4 ms·
> Can you elaborate on this a bit more? I honestly don't understand what that even means. Small manufacturing issues that usually don't get noticed, can fail a
by capableweb 3y ago
> Can you elaborate on this a bit more? I honestly don't understand what that even means.
Small manufacturing issues that usually don't get noticed, can fail a card if you put it through 100% load for a long period of time. Cryptocurrency miners regularly would downclock cards for cost reasons (energy/$) and because consumer GPUs also aren't made for sustained 100% load (across days).
- Meculpu 3y agoGPUs are plenty of capable running 24/7 without issues for a long time. Plenty of cloud providers have consumer grade GPUs. Cryptominers don't have an issue running it at 100% either. They have heat issues and power efficiency issues not GPU Chips breaking. The only instance I know are if you use a very particlar load type which would normally never happen like 24/7 hitting one register only or moving data only through the bus without anything else. Those issues are rare.
- capableweb 3y ago> GPUs are plenty of capable running 24/7 without issues for a long time. Running, yes. But running 24/7 at 100% load for days without any sort of break or underclocking it is a surefire way of making it fail quickly. If you know any people doing cryptocurrency mining, ask them :)
- CydeWeys 3y agoThe guy you're replying to knows when those issues pop up and has relevant experience here. I'd listen to him. I too have experienced hardware failure at 100% load for months straight, and have resorted to underclocking for power efficiency and longevity reasons. The performance/cost/longevity curves are not linear. Just take a little bit off the top and you do much better.
- 300bps 3y agoEthereum miners often underclock the Core and overclock the Memory in their GPU because it’s such a memory intensive operation.