Agisoft Metashape 2.0.0 pre-release

[Alexey Pasumansky]

Hello Renato,

Do you observer the same issues, when only internal (5600M) GPU is used for processing and Metashape? Also whether the same slow down is observer, when only one eGPU is used (internal GPU - disabled)?

[RHenriques]

Hi Alexey

It is slower in any situation. With the internal AMD it's less noticeable because, with any eGPU, only the internal GPU seems to be used often but, however, it's a lot slower than version 1.8.4. The external eGPU's have very small periods of usage. With only one, two or three eGPU's, with or without the internal AMD active, things are slower after 8% of the ongoing task. The problem seems to be in the kernels building and loading step. Try to see what's the diference in code between the 1.8.4 and 2 version, in this step, because 1.8.4 it's working just fine.
Cheers

[Mak11]

Hi Alexey

It is slower in any situation. With the internal AMD it's less noticeable because, with any eGPU, only the internal GPU seems to be used often but, however, it's a lot slower than version 1.8.4. The external eGPU's have very small periods of usage. With only one, two or three eGPU's, with or without the internal AMD active, things are slower after 8% of the ongoing task. The problem seems to be in the kernels building and loading step. Try to see what's the diference in code between the 1.8.4 and 2 version, in this step, because 1.8.4 it's working just fine.
Cheers

This is specifically on MacOS right?

[GRGRSS]

Hi, 
What are  all the GPUs supported by this new version?
Thanks

[RHenriques]

Hello

Version 2.0.0 build 15461 continues to have the problems previously noted in MacOS.
Using only a MacBookPro 16" intel and the included AMD Radeon Pro 5600M, using the same dataset, the build point cloud times are:

Metashape version 1.8.4 - 98.99 sec
Metashape version 2 - 607.864 sec

Best regards

[RHenriques]

Hello

Version 2.0.0 build 15501 continues to have the problems previously noted in MacOS.
Processing times seem to be slower than last beta. Is there any tweak or improvment that can be done from my side?
In its present state, this version is almost unusable to obtain Point Clouds using macOS. Please try to compare the code with 1.8.4 because this version is working great in this step.

Using only a MacBookPro 16" intel and the included AMD Radeon Pro 5600M, using the same dataset, the build point cloud times are:

Metashape version 1.8.4 - 98.99 sec
Metashape version 2 - 644.256 sec

Best regards

[Alexey Pasumansky]

Hello Renato,

Do you have logs for processing with only Radeon Pro 5600M enabled in both versions of Metashape? I can also suggest to start the calculation two times in a row for both versions, in order to have the kernels already compiled for the second run - it will make the timing in the log more representative.

Meanwhile we have performed some tests on MacBook Pro with AMD card and observe only about 10% time difference for about 100 photos dataset.

[RHenriques]

Hi Alexey

If the diference between 1.8.4 and 2 versions was about 10% processing time it would be great, considering the great new aditions of version 2.
However my processing times are a lot worst than that.
As you suggest, I'm sending the processing logs from the two versions using 2 processing requests in a row. In both cases, as you will see, processing times were worst in the second time. Considering this, probably the "compiling kernels" step is not the culprit. Things seems to slow a lot after this step.
See logs attached.
Best regards

[spader]

I am noticing a significant slowdown between 1.8.4 and 2.0, latest version, as well. (Imac Prop, 3.2 GHz 8-Core Intel Xeon W, Radeon Pro Vega 56 8 GB, Montery). Alexey or RHenriques, do you mind sharing that small dataset so we have someting similar to test on? Alexey, is it the logs that would be of best help, or can we do something else to troubleshoot?

[mattcalavera]

Hi,

we have floating licenses and seems they did NOT work with pre-release version.

Is there a way to make them work?

TNX

MF

[Alexey Pasumansky]

we have floating licenses and seems they did NOT work with pre-release version.

Hello Matt,

Metashape Pro 2.0 is using a new activation system and would not be compatible with the license activation made in 1.x versions (this includes the floating licenses activated via RLM). New floating licensing server utility will be available in January. Meanwhile, if you are interested in version 2.0 evaluation, you can use trial licenses. For the trial license extension you can contact our support team (support@agisoft.com).

[Alexey Pasumansky]

As you suggest, I'm sending the processing logs from the two versions using 2 processing requests in a row. In both cases, as you will see, processing times were worst in the second time. Considering this, probably the "compiling kernels" step is not the culprit. Things seems to slow a lot after this step.
See logs attached.

Hello Renato,

Thank you for additional information.

We will try to prepare the test version for you and spader this week, that you should be able to test on MacOS configuration, hopefully the reported slow down issue would be resolved in that version and, if so, it will be also fixed in the next public pre-release update.

[RHenriques]

we have floating licenses and seems they did NOT work with pre-release version.

Hello Renato,

Thank you for additional information.

We will try to prepare the test version for you and spader this week, that you should be able to test on MacOS configuration, hopefully the reported slow down issue would be resolved in that version and, if so, it will be also fixed in the next public pre-release update.

Hi Alexey

Just tested the version you sent (15512) and it's great!! The processing is more or less on pair with the 1.8.4 version. Probably 8 to 10% slower but much faster than the last beta. The problem seems to be solved. I'll test with all the external GPU's also but, probably, the problem is also gone with that setup. If something will go wrong I'll let you know.

By curiosity, and if it is possible to know, what was the problem?
Best regards

PS: All seems to be fine with the external GPU's. Now the processing times are on pair with version 1.8.4. All solved!!

[Ronyfly]

Bonjour à toute l'équipe Agisoft,

Nouveau sur le forum, mais ancien utilisateur de Photoscan/Metashape, je vous remercie encore infiniment pour l'énorme travail effectué depuis les premières versions.
Étant dans le domaine de l'archéologie, nous mettons à rude épreuve le logiciel.
Je voulais savoir si la représentation graphique de la région pouvait évolué sur la prochaine version 2.0.
En effet, avec les nuages denses, il est souvent difficile de voir, dicerner le bloc région dans ces amas de points et en définir ses limites.
Autres suggestions : Avez-vous réfléchi à comment faire évoluer les profils sur l'éditeur de mesure. Il est pour le moment en 2.5D.

Bonne continuation ! hourra !

Ron

[Alexey Pasumansky]

Hello Renato,

Good to hear, that this reported issue has been finally resolved.

The issue has been related to the additional RAM instability checks introduced in the most potentially problematic parts of processing algorithms (based on the received user reports), which should allow to terminate the processing with the error/warning message only, avoiding hard crashes.
Thanks to your logs we were able to find out the flaw in one of these checks, which was caused by threads synchronization and could be observed on systems with big number of CPU cores. The fix will be included in the next public pre-release update.