Artifracting, is it my 3080 or G9?

[Kashinoda]

My screen went black today and after a restart went black again as soon as the Windows Login screen appeared. No problems in safe mode or anything but couldn't get into Windows normally.

Couldn't for the life of me repair it so I just went with a new Windows install. I'm now getting these strange visuals glitches every now and then. I'd seen them once before but now they're frequent:
https://www.youtube.com/watch?v=BfPsDnPj-RM

Things I've tried

• Various nVidia Drivers
• BIOS update on GPU
• Firmware downgrade on monitor
• Different frequencies/resolutions

The G9 is a notorious **** show, but the glitches shown and how they behave seems to point to GPU (maybe I'm wrong). I'm hoping DisplayPort cable.

No spare monitors lying around at the moment so a real bitch to test

 

[Kashinoda]

Got a tv you can plug into?

I do, I'll have to lug it down tomorrow and see if I can replicate the error. It's annoyingly random!

 

[Kashinoda]

Just watched the video and they are a bit odd. Mate I hope that your unavailable gpu isn't at fault.

You and me both. Unfortunately it looks a bit hopeless. It seemed okay for a while then suddenly lots of artifacts followed by a black screen, audio went shortly after.

Same as before now, screen goes blank as Windows loads. Happened on the TV downstairs too using HDMI.

I've tested RAM and removed other PCIE devices, I think I should stop deluding myself at this point

In one last feeble attempt to avoid an RMA I'm giving Windows 7 a go.

 

[Kashinoda]

Loos of video on windows is a strong indicatof of GPU issues. If you have a spare hour, u can try run Mats mods to test gpu memory, also doubble check power supply to the gpu.
While in bios GPU is in low state of power, but once it goes to windows it powers up and if there is an issue with power delivery or memory or gpu chip itself then it is likely to crash or go black screen.

There doesn't appear to be a MATS/MODS leak for 400.226, which you'd need for RTX3xxx cards. Are there any other diagnostics?

 

[Kashinoda]

Nevermind, it's croaked now. Was doing some Heaven benchmarks and it wouldn't run without severe artifacts at stock. Knocking the memory down 500MHz made it stable for quite a while but eventually it crashed and now it won't even show when the BIOS is loading up.

Now the question is RMA with MSI directly or go through the retailer I bought from, both will be hell I imagine. Fun

 

[Kashinoda]

Cheers for the heads up guys, booked in a return with the retailer.

£100 for a stop-gap 1030, mad world. At least it'll slot into my HTPC when this comes back in 2023

 

[Kashinoda]

A bit of an update on this.

I filled in a return form with the retailer but they never got back to me. I was planning on just taking in store tomorrow instead but wanted to have another look so I could document the problems.

Computer booted up fine and the card is now working perfectly. I'm running tests to try and replicate the issues I was having last time. So far:

• 30 minutes TimeSpy Extreme - No Issues
• 30 minutes OCCT VRAM Test (95% VRAM) - No Issues
• 30 minutes OCCT 3D Test - No Issues
• 30 minutes MSI Kombustor (FurMark Donut, 6500MB VRAM, artifact scanner) - No Issues
• GPUMemTest 1.2, 3 versions running sequentially - all available VRAM used - No Issues

Currently running a 2 hour memtestg80 (https://github.com/ihaque/memtestG80) script at max VRAM, will see what I wake up to. EDIT - Also no issues

The only two variables that have changed are a new Display Port cable and (yet another) fresh Windows installation. It can't be the cable as it doesn't really make sense with the sort of artifacting in my video and stuff like this:



Maybe it's a power delivery issue, though I have a newish Seasonic Focus GX Gold 650W and at max system load (FurMark and Aida stress test simultaneously) I only ever read 512W from the socket. I know the 3080 is prone to transient spikes but I wouldn't expect them to manifest in the errors I was seeing.

Bamboozled right now.

 

[Kashinoda]

i would be looking to RMA the card

Well I did try to RMA it

The problem I have now is it's magically working and no matter what test I throw at it I can't replicate the problems I was having previously.

 

[Kashinoda]

how can he RMA the card if its working. Take it thro its pases keep testing it with 100% loads for few days. if its intermitend it will fail specialy on full loads.

FurMark is a tool to load power supply rails to the gpu.
3dmark is a good tool to load Silicon die and call out those artifacts.

Artifats can be memory or gpu die related. so put that card to work and see if it last a week. If it does then worry not.

