The RX 6000 series Owners' Club

[Felix123BU]

So have people tried any Ray Tracing benchmarks on 6900Xt/6800XT cards? It seems that both these cards are placed between 3070 and 3080 for RT performance (reasonable settings).

Not so much benchmarks, but in RT games they are between the 3070 and 3080 in regards to RT performance generally speaking.

 

[xenosys]

So have people tried any Ray Tracing benchmarks on 6900Xt/6800XT cards? It seems that both these cards are placed between 3070 and 3080 for RT performance (reasonable settings).

Native RT performance for the 6800XT is between the 3070 and 3080 indeed, with the 6900XT being closer to the 3080 and 6800XT closer to the 3070 in terms of 3D Mark scores.

The RTX 3070 averages around 8300 in Port Royale, whereas the 6800XT is around 9500, 6900XT is 10300 and RTX 3080 is 11500. I can personally hit around 10600 in PR with my 6800XT under water and pushing 2.7Ghz, so if you manage to luck out and get yourself a good chip, you can push it a bit more.

In games however, there isn't much to separate the 6800XT and 3070. Watchdogs Legion delivers better performance via the 6800XT by a couple of frames when RT is turned on for example.

 

[Godhand007]

Not so much benchmarks, but in RT games they are between the 3070 and 3080 in regards to RT performance generally speaking.

I was going by performance in Metro EE, synthetics and some AMD sponsored titles. Though if one is limited completely by RT, it seems that 6900XT barely exceeds 3070, at least in Metro EE. The thing is, game scenes are not just completely rendered RT effects and thas why RX 6900 XT can almost match rtx 3080 in Metro EE benchmark with all RT effects on but it falls behind in RT limited scenarios. So the problem is we don't know what amount of RT is good enough and what is pasted in RTX games (almost all sponsored by Nvidia) like 64x Hairworks or Teseelation times 30.

Native RT performance for the 6800XT is between the 3070 and 3080 indeed, with the 6900XT being closer to the 3080 and 6800XT closer to the 3070 in terms of 3D Mark scores.

The RTX 3070 averages around 8300 in Port Royale, whereas the 6800XT is around 9500, 6900XT is 10300 and RTX 3080 is 11500. I can personally hit around 10600 in PR with my 6800XT under water and pushing 2.7Ghz, so if you manage to luck out and get yourself a good chip, you can push it a bit more.

Yeah. That seems about right. I can get ~11100 Plus on my RX 6900 XT. The thing is, is this the correct performance for reasonable RT? For e.g. For RT Normal has good performance below but RT Ultra just tanks 6900 XT. Also, watch that they have turned down the raster setting one notch to make scenarios RT limited since increasing the raster setting helps alleviate stress on AMD in this particular title.

 

[Space Lynx]

I wish ray tracing never happened, if they had focused instead on raw FPS, we would all be gaming at 240hz 1440p 240fps with no dips, and our mouths would drop at how smooth and immersive games had become. Ray tracing is and always will be too detrimental to FPS hit even with DLSS, the few games that support RT only a handful of them even make it look good.

oh well. /shrug

 

[Godhand007]

I wish ray tracing never happened, if they had focused instead on raw FPS, we would all be gaming at 240hz 1440p 240fps with no dips, and our mouths would drop at how smooth and immersive games had become. Ray tracing is and always will be too detrimental to FPS hit even with DLSS, the few games that support RT only a handful of them even make it look good.

oh well. /shrug

Agree with this to an extent. Though people will say that GPUs were already reaching the limits of rasteriaztion and Nvidia took a genius step toward bringing something new (of course RT had been around for some time before that). We also need CPUs and Game engines to support proper 144 fps gameplay let alone 240 fps (e-sports titles excluded).

About games looking good with RT, shadows, and lighting are good but this mirror-like puddle reflections craze is beyond me.

 

[xenosys]

I was going by performance in Metro EE, synthetics and some AMD sponsored titles. Though if one is limited completely by RT, it seems that 6900XT barely exceeds 3070, at least in Metro EE. The thing is, game scenes are not just completely rendered RT effects and thas why RX 6900 XT can almost match rtx 3080 in Metro EE benchmark with all RT effects on but it falls behind in RT limited scenarios. So the problem is we don't know what amount of RT is good enough and what is pasted in RTX games (almost all sponsored by Nvidia) like 64x Hairworks or Teseelation times 30.


Yeah. That seems about right. I can get ~11100 Plus on my RX 6900 XT. The thing is, is this the correct performance for reasonable RT? For e.g. For RT Normal has good performance below but RT Ultra just tanks 6900 XT. Also, watch that they have turned down the raster setting one notch to make scenarios RT limited since increasing the raster setting helps alleviate stress on AMD in this particular title.

