TDR stands for Timeout, Detection, and Recovery components in Windows.
It’s supposed to be there to help stop BSODs by resetting the GPU and/or driver when there’s an issue of long delay.
The GPU is taking more time than permitted to display graphics to your monitor. This behavior can occur for one or more of the following reasons:
The problem here is with the GPU drivers. They haven't been fixed since the day these APUs came out. They cause BSODs, slow OpenGL performance, visual distortion, and full on crashes requiring shutdown via reset button or holding the power button.
Here's all the random possible solutions I've come across: (FYI IT'S STILL NOT FIXED)
- Lower OC RAM speeds to 2133 MHz --> Still Crashing
- Increase voltage a little to RAM + CPU (1.2 V) --> Still Crashing
- Turn off Cool N Quiet --> Still Crashing
- Turn off Core Perf Boost and Global C-State --> Still Crashing
- Turn off auto-voltage for CPU since it can cause spikes in voltage --> Still Crashing
- Increased VRAM to 2 GB --> Still Crashing
- Replace your Memory Modules with something on your Mobo QVL List --> Will problem do this next. (Currently using Corsair LPX ram... will probably try Amazon.com: Crucial 16GB Kit (8GBx2) DDR4 2133 MT/s (PC4-17000) DR x8 Unbuffered DIMM 288-Pin Memory...
- Install AMD Chipset Drivers under the 'Optional Downloads' tab. --> Ryzen Desktop APU's
- Install AMD GPU Drivers under the Main Tab --> Ryzen Desktop APU's
--> Still Crashing
List of other drivers that may be good or bad...
- Wi-Fi Driver
--> Still Crashing
- Fixed when I selected 8bit (adjusted from 10bit) color in Radeon Settings.Even better when I selected 4:2:2 in HDMI output.
- Boot into a clean boot state (AMD Support through email told me to do this) --> Still Crashing
This website offers some good troubleshooting tips on your error: Fix: VIDEO_TDR_FAILURE (ATIKMPAG.SYS) IN WINDOWS 8.1/10 - Appuals.com
EDIT: 08/24/2018 - Deleted the last link due to Unsafe ratings from Web Of Trust about downloading their software.
Thank you very much for posting this. This is the first post that addresses all of the strange things that can be done to attempt to work around this issue of the AMD drivers. I am convinced the problem is driver related because:
As you can see this isn't a hardware problem. The Raven Ridge APU AMD drivers are known to have issues with Direct X 9 and Starcraft II displays this weakness. However it's not just DX9. Quake Champions can also be used to reproduce the same issue, although only when I'm pushing the RAM clock. That is to say, the issue hiccup in which causes the TDR failure because Windows is unable to push through the pause in iGPU response when using the drivers to address the hardware is sensitive to RAM speeds.
This build: https://pcpartpicker.com/b/j6mqqs
Following up here. I purchased QVL RAM which is more stable at 3200Mhz than either of the other more expensive G.Skill Hynix or Corsair Samsung B-die RAM kits I've tried. Unfortunately, I can still reliably reproduce VIDEO_TDR_FAILURE running Starcraft 2 on Windows 10.
pls give a little more juice on vsoc - 1.1V should be "minimum" for high ram speeds with low latency - yet pls dont go over 1.15V // ps some boards use higher voltages than bios or hwinfo shows
Thanks for encouraging me to try again at SOC. Unfortunately it didn't create stability in this particular use case with 1.1 or 1.15 SOC, even with 1.4 on DRAM. Right now I'm running the Patriot Viper memory listed on ASRock's QVL and XMP 3200 is thankfully stable for Quake Champions and a myriad of other use cases with benchmark scores and fps appreciably improved over the more expensive but less compatible RAM i started with. Anyhow, the 2400G is a marvel of a CPU, I'm loving that I can get 50fps in QC without a dedicated GPU, just amazing.
Edit: I want to mention that I reconfirmed that with 2133 as a result of the Auto configuration in the BIOS for RAM, it's stable in StarCraft 2. Also my core voltage is actually underclocked at 1.35v with my GPU at 1.2. Unfortunately I was not able to observe the change in SOC via HWMonitor when going to SOC of 1.1v. I read somewhere it's VIN4 but that's 1.85. I also read there is an offset, but couldn't guess which VIN and wasn't sure the offset was something like .3v? In any case, without a good way to check it, it seems 1.1 to 1.15 is safe and tried those to no avail.
Other sources of information as i worked on the SOC 1.1 test:
I think I might be wary of downloading that software on that page they promote. As we already discussed in another thread. The other written advice seems sound.
did you also tried to set ram resistance corretc?
my samsung b-dies have proc odt 53.3 and 30 30 30 40
What is the MAKE & MODEL Number of the RAM MEMORY you have installed. SPECCY indicates UNKNOWN with 15 gigabytes of RAM MEMORY. I copied my SPECCY RAM DATA to compare to yours. Yours doesn't indicate SPD Modules installed. My does. I have four -4 gigabytes RAM Modules by Corsair. I have a feeling all your problems may be related that you don't have compatible RAM MEMORY installed.
Mine indicate TYPE: DDR3 while yours indicates TYPE: UNKNOWN
I don't know if this is because SPECCY can't read accurately DDR4 RAM MEMORY or not. Someone will need to upload their SPECCY RAM DATA that uses DDR4 and see if it says the same thing as yours.
Plus you must have mixed Gigabytes of RAM MEMORY on the same Motherboard rather than have a Set. How do you get 15 Gigabytes of RAM MEMORY?
The only combinations are One - 8 gbyte, One - 4 gbyte, One - 2 gbyte, and One - 1 Gbytes of RAM MODULES installed to get 15 gbytes of Total RAM MEMORY.
Is you RAM MEMORY listed in the QVL List from here: ASRock > Fatal1ty AB350 Gaming-ITX/ac ?
Ryzen CPU/APU are quite sensitive to the type of RAM MEMORY installed. It is very important that your RAM MEMORY is listed in the QVL list for your motherboard.
Did you install a Set of RAM MEMORY or did you buy each RAM MEMORY separately?
Just covering basic information.