Christoffer Glans wrote:Any time when you can use the current hardware setup, clock and power draw in other software and games at the GPUs maximum without a problem and a new software comes around and shuts the card down, that is definitely on the software side. Why else would overclocking be stress tested? What do you think the stress tests do to the GPU card? Having a software shut down the card is not a sign that the GPU is overclocked and can't handle stress, it's a sign of bad code that shouldn't be in the software.
You're confusing trigger with root cause. Any userspace application that can hork a system may be doing something extremely stressful, but it's the
job of the OS and hardware to be robust against userland abuses. If I run an app that BSODs or panics the kernel, the bug is with the kernel (or drivers) or harware. The app may have triggered it, but the blame for the instability lies with one of the layers underneath. If the app "fixes" the issue, at best it merely works around triggering the underlying problem.
And that assumes Resolve is doing something "abusive." It's entirely possible it's using a less used and undertested API or instruction set. For example, AVX-512 instructions generated no end of instability early on, often highlighting insufficient cooling solutions, even though every other workload was stable on the same system. We don't blame the applications using AVX-512 for crashing systems.