Hello all! Apologies in advance for the long post, but I’m at the end of my knowledge on this. As some background, I purchased a 6900XT open box from Microcenter some weeks back. the card was bad out of the box, so I RMA’d with Asus, they sent back the card I have now. When I first got the card back, I started having issues, though the issues have changed throughout testing. I’ve tried to keep a good record of the troubleshooting I’ve done, so I’ll list everything in chronological order below.
tl;dr
I have tried a number of troubleshooting tests to figure out what’s going on with my system, details at the bottom. GPU crashes under stress tests, underperforms other systems with similar specs in games, and occasionally suffers stuttering and latency issues. I’m not sure if this is a GPU issue or a system issue, however. Though I believe the issues lie with the GPU.
To start, I initially installed it in my normal case, an ITX case that uses a riser. Once everything was installed, I immediately had huge issues with stuttering and latency issues, which I immediately thought would be the riser. I moved everything to a test bench, and had the same issues - ruling out the riser.
Taking a minute to describe the issue in a bit more depth, games would slowly start to “lag”, or run behind. The performance of the card would slowly drop, and audio would start to crackle and distort, then get stretched and slow down to the point where it was inaudible. This wasn’t isolated to a specific game, but certain games were worse than others. Cyberpunk took about 2 minutes to be unplayable, Monster Hunter World took about 10-15, Doom Eternal remained playable throughout, but the audio issues and small lags were noticeable throughout. Specifically, driving around in Cyberpunk, 1440p high/ultra settings I would start around 90-100fps, then drop to under 30. Monster Hunter would drop from 200+ fps to under 80fps (ruiner nergigante event quest, in his little arena), Doom would run at around 140-160 fps. Looking at power monitoring, the card was self-reporting lower than expected power draw, often in the 80-140w range as opposed to the expected 275w. Measuring with my Kill-a-watt confirmed power draw was low as well. Temperatures were well under control, often in mid 60s to low 70s, though sometimes going as high as 80c. Benchmarks/stress tests would complete, but with hamstrung performance. I tried both WQHL and latest drivers, and a driver reinstall for each with DDU, no changes.
After some googling, some folks said a windows reinstall could help, so that was my next step. At first, it seemed like it helped, though further testing confirmed the problem was still there, though to a much lesser extent. Now, games were playable, though performance was still below what I would expect to see. Doom had no audio crackling issues or performance issues, though the performance was a bit lower than expected. MHW was running slower that expected, but again, no issues. Cyberpunk continued to have issues, but they were much more manageable than before.
I continued to troubleshoot, with the first step being to check if, for some reason, pcie 3.0 vs. 4.0 might be causing some issues. I had set the card to run in 3.0 mode to avoid issues with my riser, and kept it like that when moving back to the bench. I tried flipping it back to 4.0, no changes. Next, I thought there might be some issues with the GPU not getting enough power, since this is all running on a SF600 600w PSU. I had never seen power draw go above 460w on my kill-a-watt, but you never know. Trying with a EVGA 1000w, as well as a Seasonic 1250w both still had the performance issues. I know both of these units are good for the wattage, so I ruled out PSU. At this point, following some other forum posts I tried testing my latency with Latencymon, it showed some serious issues and latency problems, confirming there was an issue. Previously, this system had worked fine for years so I didn’t expect any issues with hardware there, but I did check both of my SSDs to make sure they weren’t failing on me or throwing up issues, they’re both fine. As for the other components, memory and CPU are both running at stock, automatic speeds, the memory passed memtest86 overnight, so no issues there.
Continuing to troubleshoot, new drivers came out for the GPU. I figured it was worth a shot, and they have eliminated a good bit of the stuttering and audio issues. After updating drivers, the issue seemed to be gone, though performance remained an issue. there are many games where even at 4k, the GPU isn’t getting fully loaded, power draw is low, clock speed is low, etc. Everything “works” though, but at anywhere from a 20-40% performance penalty. I wanted to quantify this, and ran Time Spy for an easy number to compare to other systems. I don’t have any modern games with built in benchmarks, and getting specifics on gameplay for things like MHW and Doom Eternal can be difficult since most people don’t report what level they’re testing on. The test went fine, though my system was in the bottom 7% of all similar configurations. I chalked that up to it being an uncommon configuration, and the fact that my memory was running quite slow. I re-enabled some known-good memory settings, and things improved, moving up to the 60 percentile. I wanted to confirm that everything was ok, so I ran the Time Spy stress test and my system black screened. I figured it was probably memory being finicky, maybe I set a timing wrong, so I reset to stock and ran again - also a black screen. The system wouldn’t display anything after rebooting, so I had to boot to safe mode and uninstall drivers, but I was able to recover the system.
