Should I be concerned about my RX6900XT memory

[rav3n72]

Just a quick question for the experts out there. Every review I can find on the net overclocks the RX6900XT memory from 2000 to 2150 without any problems. I am not so lucky, but I was wondering is it just me or are there more cases like mine out in the wild. If I overclock my memory to let say 2050 I don't see any difference in 3dmark, but above that (2100) I see that the scores drop a little bit. At 2150, the card feels a bit unstable even in windows. I have the card now for about 7 months and it just stays like that and not degrading.

If I set the timing to fast (which I use 24/7) the 3dmark scores are noticeably higher, which is strange that the memory behaves so much differently between the timing and speed. So in short, I am running the card with fast timings, 10% power (311W) and stock clocks (+- 2580) 24/7 at 2k 165hz gaming and I have not experienced anything wrong with the card.

So, have you guys experienced this? Could it be that they changed from samsung to hynix, etc and the reviewers samples are different? Lastly, it is not cooling as I am getting the exact same results with stock cooling as with water cooled.

I am happy with this card and will keep it for another year or so, I am just wondering about the memory, that is all.

Thanks guys.

 

[VPII]

Just a quick question for the experts out there. Every review I can find on the net overclocks the RX6900XT memory from 2000 to 2150 without any problems. I am not so lucky, but I was wondering is it just me or are there more cases like mine out in the wild. If I overclock my memory to let say 2050 I don't see any difference in 3dmark, but above that (2100) I see that the scores drop a little bit. At 2150, the card feels a bit unstable even in windows. I have the card now for about 7 months and it just stays like that and not degrading.

If I set the timing to fast (which I use 24/7) the 3dmark scores are noticeably higher, which is strange that the memory behaves so much differently between the timing and speed. So in short, I am running the card with fast timings, 10% power (311W) and stock clocks (+- 2580) 24/7 at 2k 165hz gaming and I have not experienced anything wrong with the card.

So, have you guys experienced this? Could it be that they changed from samsung to hynix, etc and the reviewers samples are different? Lastly, it is not cooling as I am getting the exact same results with stock cooling as with water cooled.

I am happy with this card and will keep it for another year or so, I am just wondering about the memory, that is all.

Thanks guys.

Hi @rav3n72 in all honesty it is somewhat difficult to say. I know that GPUz show which memory hip you are using, in my case with my RTX 3080 Ti it is micron. Now from what I have seen, the highest stable clocks is usually with Samsung chips, not sure why but it is what I have seen with my last 3 or 4 graphics cards. However, the micron memory on my card is stable and give increased results at 1000mhz higher than stock. Please understand, I am talking about Nvidia so it might be different with AMD (ATI). Unfortunately my experience with ATI or AMD now is very limited.

 

[dainluke]

No one can do 2150. Like no one. Just because someone thinks they’ve completed a stable OC doesn’t mean they actually have. The card also loosens its own timings above 2120 I believe. 2120-ish seems to be the limit for most of the RDNA 2 card samples. At least none of the cagillion RDNA 2 cards I’ve owned could do higher than 2100-2120. Perhaps I’m just a bad overclocker; GPU OCing is extreeeeeeemely difficult after all.

The only 69XT that reliably does high memory clock is the XTXH LC version with the 18Gbps Samsung chips. That card can reliably do 2370MHz-ish. All Samsung G6 chips have similar potential tbh. The 14Gbps chips all do around 16, and the 16Gbps chips can all do just shy of 17Gbps. The 18Gbps chips do just shy of 19Gbps.

The 15Gbps base chips in the 3060 actually also cap out at 16Gbps, NVIDIA kinda just added a factory OC for marketing purposes. I learnt that when I accidentally clocked one to 17Gbps without realising the base speed was higher and saw massive artefacting in Cyberpunk-2077. That was kinda funny ngl.

From my experience, G6X is a bit more of a mixed bag. Most G6X I’ve encountered is good for 21Gbps (except this one trash 3080 I had that capped out at 20 lmao). The 3090 is good for around 21.5 and interestingly enough, both 3080 Ti’s I owned were also good for 21.5. I’d wager to guess that NVIDIA are using the same bin of G6X in the 3080 Ti as the 90.

Also, memory temperature definitely plays a role. Not a major role but it definitely helps not to be sitting at 100-110c. It still boggles my mind how people are comfortable with those temps on their G6X, but then aren’t comfortable with 1.4 Vcore on Intel silicon.

 

[rav3n72]

@VPII According to GPU-Z I have samsung chips

@affxct Thanks, that explains a lot.

 

[Juan91]

I played around memory overclocking on gpus only slightly.

On Amd Radeon software you should be able to choose between two memory timing profiles and the other profile might be tighter which might gain you some performance or reduced latency.

Your memory chips might be running hot whilst the core is cool on a certain fan percentage so you can try setting the fan speed to 100% and try memory OC then, id it's better or stable then you can try and resolve that issue.

Last option is to research into modding vbios for memory timings (crypto mining sites might be of better help)

But I doubt even with the best memory overclock on ambient air temperature that your performances will increase more than 10%