Metashape 1.7.3 not detecting all available hardware

[RHenriques]

Metashape 1.7.3 is not detecting all available eGPUs on Macos. In exactly the same conditions the GPU tab in both versions appear as attached images.
Best Regards

[Alexey Pasumansky]

Hello Renato,

Which output is shown in the Console pane, if you use the following line input:

Code: [Select]

Metashape.app.enumGPUDevices()

[RHenriques]

Hi Alexey

Here is the echo from 1.7.3:

Metashape.app.enumGPUDevices()
2021-05-30 12:31:41 Found 3 GPUs in 0.001334 sec (CUDA: 0.000403 sec, OpenCL: 0.000913 sec)
Out[2]: 2021-05-30 12:31:41
2021-05-30 12:31:41 [{'name': 'AMD Radeon HD - FirePro D500 Compute Engine',
2021-05-30 12:31:41   'vendor': 'AMD',
2021-05-30 12:31:41   'version': 'OpenCL 1.2 ',
2021-05-30 12:31:41   'compute_units': 24,
2021-05-30 12:31:41   'mem_size': 3221225472,
2021-05-30 12:31:41   'clock': 725,
2021-05-30 12:31:41   'pci_bus_id': 0,
2021-05-30 12:31:41   'pci_device_id': 0},
2021-05-30 12:31:41  {'name': 'AMD Radeon HD - FirePro D500 Compute Engine',
2021-05-30 12:31:41   'vendor': 'AMD',
2021-05-30 12:31:41   'version': 'OpenCL 1.2 ',
2021-05-30 12:31:41   'compute_units': 24,
2021-05-30 12:31:41   'mem_size': 3221225472,
2021-05-30 12:31:41   'clock': 150,
2021-05-30 12:31:41   'pci_bus_id': 0,
2021-05-30 12:31:41   'pci_device_id': 0},
2021-05-30 12:31:41  {'name': 'GeForce GTX 1080 Ti',
2021-05-30 12:31:41   'vendor': 'NVIDIA',
2021-05-30 12:31:41   'version': 'OpenCL 1.2 ',
2021-05-30 12:31:41   'compute_units': 28,
2021-05-30 12:31:41   'mem_size': 11811160064,
2021-05-30 12:31:41   'clock': 1139,
2021-05-30 12:31:41   'pci_bus_id': 0,
2021-05-30 12:31:41   'pci_device_id': 0}]

The same echo from 1.7.1

Metashape.app.enumGPUDevices()
2021-05-30 12:30:57 Found 4 GPUs in 0.000635 sec (CUDA: 0.000541 sec, OpenCL: 7.9e-05 sec)
Out[3]: 2021-05-30 12:30:57
2021-05-30 12:30:57 [{'name': 'AMD Radeon HD - FirePro D500 Compute Engine',
2021-05-30 12:30:57   'vendor': 'AMD',
2021-05-30 12:30:57   'version': 'OpenCL 1.2 ',
2021-05-30 12:30:57   'compute_units': 24,
2021-05-30 12:30:57   'mem_size': 3221225472,
2021-05-30 12:30:57   'clock': 725,
2021-05-30 12:30:57   'pci_bus_id': 0,
2021-05-30 12:30:57   'pci_device_id': 0},
2021-05-30 12:30:57  {'name': 'AMD Radeon HD - FirePro D500 Compute Engine',
2021-05-30 12:30:57   'vendor': 'AMD',
2021-05-30 12:30:57   'version': 'OpenCL 1.2 ',
2021-05-30 12:30:57   'compute_units': 24,
2021-05-30 12:30:57   'mem_size': 3221225472,
2021-05-30 12:30:57   'clock': 150,
2021-05-30 12:30:57   'pci_bus_id': 0,
2021-05-30 12:30:57   'pci_device_id': 0},
2021-05-30 12:30:57  {'name': 'GeForce GTX 1080 Ti',
2021-05-30 12:30:57   'vendor': '',
2021-05-30 12:30:57   'version': '',
2021-05-30 12:30:57   'compute_units': 28,
2021-05-30 12:30:57   'mem_size': 11810963456,
2021-05-30 12:30:57   'clock': 1582,
2021-05-30 12:30:57   'pci_bus_id': 79,
2021-05-30 12:30:57   'pci_device_id': 0},
2021-05-30 12:30:57  {'name': 'GeForce GTX 1080 Ti',
2021-05-30 12:30:57   'vendor': '',
2021-05-30 12:30:57   'version': '',
2021-05-30 12:30:57   'compute_units': 28,
2021-05-30 12:30:57   'mem_size': 11810963456,
2021-05-30 12:30:57   'clock': 1582,
2021-05-30 12:30:57   'pci_bus_id': 237,
2021-05-30 12:30:57   'pci_device_id': 0}]




Both versions are running simultaneously to get the above data, so the mchine is in the same conditions.
Cheers

[Alexey Pasumansky]

Hello Renato,

Do you have 1.7.2 by the hand to perform the similar test?