Yeah I've been doubling down on the memory testing as it previously couldn't make it through 10 seconds in Heaven benchmark without artifacting everywhere, underclocking the memory seemed to make it stable for a period. Maybe that was coincidence.

I'll keep going and hopefully chalk it up to the GPU gods.

 

[Kashinoda]

I've run the 3080 through the ringer all week and it hasn't had so much as a frame stutter let alone an artifact.

I have another possible culprit but like the Display Port cable I think it's a stretch / don't understand how it could be connected. Maybe someone with a better understanding than me might know.

I have a pair of Ballistix Sport LT 3200MHz sticks which have been running overclocked since Zen2 launched 2 years ago. (3773/3600Mhz, 16-19-12-12-32). These were stable after a 24 hour stress test back when I dialed them in. In addition to the black screens I recently had a couple of blue screens (kernel related, ntoskrnl.exe) which I chalked off to nVidia drivers but I now believe were memory issues.

I've just tried to apply my OC to these sticks and I can't even push them to 3333MHz, the memory overclock fail safe doesn't even kick in and I have to BIOS reset if I do anything outside 3200Mhz 16-18-18-18-38.

I guess the questions I have are:

• Is it possible memory issues can cause visual artifacts at all? I'm thinking the artifacts shown in my video weren't typical GPU artifacting.
• Does memory usually deteriorate like this or could it be the memory controller on the 3700X has gone?

 

[Kashinoda]

**SNIP**

Haha, that's the most reassuring thing I've read yet mate. I guess if stock levels were normal I would have let this go a wee bit sooner instead of turning into this.



I will say though that the memory chips on these cards do run hot. I'm now running an undervolt (850mv) from boot and performance is actually better. I've also massively upgraded the case and airflow, need to make sure it has the best possible chance

 

[Kashinoda]

Its funny I've just stumbled across this thread. Very rarely i get little black boxes on the screen when the gpu is idle so when I'm just using internet browser or on desktop etc. Gaming and under full load is completely fine. Its also a 3080 with a R5 3600x.

Maybe the gpu wasn't seated correctly in the pci-e slot on your pc?

Early teething issues still with these drivers I bet.

On the PCI-E slot you could well have a point, there's a lot of sag on the card too and in my current mATX casse its wedged on the front fans. I have a BeQuiet 802 coming this week and a support bracket for the card.

 

[Kashinoda]

Yeah I have HWinfo but it doesn't seem to monitor the memory temps of the 3070. Only 3080 and above. Maybe only works on GDDR6x?

Try GPU-Z, though it might depend if your card even has a sensor at all. I imagine the FE does though.

With regard to voltage and temps. I run a series of 5 minute runs in Port Royal looped, ignore the average column I was resetting the sensors too early.

MSI RTX 3080 Gaming Trio at stock (current, min, max, avg)


150mv undervolt, 1875MHz clock, stock fan curve
Difference to max readings vs stock:
-3.8 degrees on the GPU
-4.0 degrees on the memory junction
-783 RPM on the cooler (very noticable)


The memory junction temp is still high, but it's a barely making noise.

150mv undervolt, 1875MHz clock, MSI Afterburner default fan curve
Difference to max readings vs stock:
-15.5 degrees on the GPU
-14.0 degrees on the memory junction
-375 RPM on the cooler (noticable)


Now we're massively down on all temps whilst still using less cooling compared to stock.

I believe the card will start to throttle at 110 degrees on the memory junction.

With undervolting my performance is better too:
Firestrike @ stock: 17,501 GPU Score, https://www.3dmark.com/spy/20101266
Firestrike @ 850mv 1875Mhz: 17,597 GPU Score, http://www.3dmark.com/spy/20101415

Also power consumption drops from 340W to 250-70W

 

[Kashinoda]

Kashinoda

I run about same memory temps at stock, just done a Timespy run and memory peaked at 86C. But I'm running full fat 350W (not overclocked, but 117% power) and the quiet Bios on the GPU, so core temps are around 65-70C

https://www.3dmark.com/spy/20129310

I can score a bit higher and run cooler (58C) if I have the bios set to Performance, but I've not bothered changing it back from when I updated the GPU bios for Resizeable bar.

Yeah your ASUS TUF is a much better card, they actually tried

The MSI Gaming Trio is meant to be the same tier but the cooling is crap without replacing thermal pads and for a high end offering it doesn't even have dual BIOS. I won't be going with MSI in the future, though the mindset these days has to be 'buy anything that's available'

That said, Timespy doesn't stress the card as much compared to Port Royal - worth running that looped for 5-10 minutes to see where your temps end up.