Drivers & Software

[rawintellect]

It takes a big man to admit when he's wrong. I'm admitting it...somewhat. I built this computer a year ago. It's a Threadripper 2950x with 64GB of Corsair Vengence RAM 2 M.2 2TB drives 1 2TB SSD and 3 HDD. It was built for MOSTLY a workstation due to my job as a mainframe analyst. When I built the rig I just slapped two RX590s in there because as stated gaming wasn't a big concern. 

Well that changed with COVID. As I spent more time at home I decided to purchase a Powercolor RX 5700XT (overclocked version). So as soon as I put in the RX 5700XT the BSODs/black screens began. Puzzled but seeking refuge here in AMDs forums I saw that there were NUMERIOUS issues with early versions of the drivers with these cards. So I figured I would work with any issues as I'm not exactly a computer noobie. I've been building rigs since I was 12 in 1982. 

And I did! Here is the long list of what I've tried to fix these issues:

First I started with the BIOS. I made sure everything was stock. No overclocks.

Checked the hardware. Used many different mem-test software checking programs. Even used windows memory diagnostic extended tests. Nothing.
Ran PRIME95 to check the CPU. Nothing.

DDU uninstalled in safe mode. Too many times to count. Factory reset too!

Then made my own more aggressive fan cure as I thought it was overheating.

Then undervolted the card for less heat.

Then added a TdrDelay to the registry.
Then checked the permissions of all files installed by the driver I actually found a permissions issue with saving custom tuning profiles. Fixed. Search out my thread I created for that answer.
Then started limiting framerates using chill since FRTC had been removed. This actually helped a lot. 
Checked for viruses/malware. Nothing.
Formatted and reinstalled windows.
Changed to HDMI cables from DP. Changed back to DP and tried all the other DP inputs on the card.

By now I was nearly a year in and due to all of the things I did above my BSODs/Black screens were only happening infrequently. But NVIDIA was looming and I bought a new laptop that ran an RTX 2060 super and saw how flawless it was working. I came to have legitimate concerns that a desktop that I built and cost considerably more in parts wasn't performing as well as my laptop. I was making plans to give up on NAVI.

Then I found it. I'm not sure if this was the problem all along as there WERE confirmed issues with the drivers but over the months I've seen others state that these issues had been resolved. But here was the issue: AMAZON. When building this computer I ordered all my RAM at once. Specifically CL18 ram. Cas latency is far more important than the MHz a RAM stick runs at. 

Since I ordered all my ram at once it NEVER occurred to me that these RAM sticks would not be identical. They weren't. Two of them were CL19. Yes!!! CL19. I revealed this by using the TAIPHOON software to review the product number for the RAM and it showed two of the sticks were CL19 and made by Samsung. While the other 6 were Micron and CL18. 

I immediately went back to my order history with AMAZON and saw that I did indeed order CL18 and paid for CL18. I got 6 CL18 and 2 CL19 sticks. Upset because these sticks had been in my system for almost an entire year I was concerned about complaining to AMAZON now. But to make this long story short I photographed the sticks and sent them to customer service. They immediately RMA'd the two sticks and sent me what I paid for.

I knew the moment I removed those two CL19 sticks the problem was fixed because I could CAUSE BSODs to happen by turning off Chill and allowing the FPS to ramp up to max. I used the most demanding game I had as a test. Death Stranding. I cranked up all options for the game in both Adrenaline and the game itself.  0 BSODs.

My PC has now been running for a solid 3 weeks with no BSODs or Black Screens. As stated before nothing failed any mem tests I put the rig through. I only found this by chance using TAIPHOON. 

I hope my tale of woe helps someone and I apologize to AMD for believing my card was at fault. I might even attempt overclocking this rig in the future. I'm definitely going to have a serious look at Big NAVI now.

P.S. Fix your permission issue with custom tuning profiles and add HWS or HAGS and I'll be eternally grateful. 

[defectivebydesign]

Motherboards need to run RAM at the same timings, and mixing brands/timing can cause compatibility issues. This was the fault of the vendor for sending the wrong RAM.

That said, people can try to mix brands and timings, but only by running all the RAM at the same settings as the worst RAM sticks. So, if one ram stick is CL19, and one is CL18, you have to run ALL of them at CL19. If you run them all at CL18, you are overclocking the CL19 RAM, which isn't always possible.

This is one of the biggest problems with the 3rd party DRAM Calculator, which always tells people to run their RAM sticks at overclocked timings the RAM does not work at, and only really works for high ends sticks that can overclock.

Ryzen CPUs then have had very weak memory controllers until the 3000 CPUs, which do not support 4 sticks very well. This also includes BIOS support, which has been spotty until very very recently.

Also, unless you are running software that explicitly makes use of mGPU, dual GPUs are useless on AMD because they removed Crossfire support. It is now a very niche feature that only works under niche use cases.

[rawintellect]

defectivebydesign wrote:

Motherboards need to run RAM at the same timings, and mixing brands/timing can cause compatibility issues. This was the fault of the vendor for sending the wrong RAM.

I know this. I stated this.

That said, people can try to mix brands and timings, but only by running all the RAM at the same settings as the worst RAM sticks. So, if one ram stick is CL19, and one is CL18, you have to run ALL of them at CL19. If you run them all at CL18, you are overclocking the CL19 RAM, which isn't always possible.

I took the default settings. With 6 sticks of ram running CL18 and 2 CL19 the board still showed them running at CL18. 

This is one of the biggest problems with the 3rd party DRAM Calculator, which always tells people to run their RAM sticks at overclocked timings the RAM does not work at, and only really works for high ends sticks that can overclock.

I didn't use a DRAM calculator. I took the default settings.

Ryzen CPUs then have had very weak memory controllers until the 3000 CPUs, which do not support 4 sticks very well. This also includes BIOS support, which has been spotty until very very recently. 

This is a Threadripper 2950x on a x399 board. It supports 8 sticks of RAM.

Also, unless you are running software that explicitly makes use of mGPU, dual GPUs are useless on AMD because they removed Crossfire support. It is now a very niche feature that only works under niche use cases.

As stated the original purpose of the build was a workstation. Compiling and decompiling system dumps from a mainframe is both CPU and GPU intensive. Crossfire definitely alleviated the lag associated with doing this task. Which is also why I have so much storage on this machine.