how to analyze the failures of your graph

The drivers and the hardware attached to them by AMD, for some time now, have improved a lot. The leap in quality, performance and few problems is evident, especially from the RDNA architecture as such. But, as always happens in this sector, nothing is perfect, so with the intention of being able to detect and correct such errors, especially in the software part, AMD has introduced Radeon GPU Detective, or RGDin its version 1.0, what exactly does it do?

Although the Adrenalin drivers already carry the call “Tool to report bugs to AMD”, as we say, included in the controllers, where we can tell those of Lisa Su with hair and signs everything that has happened to us, this falls within the scope of the user as such. For professionals, or very advanced users, comes this new utility with a lot of potential.

AMD Radeon GPU Detective (RGD) 1.0, now you can know what happened to your graphics card before it crashed or crashed

The company bills its software as “a tool for GPU postmortem failure analysis,” but what exactly does it do? Well, AMD explains it as follows:

The tool allows developers to retrieve and analyze GPU crash dumps and produce information that can help reduce the search for the root cause of said failure. This information includes page fault details, resource details, and execution markers that reflect what GPU work was in progress. in the moments before the blockade.

This first release of the tool supports Windows GPU Crash Analysis (“TDR”) in Direct3D 12 applications and requires an RX 6000 series (RDNA 2) or RX 7000 series (RDNA 3) card, plus the latest software. AMD Radeon Adrenalin released.

AMD Radeon GPU Detective (RGD) 1.0 allows developers to set the driver to crash mode using the Radeon Developer Panel (RDP) before the crash is reproduced. After the lockdown, an analysis file is generated with details that can help shed light on the cause of it:

• Execution flag information indicating which rendering passes and drawing calls were in progress during the block. The application developer can add custom markers using the AGS (AMD GPU Services) library.
• If the crash was determined to be due to a page fault:
  • Detect the offending virtual address.
  • Show details about any resource (such as heaps, textures, and buffers) that resided at the offending virtual address, including developer-supplied resource names.
  • Memory events timeline (such as Create, Destroy, and Evict) filtered for only the relevant resources.

The fault analysis file is generated in text format by default. The tool can also be configured to generate the analysis file in JSON format to support automated processing.

Minimum requirements for RGD

Considering that AMD Radeon GPU Detective (RGD) 1.0 is part of the calls Radeon Developer Tool Suite or RDTS, it is normal for the company to set a minimum requirement to be able to use it.

This is the Adrenalin 23.7.2 driver version, if possible a higher one, since it is to be expected that the reds are already developing a new version of RGD. Likewise, this is open source and can be found in the Github repository, where we can even put our comments, impressions or incorrect functions that we find. you can download Radeon GPU Detective (RGD) from the gouopen.com website.