There is an issue with 0.7.5 version:
random polaris video card in the system stops mining (showing 0 hashes) and starts to warm up until the thermal protection is triggered, driver 19.12.2. On version 0.7.1 everything is stable.
test system: RX580+RX580+Vega56
algo: daggerhashimoto
Hi! Would this be immediately from the start or after mining a while? It’s hard to see what the miner could do to increase pressure on the card as you describe without the driver comm with card being busted. If it happens immediately, can you run with —debug and pm me the output? If not immediately, any other clues about what is happening when you run into these situations? Also, does the miner detect the gpu as DEAD or not?
Edit: last, are these 4GB or 8GB 580s, or a mix even?
Hi, it happens randomly from few minutes to few hours.
test system: rx 580 8g, rx 580 4g, vega 56.
As I understand it, because the dag file is now divided, the overclocking potential of the memory has decreased (a person from the local forum also complained about an increase in the number of hw errors). This can explain the fact that mining stops on one of the video cards. But in 0.7.5 version the miner does not report that vc is dead, it reports 0 mh/s, driver does not reset card settings,
fan stops spinning (the miner shows that the fan is spinning), and vc goes overheating. Zero RPM function is disabled.
Correct, all kernels on all cards are now prepared for a > 4GB dag. This means split allocation into two buffers. I don’t believe this should affect oc capabilities in itself, but compute has increased somewhat because the address calculations are now more complex with the split dag. No choice there though, Windows and amdgpu-pro don’t support the bulky allocations
.
Your issue is quite confusing though. We don’t make progress hashing but we’re not stuck in a kernel enqueue, or else the card would be detected dead. I wonder if it crashes calling the Windows ADL library, reading clocks and temps, we’ve done a lot of changes there lately (released in 0.7.2). Just makes no sense that your fan stops spinning etc, we don’t do any clock or fan control, only read-only calls.
Can you run with —no_gpu_monitor to check? And thank you for testing and reporting in general, always grateful!
— K
