Suitable hardware for Agisoft matashape Pro ver 2.0.2

[Nuiosk]

I will assemble a new PC to process 300- 1000 Agisoft photos. I also spent hours on this article.
https://www.pugetsystems.com/labs/articles/Agisoft-Metashape-1-6-5-AMD-Ryzen-5000-Series-Performance-1963/

And found the incredible performance of the Ryzen 7 5800X. However, the results run on Agisoft 1.6.5.

My question is: Does the Ryzen 7 5800X still work well on Agisoft Matashape Pro version 2.0.2?

If yes, I wish to run a Ryzen 7 5800X coupled with Radion 6700XT  on DDR 4 64 GB for 300- 1000 Agisoft photos. Otherwise, should I shift to Ryzen 7 7900X, 7900X3D, or Ryzen 9 5950X? 

Any suggestions are welcome.

Thank you very much.

[Bzuco]

I am on similar computer(but on INTEL and Nvidia platform) and 300-1000 photos are totaly OK. Better CPUs will help you if you will be mostly creating only pointclouds, where CPU cores make a lot of sense.

[Nuiosk]

I am on similar computer(but on INTEL and Nvidia platform) and 300-1000 photos are totaly OK. Better CPUs will help you if you will be mostly creating only pointclouds, where CPU cores make a lot of sense.

What about the number of MB of L3 cache?  Is it significantly effective on the CPU's performance on Matashape Pro? Ryzen7900X3D has 96 MB of L3 cache while Ryzen7900X has 64 mb.

[Bzuco]

I have concerns that metashape code is not using L3 cache at all in terms of vectorizing data using modern processor instructions. The processed data are large and in uncompressed formats stored in RAM and cannot fits even in large L3 cache. Maybe if the data are processed at smaller pieces, this is question for devs.

7900x has 32MB cache for each 8-core chiplet so 2x32MB. X3D variant has the whole L3 cache shared between all cores of all chiplets. The X3D cache is consumming significant amount of energy, not ideal for all core CPU frequency if L3 cache is not used at all.

Instead of 5800x you should grab 7700x - better manufacturing process, much higher performance per watt. If you have larger budget, then 7950x is the best what you can have.

If your projects are using at least 16Mpix photos and you will be processing depth maps at highest quality(full photo resolution), then it makes a lot of sense buy 5800x/7700x and spend more money on faster GPU.

[Nuiosk]

I have concerns that metashape code is not using L3 cache at all in terms of vectorizing data using modern processor instructions. The processed data are large and in uncompressed formats stored in RAM and cannot fits even in large L3 cache. Maybe if the data are processed at smaller pieces, this is question for devs.

7900x has 32MB cache for each 8-core chiplet so 2x32MB. X3D variant has the whole L3 cache shared between all cores of all chiplets. The X3D cache is consumming significant amount of energy, not ideal for all core CPU frequency if L3 cache is not used at all.

Instead of 5800x you should grab 7700x - better manufacturing process, much higher performance per watt. If you have larger budget, then 7950x is the best what you can have.

If your projects are using at least 16Mpix photos and you will be processing depth maps at highest quality(full photo resolution), then it makes a lot of sense buy 5800x/7700x and spend more money on faster GPU.

According to your comments, I decided to go with 7700x and my GPU 6700XT. 

Thank you very much for your very kind assistance.