The Radeon RX 6900/6950 XT Owners Thread.

[BKUK]

Had a quick play using the settings above in the standard Amd software.



Reading back through the threads it looks like other methods would squeeze more? May try MPT as described try and hit 17K

 

[Kaapstad]

System arrived today just about up and running. And advice on first to do’s? Just about to run 3dmark.

Welcome and on TROH.

Spoiler

The Roll of Honour.

Date 30/03/2021, Make Sapphire Nitro+, BKUK, Link
Date 18/03/2021, Make Sapphire Nitro+, xR00Tx, Link
Date 16/03/2021, Make AMD Reference, Sumanji, Link
Date 11/03/2021, Make PowerColor Red Devil, mikehhhhhhh, Link
Date 09/02/2021, Make AMD Reference, Pourma, Link
Date 06/02/2021, Make PowerColor Red Devil, John Mike, Link
Date 04/02/2021, Make PowerColor Red Devil, djbully, Link
Date 04/02/2021, Make Sapphire Nitro+, atomic7431, Link
Date 03/02/2021, Make AMD Reference, snips86x, Link
Date 03/02/2021, Make XFX Merc 319, LtMatt, Link
Date 03/02/2021, Make AMD Reference, SumpNut, Link
Date 03/02/2021, Make Asus Strix LC, JediFragger, Link
Date 31/01/2021, Make PowerColor Red Devil, littlepuppy, Link
Date 30/01/2021, Make AMD Reference, ExRayTed, Link
Date 27/01/2021, Make AMD Reference, wookiee87, Link
Date 26/01/2021, Make AMD Reference, muziqaz, Link
Date 26/01/2021, Make AMD Reference, Conanius, Link
Date 24/01/2021, Make PowerColor Red Devil, eeii, Link
Date 23/01/2021, Make PowerColor Red Devil LE, Too Tall, Link
Date 15/01/2021, Make AMD Reference, Overlag, Link
Date 14/01/2021, Make XFX Merc 319, Seanoog, Link
Date 14/01/2021, Make PowerColor Red Devil, Heofz, Link
Date 13/01/2021, Make XFX Merc 319, ch99mr, Link
Date 10/01/2021, Make AMD Reference, Breezer, Link
Date 10/01/2021, Make XFX Merc 319, ronski, Link
Date 07/01/2021, Make AMD Reference, Biffa, Link
Date 06/01/2021, Make PowerColor Red Devil, heartburnron, Link
Date 04/01/2021, Make Sapphire Nitro+, Mister_tangerine, Link
Date 01/01/2021, Make Sapphire Nitro+, Jumper118, Link
Date 01/01/2021, Make AMD Reference, Jumper118, Link
Date 31/12/2020, Make AMD Reference, NathG79, Link
Date 30/12/2020, Make PowerColor Red Devil, Nugsy88, Link
Date 27/12/2020, Make AMD Reference, LtMatt, Link
Date 26/12/2020, Make AMD Reference, Abhishek Shukla, Link
Date 24/12/2020, Make AMD Reference, ch99mr, Link
Date 24/12/2020, Make Sapphire Nitro+, hornytoe, Link
Date 23/12/2020, Make AMD Reference, Diplodocus, Link
Date 23/12/2020, Make AMD Reference, DeathStorm, Link
Date 23/12/2020, Make PowerColor Red Devil, Jonnym28, Link
Date 17/12/2020, Make AMD Reference, DustyMiller, Link
Date 16/12/2020, Make AMD Reference, Curlyriff, Link
Date 16/12/2020, Make AMD Reference, Chewie, Link
Date 16/12/2020, Make AMD Reference, Too Tall, Link





To get on the Roll of Honour all you have to do is post a pic with your user name on it of your Radeon RX 6900 XT.​

 

[Andytom69]

Hi.. a greeting to all the guys in the forum, thanks for the space you grant.
I am a new owner of a custom amd 6900xt, never owned this brand. I have always had Nvidia that seems to be fine automatically on a bit of everything. Amd radeon seems to need some special settings, for example I have a benq 4k ips 60 freesync monitor, and basically there seems to be something strange about the performance , especially in some games.
Can you give me some basic tips on how to use this video card? thanks
[URL=http://www.imagebam.com/image/85a50f1373627380] [/URL]

 

[LtMatt]

Hi.. a greeting to all the guys in the forum, thanks for the space you grant.
I am a new owner of a custom amd 6900xt, never owned this brand. I have always had Nvidia that seems to be fine automatically on a bit of everything. Amd radeon seems to need some special settings, for example I have a benq 4k ips 60 freesync monitor, and basically there seems to be something strange about the performance , especially in some games.
Can you give me some basic tips on how to use this video card? thanks

Could you be a bit more specific about your system, what games you are having issue with, what settings you are using, and what the exact issue is?

