PC for Metashape in 2023

[NikosMavronichtis]

Hi,

In order to accelerate my workflow I am considering investing in a new PC and being able to use it for the next 4-5 years without issues.

Usually I use Metashape with a typical dataset of 20 laserscans and 1500 photographs. I need high details in models so the settings are never low.
I have also begin to learn Blender as mesh/texture editing tool (What's your opinion about that?? Any alternative recommendations are very welcomed!)

I have a limited budget and have in mind a base configuration a PC with:

CPU i13700k
RAM DDR5 64GB 6000Mz and
GPU 4070ti (12GB VRAM/7680 CUDA cores)


Is the above configuration (overall costs about 2600€ with the other essential components) decent for the projects I am working with?

Should  I need to spend extra money (maximum 500€) and if so, should this be on:

1) a GPU 4080 with extra 4GB VRAM and 2050 more CUDA cores (+500€)
2)  extra RAM of 64GB (+250€)
3) a CPU 13900k with extra 8 cores and extra 8 threads (+150€)


Thank you in advance

 

[jmaeding]

My experience is the extra 8 cores are definitely worth it.
64 gb ram should be plenty.
I have an nvidia card with 4 gb ram like you and its very good.
I would recommend a secondary internal SSD drive of 1 tb.
Run your projects on that, and then dump to external drive to make space later.
1500 photos is a lot, I typically do 400 or less for my drone flights.
The 20 laser scans part seems huge to me.
Maybe if doing mechanical stuff they are not too bad, our civil ones are 100 million a scan.

The other thing is its not great to buy high end now, and have it last 5 years.
Its better to go medium high now, and buy new in 3 years. Things change too fast.

[NikosMavronichtis]

Thanks! So the best worth spending money should be in a better CPU than the 13700k which is already decent I think.

If anyone has the 13700k or the 13900k, I would like a comment or recommendation about the utilization of the CPU by Metashape.

[Bzuco]

13700/13900 have 8 "big" cores with HT(hyperthreading) and rest "small" cores without HT.
You will benefit more from AMD Ryzen 9 5950X or never 7950X which both have 16 "big" cores with SMT(Simultaneous Multithreading).

GPU: more cuda cores matters, but you need to feed them enough with data. So 4070ti or better 4080 make sense if you will be calculating depth maps on ultra high or high. If you choice medium or lower quality, you will see GPU more in idle state(transferring data to GPU) then calculating state. It also depends on photo resolution you are taking.

[Medive]

Hi,

I would also recommend the Ryzen, in combination with the 4080 a good setup for both Metashape and Blender. Of course it always depends on what you want to do with Blender in the end, but with high resolution textures from photogrammetric models and a meshing with several million vertices it is a good choice.
You can always upgrade the RAM in a year or two if necessary.
If you are looking for a suitable system for the next 4-5 years and are just starting out with Blender, you will inevitably encounter scenarios where you will need the power.

[andyroo]

Seems like worth posting here. I recently did a couple builds - both Ryzen 9 7500x:

• 64GB 6GHz DDR5 RAM, 2x AMD RX 7900 XTX
• 96GB 5.2GHz DDR5 RAM*, 2x NVidia RTX 4090

*will be 192GB 5.2GHz RAM when my RMA is done, I had one bad stick.[/list]

They benchmarked pretty similarly (within 1%) - except for on tiled model and texture because I don't know how to get SPIR support with AMD drivers yet (but that's not a big deal for me because that's not in my workflow).

I benchmarked using Puget Systems' extended Metashape benchmark because it actually stress-tests the cards and the CPU a bit and does everything I care about and writes to a nice file so I can compare machines.

I attached a chart and the modified benchmarking script where I fork things depending on metashape versions. It covers 1.5.something to 1.6.6 then skips to 2.0 because I didn't build any new workstations for a couple years.

[Bzuco]

@andyroo
Can you check how the GPUs were utilized during depth maps calculation?
In my attachment top graph is medium quality(half the time GPU is not utilized) and bottom is high quality(much much better utilization)...I am using rtx 2060 super and 18Mpix photos.
How big(resolution) photos are in the puget bench used?  ...my gpu graphs are from my project.

[andyroo]

These images are also 18MP. I didn't look at individual GPU usage. Both machines are busy for the next several weeks at least.

[TexasUAS]

I've processed 15-20,000 photo data sets for orthophoto generation.

My setup is:

Ryzen 9 7950x
64GB DDR 5 running at 6000mhz
RTX 4090

I've found Metashape is very heavy in CPU processing, followed by GPU which only runs for certain operations such as depth map generation, followed by memory usage, which for large data sets I've seen all my memory be allocated. I've thought about upgrading but currently nothing above 64gb of DDR5 will run at 6000mhz.

Before I upgraded to the 4090 I had an RTX 4070 Ti OC, it is plenty for most users.