CloudCompare forum

Noobie question: CC was running as well as could be expected (PC built in fall of '19: Ryzen 5 2600, 64MB ram, RX 580), but recently started bogging down and hanging up. GPU is maxing out, but RAM is running at max 28%. The CPU looks bored and does hardly anything. Not sure if some GPU settings need resetting or if its some other global issue. Working on a cloud with 280 M points, just trying to segment and getting a slew of things that are "not visible in the active 3D view!" If I accept my clip, I eventually get a "[ccPointCloud;;resize] Not enough memory!" Since my GPU is maxing out, I'm assuming that it means my GPU memory. But it leaves me wondering what a bunch of RAM does if your bottleneck is the GPU memory.

I'm going to guess I'm wading into waters that are over my head... but I'm really hoping its something easy and obvious (ha!). I'd say my system was undersized (which it likely is), but it was working adequately up until recently. Ah, technology!

Thoughts?

Thanks all!

b

Which release are you using?

And did your issues appear with a specific driver update maybe?

And no, ccPointCloud::resize should only care about the main memory. That's a bit weird.

Hey Daniel,

Thank you for your reply!

Looks like I'm running 2.10.2. I'll try upgrading an see if that helps. BIOS was just updated and I'm running the most up to date GPU drivers.

b

ps, feel free to ignore the second version of this post I sent this morning. I should have expected that the first one would be reviewed before it posted.

Hey Daniel,

Just thought I'd let you know that I found a solution to my memory issue here:

https://windowsreport.com/windows-10-is ... g-all-ram/

CC is still running slow, but it is working!

Thanks again for you help!

b

Hey Daniel and all,

Just a follow up here. After continuing to feel like my system just wasn't robust enough for point cloud work, I upgraded last week to a Ryzen 9 3950x (from Ryzen 5 2600), and a Radeon Pro wx7100 (from rx580), ram is still 64 gb at 2133 on an ASUS Prime B450A MB. Clean install of W10 too.

Programs are running a bit better, but well below my expectations. Even with relatively small clouds (<30M points) CC (and the rest of the PC processing softwares I've been evaluating this month, Scene, TBC, etc) bogs down just trying to zoom around.

I'm clearly out of my depth and don't know don't even know how to identify the bottleneck, if there is one. My task manager suggests that my CPU and/or my GPU and/or my RAM is being throttled somehow. None of them rise above around 25% when I'm processing. Maybe this is normal, I don't know. Perhaps my expectations of my PC's performance are just too high. But while my PC build certainly isn't a beast, given what I have read about other folks builds, I expected to be able to manage small clouds with little issue.

I've run Speccy, but nothing jumps out and I don't know enough to know what to look for.

It is worth noting that performance increases dramatically in CC when I chop my cloud down to just a small portion.

The only other culprit I can think of is knowing that my existing RAM is slower than my CPU (2133 vs 3200). But, from what I've read, that should make that much of a difference.

If it is what it is, and I'm probably getting as much performance as I can expect, then so be it. But if anyone has a rabbit-hole they think is worth jumping down, I'm at a loss for what to do next.

Thanks all!

Not sure if that can help in your case, however I turned off 'Decimate clouds/meshes over' option. While working with large clouds (>500M) decimating was taking too much time and memory.

I always have decimation turned off, as long as vertex buffers are being used. As GPU memory rises I feel like we might want to change the default options, if there is room for the point cloud to live in the GPU vertex buffers then it should never be decimated.

I recently had a discussion with people at AMD. I myself observed very bad performances with a top notch pro Radeon card (which was like 3 to 6 times slower than an NVidia GeForce 1080). They benchmarked CC's pipeline and realized their OpenGL driver didn't like at all the way we are displaying texts in the 3D view (as small textures, recreated at each frame).

They told me that fixing this should make AMD graphic cards much more responsive... But it will ask quite some work to do it on our side! So AMD users will have to wait a little bit (and pray that this is the only cause of downgraded performances for AMD cards of course...).

Hi Daniel,

I discovered this thread after our conversation the other day regarding the issues I was having loading meshes between 0.9 Gb and 2.5 Gb. It turns out we are using the AMD Radeon Pro Vega Duo II graphics card on a Mac Pro configuration (1 card each in slots 1 & 2). I was wondering if there has been any resolution to the issue that you mentioned below? Also, I saw in some other posts that CloudCompare only loads the image into one GPU, is there a way to utilize both GPU's to optimize performance?

Thanks for being so responsive on this forum, it's very helpful!

Yes, normally the performances have been improved on ATI/AMD cards. It was due to the way we were displaying texts on the 3D views...

And I don't think we can dispatch / use the 2 GPUs (we use standard OpenGL). Moreover, mind that we only load clouds on the GPU, not meshes.