View attachment 201382

Interesting graph, thanks. It would seem that even the 2080ti/3070 stands up well to the 6900XT when RT is extensively utilised. The less RT becomes a factor, the further away the 6900XT pulls as you say.

 

[Godhand007]

Interesting graph, thanks. It would seem that even the 2080ti/3070 stands up well to the 6900XT when RT is extensively utilised. The less RT becomes a factor, the further away the 6900XT pulls as you say.

But that's the point. Is RT being heavily utilized? The problem is that we don't have a proper RT showcase game from AMD to give us an alternative to Nvidia-sponsored RTX games. Also, you might want to read the relevant bits of the article as it seems that Hairworks might be causing huge performance dips on AMD. The best sort of "objective" test that we have is 3D Mark Port Royal where RX 6900 is ~10% behind an RTX 3080.

 

[Chomiq]

RT is not going away, it is here to stay and is needed for actual progress in realtime 3D graphics.

 

[Godhand007]

RT is not going away, it is here to stay and is needed for actual progress in realtime 3D graphics.

I don't think anyone here is saying anything remotely close to that. There are many things needed for progress in graphics Technology, from HDR, AI Upscaling/conversion, RT, and many other things if one is talking about 3D games.

 

[Kabouter Plop]

I got a 6900 XT a month ago put it on water i get gpu temps now of 50c and 65c max hotspot usually avg out at 60c 62c it only peeks for like 1 second to 65c with a triple 360mm radiator setup fans only running about avg 1000 rpm usually very silent with room temp around 22c
This is my first fit if heard thermal pads on some chips ekwb recommends can sometimes vary in height, altho if seen users that failed fit their block with hotspot temps above 80c and 90c so i do not think i fitted it improperly, thermal paste if used X patern with mx4 temps have improved it started at 68c max hotspot temp 1 hour after refilling loop before i bleeded out all air, if i set my old fan profile i had with gtx 1080 my gpu idles at 37c hotspot else it idles at 33c hotspot my fans do stop spinning a little at idle, while if i use my old profile it stops spinning for a lot longer.

 

[Felix123BU]

I got a 6900 XT a month ago put it on water i get gpu temps now of 50c and 65c max hotspot usually avg out at 60c 62c it only peeks for like 1 second to 65c with a triple 360mm radiator setup fans only running about avg 1000 rpm usually very silent with room temp around 22c
This is my first fit if heard thermal pads on some chips ekwb recommends can sometimes vary in height, altho if seen users that failed fit their block with hotspot temps above 80c and 90c so i do not think i fitted it improperly, thermal paste if used X patern with mx4 temps have improved it started at 68c max hotspot temp 1 hour after refilling loop before i bleeded out all air, if i set my old fan profile i had with gtx 1080 my gpu idles at 37c hotspot else it idles at 33c hotspot my fans do stop spinning a little at idle, while if i use my old profile it stops spinning for a lot longer.

Those temps are perfectly fine, no need to worry

 

[Kabouter Plop]

Those temps are perfectly fine, no need to worry

What kinda temps you get with rage profile 270.300w power limit on water ?

 

[Felix123BU]

What kinda temps you get with rage profile 270.300w power limit on water ?

My 6800XT on water with slightly raised power limits via MPT(MorePowerTool) (so around 350w max) gets 29c idle, around 43c on load, around 56c on hotspot, but that with Liquid Metal between the block and chip, with normal paste I got basically the same temps you now have. LM reduced temps by around 10c with the same setup.

 

[Kabouter Plop]

Don't think i would dare mess with LM and risk damaging the card, altho i am sure it could be aplied safely.

 

[Felix123BU]

Don't think i would dare mess with LM and risk damaging the card, altho i am sure it could be aplied safely.

Well, LM is not really needed except if you want the lowest possible temps, that's why I said your temps are perfectly fine as they are

 

[turbogear]

Well, LM is not really needed except if you want the lowest possible temps, that's why I said your temps are perfectly fine as they are

Yes especially on 6800XT LM is not a must.
With good thermal pastes like Thermal Grizzly Kryonaut or MX-4/MX-5 one can get decent temperatures but of course nothing compared to LM.

The LM is amazing, I was hitting 92°C on Hotspot at 400W in benchmarks like Fire Strike with my 6900XTU Liquid Devil Ultimate with original factory paste.
I have changed to LM and the results are unbelievable.
In same benchmarks I don't go higher than 65°C on hotspot with 10°C delta to GPU temperature.
In Furmark, hotspot is staying around 75°C.

