Lighthouse2 / RTX path tracing benchmark

Started by jbikker, February 17, 2020, 05:17:44 PM

Previous topic - Next topic

0 Members and 1 Guest are viewing this topic.

jbikker

An alpha build of the Lighthouse 2 path tracing benchmark is now available:
http://www.cs.uu.nl/docs/vakken/magr/2019-2020/files/Lighthouse2%20RTX%20Benchmark%20vA0.2,feb17.zip
The source code for this package is available from github:
https://github.com/jbikker/lighthouse2
Lighthouse 2 is open source.

JeGX

Thanks for the news, I moved the topic in this forum  ;)

jbikker

Thank you. :)
I uploaded a better version; next event estimation greatly improves the image quality.
Video:
https://t.co/EFyjpJT9Ca?amp=1
Executable package:
https://t.co/ukAlREPJqh?amp=1
Source code:
https://t.co/uyke4Uzc2W?amp=1

Updates on twitter, @J_Bikker.

nuninho1980

#3
@jbikker

Thank you. :) How do I increase spp (samples per pixel) from "2" to "32" for non-RTX vA0.3 executabled (extension ".exe")? - Because I get min 30 and max 55 fps running my GTX 980 Ti SC+@VRAM 3505MHz.

jbikker

Add a command line parameter, e.g. benchmark.exe 4 will render using 4spp.
Alternatively, get the project from github; at the top of main.cpp in the benchmark project you will find the default spp. This will also let you play with other 'render cores'. For pre-RTX there is the OptixPrime core; for RTX there is the Optix7 core but also the filtered Optix7 core and a Vulkan core. Modifiying and creating cores is what Lighthouse 2 is for. :)

nuninho1980

Quote from: jbikker on February 18, 2020, 12:02:31 PM
Add a command line parameter, e.g. benchmark.exe 4 will render using 4spp.
command line (cmd) running Windows 10 x64:
C:\Users\ricar\Desktop\Lighthouse2 RTX Benchmark>"benchmarkapp 2spp (NORTX).exe" 32
execute this application with 32 spp but it can't run.

jbikker

You need a lot of VRAM for 32spp. What's your card?

nuninho1980

#7
Quote from: jbikker on February 18, 2020, 07:27:48 PM
You need a lot of VRAM for 32spp. What's your card?
I've said you my videocard. lol

C:\Users\ricar\Desktop\Lighthouse2 RTX Benchmark>"benchmarkapp 2spp (NORTX).exe" 2
2 spp as default but I can't run (Window doesn't appear and 100% CPU usage). Too high VRAM doesn't cause that issues. 2 spp uses 1347 MB of VRAM.

Stefan

The pre-compiled binary doesn't seem to accept command line parameters.

Edit main.cpp as suggested and rebuild the code
scrspp = 16 uses about 3.5 GB VRAM

After a minute my laptop starts thermal throttling  :P



jbikker

#9
Version B0.5 of the benchmark is now available, which actually makes it a benchmark:

https://twitter.com/j_bikker/status/1230815837784805376

Let me know how your GPU fares, especially if you have an overclocked or otherwise extreme machine.

- Jacco.





(EDIT: reduced image size to reasonable)

Stefan

#10
benchmarkapp 2spp (preRTX).exe  fails with Pascal GPU

7FF8C3A30000: loaded D:\Downloads\Lighthouse2 RTX Benchmark vB0.5,feb21\coredlls\release\RenderCore_Optix7.dll
7FF887F40000: loaded D:\Downloads\Lighthouse2 RTX Benchmark vB0.5,feb21\nvrtc64_102_0.dll
7FF8F6730000: loaded C:\WINDOWS\SYSTEM32\dbghelp.dll
7FF88BBB0000: loaded C:\WINDOWS\system32\nvcuda.dll
7FF8E7B20000: loaded C:\WINDOWS\SYSTEM32\nvapi64.dll
7FF8F6B10000: loaded C:\WINDOWS\SYSTEM32\dxgi.dll
7FF880550000: loaded C:\WINDOWS\System32\DriverStore\FileRepository\nvami.inf_amd64_ef3e459f58adb6da\nvoptix.dll
7FF87EAF0000: loaded C:\WINDOWS\System32\DriverStore\FileRepository\nvami.inf_amd64_ef3e459f58adb6da\nvrtum64.dll
7FF89E1D0000: loaded C:\WINDOWS\system32\nvfatbinaryLoader.dll
7FF8A3360000: loaded C:\Program Files\NVIDIA Corporation\NVSMI\nvml.dll
7FF8FA8C0000: loaded C:\WINDOWS\System32\psapi.dll
7FF8F870C4C8: unknown exception code C0000409 (exc.code c0000409, tid 21188)

EDIT:



JeGX

Here is my score with the new version:

