Methods for testing rendering speed

[QizW]

Hello, when I was reproducing your code, I found that the FPS tested by my rendering function is much lower than the results in your paper. Could you please advise how to resolve this issue?

[fastgs]

Hello, I first need to confirm a few things:

1. Is the reported training time correct?
2. What operating system and GPU are you using?

[QizW]

Hello, I first need to confirm a few things:

1. Is the reported training time correct?
2. What operating system and GPU are you using?

I conducted the test on a single 4090 on Windows11, with no other programs running during the test, and the Cuda version was 11.8.
During training, my training time was also longer than the time mentioned in the text, and I found that my GPU utilization did not reach 100% (about 60%).

I am also very troubled by this issue.

[fastgs]

Sorry! I haven’t tested this on a Windows system, so I’m not exactly sure what the cause is. If you have access to an Ubuntu system, you could try running it there. Otherwise, I can try testing it on Windows later to see what the issue might be.

[QizW]

Hello！ I attempted to run your code on the Liux system, but the rendering and training speeds still failed to reach the metrics mentioned in your text. Below are my environment settings and the log output. It seems that the GPU is not being fully utilized. I would like to ask if you have any better suggestions to help me modify the environment or the relevant code to speed up the process.
The data set I tested was the base mode of the "train" dataset.

[fastgs]

Hello! You can try setting up the environment directly using the command conda env create --file environment.yml. You may also refer to issue #11. If you have other servers available, you can try running it on those as well.

[QizW]

I have tried to set up the environment directly using the command conda env create --file environment.yml on an Ubuntu system. But the problem is the same，even the speed has become slower... I found that the power and utilization rate of the GPU can only reach half. Could you please offer any suggestions for improving or accelerating the process of file reading or CUDA processing?

[Shiwei-Ren]

Hello! The issue you encountered is indeed quite strange. Could you send your WeChat ID to renshiwei@mail.nankai.edu.cn? We can communicate via WeChat.

[Assld]

Has this issue been resolved? I encountered the same problem. I tested it on two different 4090 servers, but I still couldn't reproduce the training speed, and the GPU utilization was also 60%. I asked GPT, and it said it might be a CPU I/O bottleneck issue. The CPU frequency of my servers is 2.10GHz. @fastgs @QizW @Shiwei-Ren

[Shiwei-Ren]

@Assld Hello! The issue has not been resolved yet, as I don’t have access to a Windows-based machine for debugging. The CPU I’m using is an AMD Ryzen Threadripper PRO 5995WX with 64 cores, with a base frequency of 2.70 GHz.