GeoSLAM Hub processing speeds

[Jonagoldie]

In search of faster point cloud processing speeds, we upgraded our machine from 6-core Intel i5 8600k to 16-core AMD 7950x.

For our surprise, there was literally no difference in processing time. How can this be possible? Is there something we are missing? Seemed like processor was not utilized properly, on i5 8600k all of the cores had reasonable utilization, now on 7950x only few of the cores are being utilized and total utilization of cpu went from around 60% to 15%..

[Kruse]

Just talking CPU's for now, but there could be other factors such as RAM, Storage, GPU, cooling, etc...

This is most likey due to how the software operates. Like any application, some rely more on a single core (or 2 or 3) so what makes a bigger difference is that you choose a CPU that has the highest single core speeds (7950X can boost to 5.7GHz while a 14900K can boost to 6GHz compared to your 8600K's max boost of 4.3GHz). Other applications are better able to use all the cores and could take advantage of your 7950X's more cores/threads and would definitely be faster. Some applications are a hybrid and will bounce between single core and multi core loads, so you need to choose a processor that will be more well rounded.

Every software is different in what they need to best optimize the processing times... I use Register 360 and a computer with with a 14900K CPU will run circles around a Threadripper 5995WX with around 4 times the amount of cores/threads simply because it has a balance of fast single core speeds, but enough multithreaded power to efficiently chug through the data. Having more cores doesn't always mean more performance if the software can't use all those cores. In this case, you might have been better off choosing something like a 14900K for slightly faster single core speeds.

I'll let others reply back with specifics on your software if there are settings or things to change, but just my two cents for now.

[Jonagoldie]

Virtually all the components got an upgrade except the GPU. RAM went from 32gb of ddr4-3600 to 96gb of ddr5-6400 (twice the speed), storage went from Samsung 970 m.2 ssd to Crucial T700 pcie 5.0 m.2 ssdm almost 10 (about 5 times the speed)

We used AMD:s integrated GPU instead of RTX 3090 of the old system since we observed no GPU load when processing.. I doubt this would make a difference..

Some kind of improvement in processing speed would surely be expected no matter what, unless there is a clear preference for Intel cpus..

[max72]

Can we informally cross check processing time of the same project or set of scans?
Maybe we can get a better understanding of the hardware influence. I have an i7, and I'm pretty pleased of the processing time.
Massimo

[MarkJLane]

now on 7950x only few of the cores are being utilized and total utilization of cpu went from around 60% to 15%..

Hi Joona, GeoSLAM Hub and Connect benefit more from core speed than core count, so the extra cores doesn't necessarily provide a performance improvement. The software doesn't use all cores, which explains why your CPU utilization went down when you moved to more cores.

If you'd like to accelerate your processing, you could try limiting your import queue size to one and maybe even elevating the priority of nrrslam.exe in task manager.

[Jonagoldie]

now on 7950x only few of the cores are being utilized and total utilization of cpu went from around 60% to 15%..

Hi Joona, GeoSLAM Hub and Connect benefit more from core speed than core count, so the extra cores doesn't necessarily provide a performance improvement. The software doesn't use all cores, which explains why your CPU utilization went down when you moved to more cores.

If you'd like to accelerate your processing, you could try limiting your import queue size to one and maybe even elevating the priority of nrrslam.exe in task manager.

Is there any multithreading improvements to be expected in near future for the Connect?