I have posted this a few years ago in the reply section of the Anandtech news item when it was reported AMD videocards based on GCN architecture and newer were crashing randomly and investigation was still going on.
And there seemed no clear explanation at the moment. Totally random.
I found this solution a few years back around 2019 i think for Intel graphics based videocards while googling and as it turns out, it can also be a solution for your AMD or Nvidia card.
I had the random crashes issue for almost a year and was desperate. And was so happy when this turned out to be the solution.
But if you ever seem to have random crashes with your videocard and it seems totally random, try disabling PCI express power saving. Turn off "Link State Power Management".
Correct my please when i am wrong because i am doing this with vague memories and incomplete information about the PCIexpress specification :
If i remember correctly "Link State Power Management" causes the serial links te power down but when they need to start up again, i kind of forgot about how it works but i think the link needs to be trained again and this causes considerable latency in comparison to a PICe link that remains active without power management.
This all depends on the PCIe revision, and the link power state. L0 , L1. And L2& L3. This is the case for PCIe 3.0. But probably is the same or similar for the older and newer PCIe versions.
Anyway, if you have random crashes and your videocard whatever brand may be the culprit, try this solution. I suspect the added latency when "Link State Power Management" is active causes some timeout issue or deadlock or whatever difficult problem can occur with highly multithreaded software running both on cpu and gpu.
For desktop machines "Link State Power Management" is not really necessary anyway.
So turn "Link State Power Management" off.
And also disable "USB selective suspend setting", some usb devices do not play nice and can cause hangups in games or other programs.
For desktop machines "USB selective suspend setting" is not really necessary anyway.
And there seemed no clear explanation at the moment. Totally random.
I found this solution a few years back around 2019 i think for Intel graphics based videocards while googling and as it turns out, it can also be a solution for your AMD or Nvidia card.
I had the random crashes issue for almost a year and was desperate. And was so happy when this turned out to be the solution.
But if you ever seem to have random crashes with your videocard and it seems totally random, try disabling PCI express power saving. Turn off "Link State Power Management".
Correct my please when i am wrong because i am doing this with vague memories and incomplete information about the PCIexpress specification :
If i remember correctly "Link State Power Management" causes the serial links te power down but when they need to start up again, i kind of forgot about how it works but i think the link needs to be trained again and this causes considerable latency in comparison to a PICe link that remains active without power management.
This all depends on the PCIe revision, and the link power state. L0 , L1. And L2& L3. This is the case for PCIe 3.0. But probably is the same or similar for the older and newer PCIe versions.
Anyway, if you have random crashes and your videocard whatever brand may be the culprit, try this solution. I suspect the added latency when "Link State Power Management" is active causes some timeout issue or deadlock or whatever difficult problem can occur with highly multithreaded software running both on cpu and gpu.
For desktop machines "Link State Power Management" is not really necessary anyway.
So turn "Link State Power Management" off.
And also disable "USB selective suspend setting", some usb devices do not play nice and can cause hangups in games or other programs.
For desktop machines "USB selective suspend setting" is not really necessary anyway.
Last edited: