Agisoft Metashape

Agisoft Metashape => General => Topic started by: rexsham on July 17, 2019, 05:07:13 PM

Title: AMD 5700XT support ( no SPIR support)
Post by: rexsham on July 17, 2019, 05:07:13 PM
When I run my project, I see this:

AMD Radeon RX 5700 XT (gfx1010): no SPIR support

Will it affect the performance or results of the project?
Title: Re: AMD 5700XT support ( no SPIR support)
Post by: Alexey Pasumansky on July 17, 2019, 09:33:26 PM
Hello rexsham,

If you are using latest AMD drivers and latest Metashape version, then you should not experience processing problems.

Check that you see [GPU] related processing lines in the log during GPU-supported operations and no errors.
Title: Re: AMD 5700XT support ( no SPIR support)
Post by: rexsham on July 18, 2019, 08:32:05 AM
I am using latest version of both Driver and Metashape.
I still see the same  ( no SPIR support) issue in the log.
The dense cloud generation process can proceed for an hour and it was stuck at around 20% and cannot progress anymore.
Title: Re: AMD 5700XT support ( no SPIR support)
Post by: rexsham on July 18, 2019, 10:21:12 AM
Attached is the log file I got using 5700XT and Latest Metashape and Radeon Driver
Title: Re: AMD 5700XT support ( no SPIR support)
Post by: tweezlednutball on November 07, 2019, 06:08:45 AM
Also experiencing the same issue.  Latest drivers. Version 1.6.0 build 9487
Title: Re: AMD 5700XT support ( no SPIR support)
Post by: tweezlednutball on November 07, 2019, 06:11:27 AM
Also noticing that Metashape is only reporting 20 compute units when there are 40 on the 5700xt.

Thank you.
Title: Re: AMD 5700XT support ( no SPIR support)
Post by: Alexey Pasumansky on November 07, 2019, 01:35:13 PM
The line "no SPIR support" indicates that the driver doesn't support SPIR. It is not an error, just an indication of the fact, since AMD dropped SPIR support at some point.

As for the number of compute units, we are using OpenCL call CL_DEVICE_MAX_COMPUTE_UNITS to get this number. It may be a specific of RDNA architecture that only half of the actual compute units number is returned. However, it shouldn't affect the device performance.