At this point, I was starting to suspect memory, or god forbid motherboard or CPU issues. I had the opportunity to test in two separate systems, though testing was a bit limited by time and what we had available. The first system I tested on was an old X58 system, less than ideal. It had a X5650 clocked to 4.4ghz though, and we were able to run at 4K, which helped to focus the stress on the GPU. In that system, everything ran fine, even on extended tests! Though, we weren’t able to test every game, just MHW and Doom. I was also able to test on a friend’s system, a 3900x running with a decent overclock. Everything was fine there too, and ran like a dream. No issues with stuttering, power draw, or other issues in either Cyberpunk or Doom. At this point, I assumed it had to be my memory, since “it’s always memory”. I borrowed some ram, known-good, from my bud, then went home and tried it. I had the same issues, ruling that out.
I was interested to see if I could replicate the crashing under load, and with either RAM kit, both at stock, both at XMP/known-good timings, the system will black screen during stress tests and cant load the display when rebooting. Interestingly, I went to bed after one crash, and when I tried to start the computer in the morning, the display loaded up without the need to go in and fiddle in safe mode, which further complicates things.
Looking back on the two systems we tested with, I’m concerned the X58 might not have been loading the 6900XT enough, and we didn’t really test for an extended period of time on my friend’s computer because of a lack of time. I don’t have a second system on hand I can test with here, though I’m working on an unrelated X58 project that should be ready here soon. I can use that to troubleshoot the GPU, though I still have concerns that the CPU can’t keep up, even at 4K.
Now, I find myself with a question of what is going wrong. At this point, I think it’s the GPU. I’ve ruled out the PSU, riser, and RAM, and previously the system never gave me any issues. However, with the GPU seemingly working fine in two other systems, I’m left wondering if something isn’t wrong with my system. I don’t have another card with comparable power requirements or performance to the 6900XT to test with, but everything works fine with my trusty old 290. The inverse is true as well, I don’t have a system close to my 3700x to test the GPU with, the closest I can get is a clocked to snot X58 + X5650. I’m looking for any advice or anything I might have missed here, I don’t want to pull the trigger on another RMA if that isn’t the issue, though I will if that’s what needs to happen.
I’ll include detailed specs and a line by line of all the things I’ve tested below, but please let me know if you have any questions! I have some Afterburner recordings/grabs of the power issues saved somewhere, if memory serves, and I could grab some more as well if needed.
My System
- CPU - 3700x run at stock throughout
- Mobo - Gigabyte B550I Aorus Pro AX
- RAM - Crucial Balistix Max 4000mhz 2x8gb kit run at “stock” - whatever the motherboard decided that is, as well as a “safe” 3600 CL16 setting
- Testing RAM - Corsair Vengence 3200mhz 2x8gb kit run at “stock”, as well as XMP
- GPU - Asus 6900XT TUF
- PSU - Corsair SF600 600w PSU
- Testing PSU - EVGA G1 1000w, Seasonic 1250-X
- SSDs - Intel 660P 1TB NVME, Silicon Power 1TB NVME
X58 Test System
- CPU - X5650 Clocked to 4.4ghz, known stable OC
- Mobo - EVGA X58 Classified Rev1 westmere modded for X5650, known-good
- RAM - 12gb Corsair 1600 clocked at 14XXmhz, due to FSB OC
- SSD - Generic Samsung 256gb
- PSU - EVGA G2 1000w
Buddy’s System
- CPU - 3900x known good OC on this, not sure the settings
- Mobo - Gigabyte X570I Aorus Pro Wifi
- Ram - 2x8gb G.Skill Trident Z Royal 3200mhz not sure of the settings on this, but known-good
- PSU - Silverstone SFX SX750
- that’s all I know for this one, but I know it works fine with his overclocked 3080 as well
Troubleshooting tests
- tested on bench, without riser
- Driver reinstall with DDU
- WQHL Drivers, resinstall with drivers
- Latencymon Tests
- Clean windows install
- PCI-e 3.0, PCI-e 4.0
- Disabled onboard audio, used external DAC instead
- two separate PSU tests, both known-good
- tested in X58 system
- tested in Buddy’s system
- Testing with known-good RAM
Any help would be great! I would LOVE to have just missed something obvious.