 

[BKUK]

Hit 16800 in Timespy 20324 Graphics
Min 2575
Max 2675

1065mv

Just with Wattman not MPT, However on hwmonitor max power hit 337W. Thought that would be lower stock? Or is the nitro xt higher?

 

[Wozzey]

Ty!
Yes, now my 6900xt is watter cooled!
Clock hovers around 2895mhz on Shadow of Tomb Raider.

I'm sorry but I do not believe you. Do you have any screenshot to prove this. Not meant to be an insult here but I haven't seen a single 6900xt that will actually reach clocks above 2.8ghz without a hardware mod. Let alone sustained clocks. Please post a screenshot because if what you are saying is true it's literally a world record clock for the card without a hardware mod

 

[Wozzey]

I'm sorry but I do not believe you. Do you have any screenshot to prove this. Not meant to be an insult here but I haven't seen a single 6900xt that will actually reach clocks above 2.8ghz without a hardware mod. Let alone sustained clocks. Please post a screenshot because if what you are saying is true it's literally a world record clock for the card without a hardware mod

Ok so you aren't getting that actual clock in games. Either you are just lying or you don't know what you are talking about and are mistaken somewhere (i'm 99% leaning towards this). There is way, way, way too much missinformation with rdna2 overclocking so I don't fault you for being mistaken. Here are my 1080p results in shadow of the tomb raider, where you supposedly got almost 3ghz.
1080p
https://ibb.co/mBKTqp8
https://ibb.co/HH6yZwv
I have 2 fps less than you at 1080p, I am cpu bottlenecked here just like you. However you are 96% gpu bound, and I am only 82%. Using simple math that means my card is performing 17% better than yours at supposedly much less clocks of 2.75ghz. Nothing else to talk about. Most likely you have your gpu set to 2.94ghz, yet never actually get even close to those clocks.

Really not meant to be insulting here in any way. I literally made an account just to ask you to show clocks since that would have been so ground breaking. Alas it was sadly not true. Had my hopes up.

 

[LtMatt]

I'm sorry but I do not believe you. Do you have any screenshot to prove this. Not meant to be an insult here but I haven't seen a single 6900xt that will actually reach clocks above 2.8ghz without a hardware mod. Let alone sustained clocks. Please post a screenshot because if what you are saying is true it's literally a world record clock for the card without a hardware mod

It’s true, you should see his Timespy score in the benchmark thread. He’s beating 3090’s for fun in a bench that favours green.

 

[Wozzey]

It’s true, you should see his Timespy score in the benchmark thread. He’s beating 3090’s for fun in a bench that favours green.

Sorry again, none of my posts are meant to offend, but no. He had a average/sustained clock of 2.7ghz and a peak of 2.76ghz in the test you are referencing https://www.3dmark.com/spy/19157079
That is right in line with the peak these cards can achieve without a voltage increase past 1175mv which is impossible as of right now. It also puts it right around where my card sits, so I'll go try and beat his score in the morning. Haven't ran that benchmark since I first got my card, will post results later of where I end up on the scoreboard. For reference I hold the number 2 spot in port royal https://www.3dmark.com/pr/971517

 

[LtMatt]

Sorry again, none of my posts are meant to offend, but no. He had a average/sustained clock of 2.7ghz and a peak of 2.76ghz in the test you are referencing https://www.3dmark.com/spy/19157079
That is right in line with the peak these cards can achieve without a voltage increase past 1175mv which is impossible as of right now. It also puts it right around where my card sits, so I'll go try and beat his score in the morning. Haven't ran that benchmark since I first got my card, will post results later of where I end up on the scoreboard. For reference I hold the number 2 spot in port royal https://www.3dmark.com/pr/971517

It's okay, no offence taken here my friend.

It's normal to have a lower average clock in synthetics though. I suspect his in game core clock is higher than in Timespy when running games. Would need to know how he is monitoring the clock, i suspect with an overlay.

The reason why i think his clocks are accurate, or close to them is due to how mugh higher his Timespy score is. Here is my Timespy run from a month or two ago. I have barely a lower average core clock than him but my score is significantly lower. AMD Radeon RX 6900 XT video card benchmark result - AMD Ryzen 9 5950X,ASUSTeK COMPUTER INC. ROG CROSSHAIR VIII HERO (3dmark.com)

I am interested to see your scores though, make sure you post them here when done.
Time Spy Standard DX 12 Bench. | Overclockers UK Forums

 

[mikehhhhhhh]

Ok so you aren't getting that actual clock in games. Either you are just lying or you don't know what you are talking about and are mistaken somewhere (i'm 99% leaning towards this). There is way, way, way too much missinformation with rdna2 overclocking so I don't fault you for being mistaken. Here are my 1080p results in shadow of the tomb raider, where you supposedly got almost 3ghz.
1080p
https://ibb.co/mBKTqp8
https://ibb.co/HH6yZwv
I have 2 fps less than you at 1080p, I am cpu bottlenecked here just like you. However you are 96% gpu bound, and I am only 82%. Using simple math that means my card is performing 17% better than yours at supposedly much less clocks of 2.75ghz. Nothing else to talk about. Most likely you have your gpu set to 2.94ghz, yet never actually get even close to those clocks.

