Luxcore 2.1 Benchmark
- FarbigeWelt
- Donor
- Posts: 1046
- Joined: Sun Jul 01, 2018 12:07 pm
- Location: Switzerland
- Contact:
Re: Luxcore 2.1 Benchmark
Benchmark 2.1 render if both devices are selected or second listed (gfx1010) only.
Other scenes render no matter which device is selected.
I think this is odd.
Does anybody have any idea?
Light and Word designing Creator - www.farbigewelt.ch - aka quantenkristall || #luxcorerender
MacBook Air with M1
MacBook Air with M1
Re: Luxcore 2.1 Benchmark
The OpenCL drivers for 5700XT are broken, at the release they weren't able to render some of the LuxMark scenes too, do they all work for you ?FarbigeWelt wrote: ↑Wed Jul 31, 2019 4:20 pm Above happens if I select first listed device (Hawaii) only.
Benchmark 2.1 render if both devices are selected or second listed (gfx1010) only.
Other scenes render no matter which device is selected.
Does anybody have any idea?
- FarbigeWelt
- Donor
- Posts: 1046
- Joined: Sun Jul 01, 2018 12:07 pm
- Location: Switzerland
- Contact:
Re: Luxcore 2.1 Benchmark
Adrenalin 2019 Edition 19.7.4 Optional from 29.07.2019.Dade wrote: ↑Wed Jul 31, 2019 4:32 pmThe OpenCL drivers for 5700XT are broken, at the release they weren't able to render some of the LuxMark scenes too, do they all work for you ?FarbigeWelt wrote: ↑Wed Jul 31, 2019 4:20 pm Above happens if I select first listed device (Hawaii) only.
Benchmark 2.1 render if both devices are selected or second listed (gfx1010) only.
Other scenes render no matter which device is selected.
Does anybody have any idea?
All LuxMark 3.1 scenes work on both cards (one choosen only) with all openCL setting, benchmarks and VR (all openCL paths, all engines).
Light and Word designing Creator - www.farbigewelt.ch - aka quantenkristall || #luxcorerender
MacBook Air with M1
MacBook Air with M1
- FarbigeWelt
- Donor
- Posts: 1046
- Joined: Sun Jul 01, 2018 12:07 pm
- Location: Switzerland
- Contact:
Re: Luxcore 2.1 Benchmark
This is a good idea and I tested it. Tonemap openCL device disabled and first device (Hawaii) rendered successful. It also does if tonemap device was set to second openCL device (gfx1010).
Light and Word designing Creator - www.farbigewelt.ch - aka quantenkristall || #luxcorerender
MacBook Air with M1
MacBook Air with M1
Re: Luxcore 2.1 Benchmark
GTX 2080Ti + Titan X (Pascal) + i7-9700K: 3 min 34s
Photon GI
However with Photon GI, Random sampler and noise threshold halt at 30, it rendered in 38.2 seconds, which is pretty amazing.
EDIT: Turning off the Automatic Lookup Radius and setting it to 50cm fixed the reflections.
I gave Photon GI a try but it did change the look of the render quite a lot. The floor bounce a lot more of warm light, which doesn't look bad, but what is more surprising is that the reflections are different? The bottom left corner where the wall meets the floor is very different.
ReferencePhoton GI
However with Photon GI, Random sampler and noise threshold halt at 30, it rendered in 38.2 seconds, which is pretty amazing.
EDIT: Turning off the Automatic Lookup Radius and setting it to 50cm fixed the reflections.
-
- Donor
- Posts: 790
- Joined: Thu Oct 04, 2018 6:06 am
Re: Luxcore 2.1 Benchmark
Bought a used MacBook Air M1, base version: 8CPU/7GPU cores, 8GB RAM+256GB SSD (modded with thermal pad).
Tested against Ryzen 3900X @ 65W TDP +2x RTX 2070 @ 130W TDP while measuring wall power in Watt.
Benchmark scene @ 256 samples, no denoiser, SOBOL progressive is faster for Mac while SOBOL cache friendly is better with Win PC.
While everyone is waiting for Cycles on Metal, LuxCore has already a working GPU solution which, running on OpenCL, might not even take a significant Rosetta toll which is ~20% for CPU render.
Tested against Ryzen 3900X @ 65W TDP +2x RTX 2070 @ 130W TDP while measuring wall power in Watt.
Benchmark scene @ 256 samples, no denoiser, SOBOL progressive is faster for Mac while SOBOL cache friendly is better with Win PC.
While everyone is waiting for Cycles on Metal, LuxCore has already a working GPU solution which, running on OpenCL, might not even take a significant Rosetta toll which is ~20% for CPU render.
Re: Luxcore 2.1 Benchmark
Very interesting to see render performance together with power consumption
Thanks.
Thanks.
-
- Donor
- Posts: 790
- Joined: Thu Oct 04, 2018 6:06 am
Re: Luxcore 2.1 Benchmark
Nope, the CPU is about half as fast natively, in LuxMark even worse because of ~20% Rosetta (X86 to ARM transl.) penalty. Above I have measured the average wall power during the main render (w/o scene setup) and multiplied with the render time to get energy amount used for the LuxBench render job disregarding time.
Here the M1 (CPU+GPU) is 1:1.67 vs Ryzen 3900 +2x RTX2070.
On Blenderartists I'm trying to get owners of the new M1 Pro/Max hardware to join in LuxCoreRender testing.