Furmark detects artefacts, but I can’t see any

[Transpanda]

I‘m once again talking about my weird mining gpu the p106-90 6gb (cut down gp106 die)

Among other things, I‘ve also been overclocking it as much as possible and I‘ve come across an issue that I wasn’t able to find anything about.
So I got the gpu running stably with the core clock at a little over 2.1ghz and the ram offset at +600mhz (samsung ram). I already shunt modded the card and the temperature is sub 70° C.

The issue arises when I set the vram clock higher. Even a bit over +600mhz Furmark starts to detect Artefacts Infrequently, but I only actually start to see them at over +710mhz

A few more things:
I know +600mhz is already pushing it, i‘m just doing this for fun. The voltage can’t be adjusted for this gpu, in software at least. The core clock starts to stutter and crash above around +260 mhz.

So basically I don’t understand why it’s detecting artefacts. Are they just really tiny, is the vram overheating, would it need a higher voltage? Also does anyone have experience with overclocking this or a similar card?

 

[Panther_Seraphin]

Visable artifacting is drastic artifacting as its multiple pixels that have to be impacted before the human eye really notices it and has to be a dramatic change as well. For example. one shade of red to another will pass the human eye test but would fail the mathematical check that software would use to flag up as artifacting.

Hence why you only visably see it at 710+ but there is failures at 600mhz being detected by software.

Its similar with System RAM where with both OS and now in DDR5 the ECC/CRC aspect will "mask" minor instability 99% of time but you can still end up some times with random lockups/crashes where you are overclocking and its not absolutley rock stable.

 

[Transpanda]

Visable artifacting is drastic artifacting as its multiple pixels that have to be impacted before the human eye really notices it and has to be a dramatic change as well. For example. one shade of red to another will pass the human eye test but would fail the mathematical check that software would use to flag up as artifacting.

Hence why you only visably see it at 710+ but there is failures at 600mhz being detected by software.

Its similar with System RAM where with both OS and now in DDR5 the ECC/CRC aspect will "mask" minor instability 99% of time but you can still end up some times with random lockups/crashes where you are overclocking and its not absolutley rock stable.

So i should aim for furmark to not detect any artefacts to avoid crashes?

 

[Panther_Seraphin]

So i should aim for furmark to not detect any artefacts to avoid crashes?

I would as it would indicate far more stability that the good ol Mk1 Eyeball. It may be fine for games etc at 650mhz 99% of the time, but there may be an edge case where it somehow causes the card to hardlock in certain games/loads.

IF that 50mhz is the make or break in terms of performance to make a game playable it may be worth it, but usually that extra 50mhz on memory is usually 0.1 FPS difference so I cannot see it being worth it.

 

[StViolenceDay]

Different benchmarks/games/engines gives different load patterns, sometimes one passes and other not.
Furmark mostly maximizes GPU die power usage, and here is the other tool targeted to maximize VRAM communication

Announcing memtest_vulkan - opensource video memory stability test

memtest_vulkan - stress test video memory for stability during overclocking or repair. Opensource & crossplatform tool written in vulkan compute. Developed as an alternative to OpenCL-based tool memtestCL Just start application, wait several minutes and stop testing by Ctrl+C. Detected errors...

www.techpowerup.com

Like artifact scanner it's designed for "launch wait-5-minutes view-results" workflow, no need to stare continuously for artifacts.