[RHenriques]

Yes

Here is the output from 1.7.2

Metashape.app.enumGPUDevices()
2021-05-30 13:00:27 Found 4 GPUs in 0.062248 sec (CUDA: 0.059628 sec, OpenCL: 0.002573 sec)
Out[1]: 2021-05-30 13:00:27
2021-05-30 13:00:27 [{'name': 'AMD Radeon HD - FirePro D500 Compute Engine',
2021-05-30 13:00:27   'vendor': 'AMD',
2021-05-30 13:00:27   'version': 'OpenCL 1.2 ',
2021-05-30 13:00:27   'compute_units': 24,
2021-05-30 13:00:27   'mem_size': 3221225472,
2021-05-30 13:00:27   'clock': 725,
2021-05-30 13:00:27   'pci_bus_id': 0,
2021-05-30 13:00:27   'pci_device_id': 0},
2021-05-30 13:00:27  {'name': 'AMD Radeon HD - FirePro D500 Compute Engine',
2021-05-30 13:00:27   'vendor': 'AMD',
2021-05-30 13:00:27   'version': 'OpenCL 1.2 ',
2021-05-30 13:00:27   'compute_units': 24,
2021-05-30 13:00:27   'mem_size': 3221225472,
2021-05-30 13:00:27   'clock': 150,
2021-05-30 13:00:27   'pci_bus_id': 0,
2021-05-30 13:00:27   'pci_device_id': 0},
2021-05-30 13:00:27  {'name': 'GeForce GTX 1080 Ti',
2021-05-30 13:00:27   'vendor': '',
2021-05-30 13:00:27   'version': '',
2021-05-30 13:00:27   'compute_units': 28,
2021-05-30 13:00:27   'mem_size': 11810963456,
2021-05-30 13:00:27   'clock': 1582,
2021-05-30 13:00:27   'pci_bus_id': 79,
2021-05-30 13:00:27   'pci_device_id': 0},
2021-05-30 13:00:27  {'name': 'GeForce GTX 1080 Ti',
2021-05-30 13:00:27   'vendor': '',
2021-05-30 13:00:27   'version': '',
2021-05-30 13:00:27   'compute_units': 28,
2021-05-30 13:00:27   'mem_size': 11810963456,
2021-05-30 13:00:27   'clock': 1582,
2021-05-30 13:00:27   'pci_bus_id': 237,
2021-05-30 13:00:27   'pci_device_id': 0}]


Cheers

PS: In previous 1.7.3 versions the detection was allright

[Alexey Pasumansky]

 Hello Renato,

Thank you for additional information. We are currently working on the solution and will provide you with a fixed version as soon as possible.

To figure out why only one GeForce card is displayed in OpenCL mode it would be very helpful to see output of the clinfo program.
If you have Homebrew installed, this can be done using the following commands:

Code: [Select]

brew install clinfo
clinfo
If you can’t use Homebrew – please let us know, we will send you a pre-build clinfo binary.

[RHenriques]

Hi Alexey

Your message caught my attention because I guess that CUDA is not being activated in my hardware using the current 1.7.3 version and that's the problem.
The appearence of only 1 Nvidia in OpenCL is an old issue and I was already aware of it. Since while using CUDA that was not a problem, I was ignoring it. By reading your message I went back to the images and SAW that CUDA is off in the latest version.  Probably it can be forced with the CUDA activation tweak. What is the right syntax?

Meanwhile, I'll install clinfo.

Best Regards

PS - I have the tweak to force CUDA to "true" and it's not being respected. I guess that the problem is the activation of CUDA. Latest "faulty" 1.7.3 versions were mounting all the hardware just fine.

[Alexey Pasumansky]

Hello Renato,

We found the issue that could be a reason of lost CUDA support in 1.7.3 release version and we'll try to fix it as soon as possible.

But the information related to clinfo output may be helpful, as we expect that in 1.8 version CUDA support could be dropped in Metashape for macOS.

[RHenriques]

Hello Renato,

We found the issue that could be a reason of lost CUDA support in 1.7.3 release version and we'll try to fix it as soon as possible.

But the information related to clinfo output may be helpful, as we expect that in 1.8 version CUDA support could be dropped in Metashape for macOS.

That would be awful to Macusers with CUDA capable machines :-(
At least it would be great to replace it by Metal for instance.

Sent clinfo output by email.

[RHenriques]

Hi Alexey

It seems that the Hardware detection problems from the previous beta are gone in build 1.7.4 (12584). So far so good.
Best regards


PS: This version also seems to spare a little bit the memory and the processing pressure on the hardware. The calculations are more stable. A few CUDA errors from previous versions are not happening now.