Really not meant to be insulting here in any way. I literally made an account just to ask you to show clocks since that would have been so ground breaking. Alas it was sadly not true. Had my hopes up.

I don't know what kind of maths they are, but they are pretty funky and far from simple

I look forward to your Timespy score - its' going to be off the charts if your card legitimately performs 17% better

 

[Wozzey]

I don't know what kind of maths they are, but they are pretty funky and far from simple

I look forward to your Timespy score - its' going to be off the charts if your card legitimately performs 17% better

you are right about that ia was very simple math that is probably wrong lol. But I was just trying to emphasize that what he's saying his card performs at and what it actually does isn't matching up. Based on his scores there is no doubt his card is good/extremely good. But most 6900xts without a power limit and water cooled will perform around there. However what isn't right is the clock speeds. Again I literally made an acount just to try and confirm this lol. It would be that impressive if it's real since not a single 6900xt has been able to actually sustain 2.8ghz+ without voltage modding. The average sustained clocks at the highest end are in the 2.7-2.8ghz range. So if he can actually show me a screenshot or video of a 100% gpu load and sustained (not just peak/blip) clocks of over 2.8ghz i will bow down before his cards silicon lottery glory and take my words back. I however highly doubt it is true. Hence my post. I live in the US do after work ill get to trying my scores. I did 2 runs last night but went to bed early. Will post updates later

Edit - I should note that just like cpus, under light work load the clocks can be sustained at a higher amount than full load. That is why I emphasized the 100% gpu load. Under less than 90%load the clocks tend to creep up. I (and I assume most 6900xts under the same config) can get over 2.8ghz sustained on lightly gpu used games, typically under 50% load or so.

 

[heartburnron]

you are right about that ia was very simple math that is probably wrong lol. But I was just trying to emphasize that what he's saying his card performs at and what it actually does isn't matching up. Based on his scores there is no doubt his card is good/extremely good. But most 6900xts without a power limit and water cooled will perform around there. However what isn't right is the clock speeds. Again I literally made an acount just to try and confirm this lol. It would be that impressive if it's real since not a single 6900xt has been able to actually sustain 2.8ghz+ without voltage modding. The average sustained clocks at the highest end are in the 2.7-2.8ghz range. So if he can actually show me a screenshot or video of a 100% gpu load and sustained (not just peak/blip) clocks of over 2.8ghz i will bow down before his cards silicon lottery glory and take my words back. I however highly doubt it is true. Hence my post. I live in the US do after work ill get to trying my scores. I did 2 runs last night but went to bed early. Will post updates later

Edit - I should note that just like cpus, under light work load the clocks can be sustained at a higher amount than full load. That is why I emphasized the 100% gpu load. Under less than 90%load the clocks tend to creep up. I (and I assume most 6900xts under the same config) can get over 2.8ghz sustained on lightly gpu used games, typically under 50% load or so.

I think the accuracy of the GPU utilisation figure reported in the Tomb Raider benchmark is questionable. I've seen a few conversations here before where the metric has seemed contradictory so my understanding was that it was at best, indicative and not to be fully relied upon. I could be mistaken, but I believe this to be true.

 

[eeii]

+15 on the power consumption
Change memory to fast timings and manually set to 2112 Mhz
Set max core clock to 2700 Mhz.
Undervolt to 1100 Mv
That should get you fairly close to maximum stable performance for both games and benchmarks. That's a good place to start anyway and then you can tweak accordingly from there if you're so inclined.

Loving these settings working a treat.

 

[heartburnron]

Loving these settings working a treat.

Glad to hear it. I can't take credit for the 2112 Mhz on the memory. We have @LtMatt to thank for that following his experimentation. You can set the memory right across to 2150 and it will likely still be stable, but for most games you actually lose a little bit of performance as Matt proved. You might find you can under volt a bit further, down to 1075 - 1080 which may allow for your average boost clocks to go a bit higher. A find 1100 is a good balance between the additional clock boost and complete stability though.

 

[Wozzey]

I think the accuracy of the GPU utilisation figure reported in the Tomb Raider benchmark is questionable. I've seen a few conversations here before where the metric has seemed contradictory so my understanding was that it was at best, indicative and not to be fully relied upon. I could be mistaken, but I believe this to be true.

