7 Replies Latest reply on Nov 25, 2016 5:03 PM by goodplay

    Confirmed Fix for DirectX Render Crash/Red Screen Freeze

    dementalist

      I have an update regarding the issues of DirectX render crashes, "display driver has stopped responding and has recovered" crashes, and red screen error (hardware) crashes. I also have a resolution I have personally confirmed after several days of testing.

       

      Further personal experimentation has revealed to me that there is a problem with AMD's default parameters — as well as a potentially serious liability for both Sapphire and AMD regarding the longevity of these cards.

       

      The default parameters defined by AMD Wattman are incorrect. The voltage is too high for the GPU clockspeed levels specified. Only by reducing the voltages for each of the seven states in AMD Wattman have I been able to completely stop the driver from crashing. I do not have to reduce GPU clockspeed by 300MHz across the board to stop this issue anymore.

       

      I was initially led to that conclusion by the experiments run by the user who started the following thread: https://forums.anandtech.com/threads/optimizing-radeon-performance-with-wattman.2482895/

       

      By simply undervolting the card slightly, I was able to completely stop the driver from crashing. I believe the default parameters for the voltages specified in AMD Wattman are so high that the GPU clockspeeds become unstable.

       

      The other problem is GPU overheating.

       

      AMD Wattman's default parameters for the fan speed and for the temperatures are incorrect, resulting in crashing with a static red screen and a prolonged error sound (hardware crash), resolvable by a hard system reboot; not to be confused with a red screen of death.

       

      Because the default parameters are incorrect, AMD Wattman does not regulate fan speed the way it should, resulting in random crashes even if the card appears to be stable at 75C to 80C on full load. The reason for the red screen crashes, in my finding, is that the temperature can randomly spike as high as 95C or more, triggering a hardware crash.

       

      In my conclusion, AMD's incorrect parameters for voltage trigger a display driver crash, and its incorrect parameters for fan speed and target temperature trigger a hardware crash. Only by switching these from auto to manual is the problem resolved.

       

      For voltages, from states 0 through 7, I specified:

       

      N/A | 410 | 735 | 985 | 1050 | 1115 | 1140 | 1260

       

      For fan speed, I set a minimum value of 1500 and a target value of 3200 so that the fan spins as it should.

       

      For GPU temperature, I set a max value of 75 and a target value of 50. This way, the fan begins spinning once the card reaches a value of 75C, and it spins so that the card gets down to at least 50C. Because of the awesome construction of Sapphire's Nitro+ cooler, my GPU gets down to about 33C.

       

      That is my confirmed fix for the Sapphire Nitro+ Radeon RX 470 8GB.

       

      I have written this post with the interest of forewarning Sapphire and AMD of potential legal problems that could arise for both Sapphire and AMD if the issues with AMD Wattman are not brought to proper attention. GPUs can and will overheat if users rely on the default parameters — which, to put it bluntly, means about 99% of the consumer base.