 pm # Do not use OpenCL CPU devices (native C++ rendering is faster)
opencl.cpu.use = 1
# Use all GPU devices we can find
opencl.gpu.use = 0
# You can use this setting to specify exactly which OpenCL devices to use

Your config file was wrong for the use of CPU only change accordingly to bold font above.

FarbigeWelt, OpenCL settings don't matter because they are using PATHCPU (i.e. C++ code, CPU-only rendering):

Dade wrote: ↑Fri Dec 13, 2019 3:35 pm
FarbigeWelt, OpenCL settings don't matter because they are using PATHCPU (i.e. C++ code, CPU-only rendering):

Code: Select all

renderengine.type = "PATHCPU"

Thank you Dade for the correction.

Dade solved our problem i think. In our test we have been using a scene which uses the old denoiser which is slow and almost single core. This caused a bottleneck on dual xeons with many cores. The intel denoiser is super fast and apparently allows to scale up more linearly. Monday we complete our test round and i will confirm. Thanks to all in the meanwhile.