i have 13700k, 64gb ram, 4090. Fusing even a small scan takes eons because it isn’t even loading the CPU up to barely 15%. I also compared this to my laptop with 4060, ultra 9 285, 64gb ram.
is there any setting im missing that makes this take longer. Coming from competitor's scanner I have never experienced such slow processing of the point cloud even for gigantic scans.
The slow fusing is because of the Laser mode and the amount of very small frames to be recalculated to rebuild the model , a slightly different method than when using Structure light mode where the frames are bigger and less frames us needed.
Laser mode can produce huge data and since it is only CPU based software , the GPU is not used in calculation.
How higher the fusing settings how longer it will take .
Try to avoid capturing extra things beside your actual object by selecting the proper brightness for your order settings .
It will work slightly faster but still not as fast as you want it to be .
Use also your system "hard drive " avoid external hard drivers .
Don’t use too high fusing settings if the details are not there. Measure the smallest detail on your object and set the fusing according to it .
For now that’s all you can do .
why is it not taking advantage of full CPU capability then?
If the CPU would run at full capacity your system RAM would run out and crash with this South of frames, right now the algorithms are set this way and not another for fusing raw scans scanned using Laser mode .
The slower processing preserving the system RAM as well.
Don’t forget that models with this amount of data using lots of processing time in other applications as well .
This don’t means that it will be this way for ever as the team is working on improvements all the time.
I think he means - wouldn’t it be possible to have a slider that selects how much cpu and ram the application uses.. 15% cpu usage is tiny.. Even excel uses more when recalculating. CPUs are multithreaded these days, so couldn’t multiple threads be used to speed everything up? Even if you don’t use the GPU
Hi @Vassago
Sorry but I don’t have any other answer for it at this moment .
The Dev. Team is working on the improvements and will find better way.for speed up the fusing mode for laser mode scans in the future.
As I said already so many times , this is related to Fusing of Laser mode scans , specifically with laser mode frames , the other mode and functions working optimal .
At the very least you could improve the timing estimates, so the user knows how many hours they have to be more productive doing something else. But you should really just fix this, performance is abysmal even on an exctremely high-end PC.
I was fusing on my dell pro max w/ an ultra 9 285hx and 64g of ram on a 9100 samsung drive. it does take a while, but it’s not hours, it’s minutes. maybe 10-15 for big scans.
I did some scans with 60k frames and it took like 10-15min to fuse. A long time for sure it seems to take, but it wasn’t hours and that’s a ridiculous amount of frames.
I would also like to see more cpu used, but if I run out of ram, IDK what will happen. Maybe there’s some sort of balance in the multi threaded environment. I’m sure they are working on it tho.
Hi Robin ,
Unfortunately the fusing is slow and that has nothing to do with his much CPU it uses , higher CPU usage would totally consume the RAM even at 64GB it would barely make it for that reason it is so slow. On my early beta testing it almost used 57GB in small scans , so definitely not the way. Laser scanning is time and resources consuming
Fusion only uses 2-4 of 24 CPU cores — GPU essentially idle — memory leak across batch jobs causes full freeze/hang
System: Intel Core Ultra 9 275HX (24 logical cores), 64GB RAM (Have 128gb kit going in next) Still will be the same im sure, RTX 5090 Laptop GPU, RevoMetro 5.8.7.415 (latest), MetroY Pro, power plan set to Ultimate Performance (core parking off, ruled out as a factor).
CPU usage during fusion:
While actively fusing a scan, RevoMetro spawns 200+ threads, but only 2-4 logical cores ever show real load (measured 90% / 81% on two cores, ~33% on a couple more, everything else idle). CPU time consumed over an 8-second sample worked out to roughly 3.4 cores’ worth of actual work — on a 24-core CPU. The fusion step appears to be single/lightly-threaded despite the huge thread count spawned.
GPU usage during fusion:
nvidia-smi shows 0% GPU utilization during fusion, with only ~2GB VRAM parked (just holding the viewport/3D scene, not doing compute). The GPU selector in settings is set to the 5090, but it’s not being used for the actual fusion math — only for rendering the environment.
Memory leak across a batch queue → full hang:
Queued 10 scans for fusion overnight. First scan peaked at ~18GB RAM (reasonable). By the time it reached the scan that eventually failed, working set had climbed to 41.6GB and then flatlined completely — 0% CPU, memory pinned, for 19+ minutes — before Windows force-closed it with an “Application Hang” event (RevoMetro.exe stopped responding). Memory not being released between sequential fusion jobs in a batch looks like the root cause. Worse: there’s no autosave/checkpoint mid-fusion, so the hang wiped out that scan’s fusion progress entirely — had to redo it from scratch, individually.
Ask: Any plans to parallelize the fusion pipeline across more cores, offload more of it to GPU compute (not just viewport rendering), and/or fix whatever’s leaking memory across queued batch jobs? Happy to share raw logs/numbers if useful for reproducing.
It would be nice to have the same features that the full field has for key frame edits before fusion in laser mode.
“Key Frame Edit is Full Field/Auto Mode only, with Laser Scanning Mode explicitly called out as unsupported since it uses a different direct-collection pipeline.”
Hi @ahowell95
Right now not possible , it will consume too much RAM of your system if the usage of the cores is increased for fusing laser scans . Full Field use totally different technology and very small amount of frames in comparison to laser mode .
Right now the software is only CPU based , GPU is only used for GPU acceleration and generate of 3DGS models .
Hopefully things change in the future .
Also feel that this should be an option that can be turned on for users with beefier systems or workstations. Instead of just stuck at the “base” allowable default everyone’s on.
I do agree , but that would be only users that have above 64GB of RAM on their system , hopefully team find another way to speed up the Laser scanners processing in the future .
thats fine
if only there was some kind way software could determine how much ram the users system has installed
Hi @bxp
This will not change anything , because the higher portion of the RAM is used by the system , the graphic card and the Revo software project. What is left will be not enough even for systems with 64GB of RAM , trust ne I tested it long before release , it was totally not worthy.
They need to find a better algorithm to process it or add GPU to processing. And the second would speed things for sure.
I am processing models in other 3D software , no matter if it is Cloud Compare or Zbrush or any other CPU based processing software it is just very slow in processing.
Now with structured light scanning you have 360 frames to process with laser 36 000 that is huge difference in processing time
For that reason I was never a big fan of a laser scanning , markers and long processing time for this technology. Something for something.
Now the Trackit SR give better opportunity with scanning if only the processing was faster even half way up.
Just remember, it is nothing Dev. Team did wrong with , there is currently not much options to speed it up at this moment.