Discussion in 'Gaming (Software and Graphics Cards)' started by King of Interns, Aug 17, 2016.
Wooow, that K40 is a real monster sized card
There are two aspects where you shouldn't push anything to the limit
1. Sudden spikes in power, which is pretty damn scary for GPUs nowadays (power density exceeds a nuclear reactor/ getting close to rocket exhausts at peak)
2. Continuous loads. Which you will notice when you do something like mining/ deep learning. I have no doubt if you mining on that 1080 it will eventually kill your 7720 or 1080 or both
turning it into a desktop card with somehting like this? (https://market.m.taobao.com/app/idl...9250.489830115&forceFlush=1&qq-pf-to=pcqq.c2c)
1. Not sure what this means in a practical sense, launching a game is a spike in power, 3d acceleration is a power spike.
2. Really should be using a continuous load to determine stability anyways. Back in the day 24 hours was considered the minimum for any meaningful test of a particular tune or overclock. All continuous loads are not equal either.
Feels like an omission of variables for a given scenario
1. I mean the sudden spike when you are running games. On recent GPUs that sudden power spike can easily twice as much as the normal TDP. (An interesting read if you can find translation for Chinese: https://www.fcpowerup.com/gpu-transient-spike-shutdown-issue/)
2. That 24 hour is also questionable. I have a RX 588 that was runnign perfectly stable for half a year for 24*7 then suddenly died (one or more of the vRAM chip became unstable). My R9 nano also shrinked after being tested stable undervolt (was stressed for at least 24 hours before vBIOS mod). OFC if you are challenging for some benchmarks that's a different story.
Separate names with a comma.