I am using an AMD 6900XT and am experiencing either complete crashes, or weird artefacts on projects that take longer than an hour to complete.
The program crashes with the message to reduce VRAM usage or limit or machine load (VRAM is reported to never exceed 3.5GB from 16GB available, system memory never exceeded 7GB of 16GB available, no other programs running, max GPU load 60%, max CPU load 40%).
Even though GPU usage and temperature is very low compared to other applications, I am experiencing extensive stability issues.
If the program does render a project, a couple of hours into the render some weird artefacts appear (black square covering approximately 1/3 of the screen, see image below). This can progressively get worse, covering 50% of the screen later and eventually blacking out the whole image.
I have tested this with Artemis Medium, High as well as with Proteus.
CPU is 3900x, motherboard b550. Windows 10 latest, AMD drivers 21.11.3 stable.
Have tried reducing GPU clock and memory clock, disabled ultra low voltage. No changes to stability or the artefacts appearing.
Can you try the reduce machine load option in preferences. Is it always consistent or is it appearing randomly?
Is the GPU well cooled, power supply sufficient? You can also try reducing max power usage of the GPU and under clock it to see if it fixes the issue?
It is appearing consistently a couple of hours into the project (tested with different files, all individually and with different models/settings).
The location where the rectangle appears is different each time though. The size differs as well, with one “tile” appearing first, then another until the whole screen is blacked out eventually with the project still rendering happily behind that black curtain. These tiles are usually between 1/3rd and 1/9th of the screen. The squares are in the output file as well as in the current frame render preview in Topaz.
I will try the reduce machine load setting overnight. The problem with that option is that it pretty much doubles the processing time, and that machine load is already very conservative when running the project under normal settings.
The GPUz log file I posted in the other thread (attached again below) and the performance numbers in this topic here show that the workload, temperature and power draw are all minimal and well below any other comparable application.
I have tried different combinations of power limit, clock speed limit, VRAM speed limit, VRAM timing reduction, CPU clock decrease. I can give it another go tonight with greatly reduced settings for the sake of testing, however am not confident based on the previous results that it will lead to an improvement. Will let you know tomorrow.