Device: GeForce RTX 2070 (#SM: 36)
Compute capability: 7.5
Device RAM: 8GB
Device RAM: 8GB
Rendercore: RenderCore_Optix7.dll
--------------------------------
samples per pixel:       2
screen size:         1600 x 900
average fps:         91.21
peak fps:            142.27
avg rps (primary): 1747.16M
peak:              2352.70M
avg rps (shadow):  1955.96M
peak:              2815.78M
Checkpoint #0: 0.014
Checkpoint #1: 0.014
Checkpoint #2: 0.195
Checkpoint #3: 0.013
Checkpoint #4: 0.201
Checkpoint #5: 0.177
Average RMSE: 0.102
--------------------------------
#frames recorded:     6699



You should compile the benchmark with /MT option, I don't like installing the VC++ runtimes....
Cool benchmark!

jbikker

Thanks. I fixed the issue; asynchronous material sync went wrong. New version:
http://www.cs.uu.nl/docs/vakken/magr/2019-2020/files/Lighthouse2%20RTX%20Benchmark%20vB0.6,feb23.zip

Quote from: Stefan on February 21, 2020, 05:26:00 PM
benchmarkapp 2spp (preRTX).exe  fails with Pascal GPU

7FF8C3A30000: loaded D:\Downloads\Lighthouse2 RTX Benchmark vB0.5,feb21\coredlls\release\RenderCore_Optix7.dll
7FF887F40000: loaded D:\Downloads\Lighthouse2 RTX Benchmark vB0.5,feb21\nvrtc64_102_0.dll
7FF8F6730000: loaded C:\WINDOWS\SYSTEM32\dbghelp.dll
7FF88BBB0000: loaded C:\WINDOWS\system32\nvcuda.dll
7FF8E7B20000: loaded C:\WINDOWS\SYSTEM32\nvapi64.dll
7FF8F6B10000: loaded C:\WINDOWS\SYSTEM32\dxgi.dll
7FF880550000: loaded C:\WINDOWS\System32\DriverStore\FileRepository\nvami.inf_amd64_ef3e459f58adb6da\nvoptix.dll
7FF87EAF0000: loaded C:\WINDOWS\System32\DriverStore\FileRepository\nvami.inf_amd64_ef3e459f58adb6da\nvrtum64.dll
7FF89E1D0000: loaded C:\WINDOWS\system32\nvfatbinaryLoader.dll
7FF8A3360000: loaded C:\Program Files\NVIDIA Corporation\NVSMI\nvml.dll
7FF8FA8C0000: loaded C:\WINDOWS\System32\psapi.dll
7FF8F870C4C8: unknown exception code C0000409 (exc.code c0000409, tid 21188)

EDIT:



Stefan

#13
Benchmark completes now successfully.

CLI reads Optix 5, but results.txt reads OptiX7, even if you delete RenderCore_Optix7.dll from subfolder.
"Device RAM" is listed twice.
Quote
Device: GeForce GTX 1060 (#SM: 10)
Compute capability: 6.1
Device RAM: 6GB
Device RAM: 6GB
Rendercore: RenderCore_Optix7.dll
--------------------------------
samples per pixel:       2
screen size:         1600 x 900
average fps:         10.71
peak fps:            20.12
avg rps (primary): 117.92M
peak:              193.32M
avg rps (shadow):  124.61M
peak:              194.26M
Checkpoint #0: 0.201
Checkpoint #1: 0.185
Checkpoint #2: 0.218
Checkpoint #3: 0.193
Checkpoint #4: 0.220
Checkpoint #5: 0.120
Average RMSE: 0.190
--------------------------------
#frames recorded:      789


nuninho1980

#14
My result on my EVGA GeForce GTX 980 Ti SC+@VRAM 3505MHz with Gf 441.87 driver running Win 10 x64 1903 and i7-4790K@4.4GHz:

LightHouse v2.0 - vB0.6

jbikker

Interesting to see that the 980Ti is so much faster than the more recent 1060.
I'll assemble a little table tomorrow and put it on a proper website.
Thanks for posting your results!

Stefan

Quote from: jbikker on February 25, 2020, 12:03:55 PM
Interesting to see that the 980Ti is so much faster than the more recent 1060.
Seems to be dependent on the amount of SMs

I noticed another issue:
At result screen GPU clock is stuck at 1809 MHz, which is faster than my boost clock (1671 MHz).
After exiting clock drops as expected.





jbikker

The benchmark has been updated. The new version has less noise: in fact, the 10spp version now has a better image than the 24spp in the previous version. Performance also improved; framerate for 10spp is about the same as for 8spp in the previous version. Download:

https://jacco.ompf2.com/lighthouse-2-rtx-path-tracing-benchmark/

Image: close-up of the blue noise sampling pattern used in this demo.


Stefan

Result screen shows now correct rendercore for Pascal.
But current clock still exceeds boost clock.