I'm sure it's not a very reliable or accurate measurement, I was more just using it as an example as that's what I had to work with. For some reason timespy seems to be giving me massive instability in the second gpu test. Not sure why since last night I was able to do a couple runs just fine. I'll post the best one here https://www.3dmark.com/spy/19425588 which is obv pretty far from roots score but not too far off. I wanted to see if I was unstable in any other benchmark and low and behold I beat my previous record and the number 1 port royal spot (for 6900xt's) by a mile lol https://www.3dmark.com/pr/973106. Not sure why timespy is acting so weird Obv port royal is stable and no game ever shows instability at my current game profile of 2700mhz/2800mz yet at those settings timespy gpu test 2 almost instantly crashes. If anyone has any insight I'm all ears. I'll do some messing around though. I probably have some other system instability I am not aware of. I just really, really want to see a screenshot or something of mr root getting 2.8ghz or more in games while under load. I'm sure the dude has a life and doesn't get on everyday lol but I just need to know wince it would quite literally be world record breaking if true. Oh well. I will post any relevant updates of my own if they come.

 

[heartburnron]

I'm sure it's not a very reliable or accurate measurement, I was more just using it as an example as that's what I had to work with. For some reason timespy seems to be giving me massive instability in the second gpu test. Not sure why since last night I was able to do a couple runs just fine. I'll post the best one here https://www.3dmark.com/spy/19425588 which is obv pretty far from roots score but not too far off. I wanted to see if I was unstable in any other benchmark and low and behold I beat my previous record and the number 1 port royal spot (for 6900xt's) by a mile lol https://www.3dmark.com/pr/973106. Not sure why timespy is acting so weird Obv port royal is stable and no game ever shows instability at my current game profile of 2700mhz/2800mz yet at those settings timespy gpu test 2 almost instantly crashes. If anyone has any insight I'm all ears. I'll do some messing around though. I probably have some other system instability I am not aware of. I just really, really want to see a screenshot or something of mr root getting 2.8ghz or more in games while under load. I'm sure the dude has a life and doesn't get on everyday lol but I just need to know wince it would quite literally be world record breaking if true. Oh well. I will post any relevant updates of my own if they come.

In my experience, Time Spy is the ultimate test for stability. I've had overclocks pass many stress tests and play hours of games yet when it comes to Timespy I've crashed. I'm sure if you lowered you clocks to something like 2600 / 2700-2730 you'll find you pass test 2 no problem. If not, then you could also consider that Timespy is very sensitive to memory; tighter RAM timings will be reflected in your scores and it's good for testing any changes you make, however if there is any instability with the RAM timings, then this could also be the cause of the crashes.

You graphics score of over 22,000 is really good though. I haven't been able to break 21,000 with mine. Here is my best effort: https://www.3dmark.com/spy/19443954

 

[Wozzey]

In my experience, Time Spy is the ultimate test for stability. I've had overclocks pass many stress tests and play hours of games yet when it comes to Timespy I've crashed. I'm sure if you lowered you clocks to something like 2600 / 2700-2730 you'll find you pass test 2 no problem. If not, then you could also consider that Timespy is very sensitive to memory; tighter RAM timings will be reflected in your scores and it's good for testing any changes you make, however if there is any instability with the RAM timings, then this could also be the cause of the crashes.

You graphics score of over 22,000 is really good though. I haven't been able to break 21,000 with mine. Here is my best effort: https://www.3dmark.com/spy/19443954

In regards to memory/ram do you mean vram or system ram?? I do have a pretty hard overclock on my system ram so if that's what you mean it could be it. I'll go downclock it and get back to you.
Also nice score, I have a custom loop on my rig so I am in the best situation to get good scores. Anyone who gets even close on air is very impressive!

 

[heartburnron]

In regards to memory/ram do you mean vram or system ram?? I do have a pretty hard overclock on my system ram so if that's what you mean it could be it. I'll go downclock it and get back to you.
Also nice score, I have a custom loop on my rig so I am in the best situation to get good scores. Anyone who gets even close on air is very impressive!

Yeah I mean system RAM. Timespy has been very unforgiving to me when I've been tweaking my system RAM timings so that could well be the culprit of your crashes.

Out of curiosity, I just tried running Timespy with your settings of 2700/2800. I was able to complete the full test and saw the clock boost to and hold over 2650 Mhz however my graphics score dropped by 2000 points so there must be some throttling or the like going on for me on air. I tried again with the voltage dropped down to 1080 Mv but still bringing the graphics score down to circa 18500. Going past 2700 Mhz is definitely only a game for the water cooled it would seem...

 

[JediFragger]

In my experience, Time Spy is the ultimate test for stability.

@LtMatt has suggested to me in the past that looping 20 runs of GPU Test 2 is usually solid for him for games.