Until now I had been running it at 2750MHz@1175mV with 2550MHz on min frequency bar without memory OC.

Now I try to tune the 6900XTU to get best possible clock after changing to LM.
I limited myself to 415W Core power and 400A TDC. I was able to get 2750MHz max bar and 2600MHz min bar with 1085mV but score in Timespy (21400) remains similar to what I had before at 2750MHz@1175mV because real frequency is in same range of 2690MHz-2700MHz. The power consumption could be lower as lower voltage is applied now.
Note this is without Memory OC. I have not Oced memory at the moment as I first want to get stable Oc on GPU before going to memory as well.

In Heaven benchmark I can run it for hours at 2850MHz max bar and 2700MHz min bar at 1100mV without issues. Clock in this benchmark hover in range of 2750MHz-2775MHz, but any setting higher than 2750HMz max clock is unstable in 3DMARK benchmarks though.
Fire Strike test for example for setting 2850MHz stop after a few minutes though power consumption doesn't go higher than 415W at the point where it stops.
Tried also to play with voltages upto 1200mV to try to get 2850Mhz stable in Fire Strike but until now was not able to get any 3DMARK benchmarks to complete for Frequencies higher than 2750MHz.

Somehow 2750MHz seems like a brick wall until now for me.

 

[Butanding1987]

I knew you couldn’t resist it.

 

[xenosys]

I'm currently getting around 53-55 degrees (hotspot of between 70 and 75) under water with Kryonaut on my 6800XT drawing around 370w overclocked to 2.7Ghz (2112Mhz Fast Timing Memory).

What sort of temp drops would I likely get by switching to LM with the above settings? 5 degrees? 10 degrees?

 

[Kabouter Plop]

Probably 10c but liquid metal can destroy aluminium potentially damage gpu if not used correctly, its probably also conducting electricity so you dont want it touching parts it should't altho i am not sure about that last one.

 

[Felix123BU]

I'm currently getting around 53-55 degrees (hotspot of between 70 and 75) under water with Kryonaut on my 6800XT drawing around 370w overclocked to 2.7Ghz (2112Mhz Fast Timing Memory).

What sort of temp drops would I likely get by switching to LM with the above settings? 5 degrees? 10 degrees?

my 6800XT with LM was -12c on hotspot and -10 and core vs Kryonaut

 

[xenosys]

Probably 10c but liquid metal can destroy aluminium potentially damage gpu if not used correctly, its probably also conducting electricity so you dont want it touching parts it should't altho i am not sure about that last one.

Really? Interesting. I've heard people use nail polish around the GPU die to ensure the LM doesn't short any components it may touch should it leak/run. I've also got myself an EK Quantum Vector block which is Nickel-plated Copper so that shouldn't pose a problem.

my 6800XT with LM was -12c on hotspot and -10 and core vs Kryonaut

That's amazing. So I could be looking at anything around 45 degrees with LM and a hotspot of 60-65 at full load. I think I've pushed my card as far as it can go when overclocking. I doubt it'll be able to push it higher than 2.7Ghz stable in games but the temp drop is worth considering.

 

[Godhand007]

Seeing guys here with such low temps on their cards; I am just thinking how much my reference card is being help up by temps. My current settings are 2665Mhz (max), 2150MHz Mem ,~430W, 1175mv (can't get stability on anything lower) which are stress tests stable. Clocks always remain above 2600MHz but junction temp can reach 110C for certain extreme situations which forces a downclock. Any Ideas on how I can increase my clocks and get to sweet that 2.7 GHz game stable settings while on air?

Also, are people here not afraid of voiding their warranty when messing around with cards?

 

[Felix123BU]

Really? Interesting. I've heard people use nail polish around the GPU die to ensure the LM doesn't short any components it may touch should it leak/run. I've also got myself an EK Quantum Vector block which is Nickel-plated Copper so that shouldn't pose a problem.



That's amazing. So I could be looking at anything around 45 degrees with LM and a hotspot of 60-65 at full load. I think I've pushed my card as far as it can go when overclocking. I doubt it'll be able to push it higher than 2.7Ghz stable in games but the temp drop is worth considering.

Nickel-plated Copper would not have issues with LM, only copper and aluminum have issues with LM, aluminum has most problem with LM, copper a bit less, and nickel plated basically none, except maybe some staining but not corrosion. In the past I also used LM on a copper heatsink, id did some very minor damage, but nothing to bad. On nickel LM is almost harmless in general.

I was amazed by how much the temps dropped after applying LM on the chip, was expecting a good drop but not as much I used normal MX4 thermal paste to cover everything around the chip so that if any LM drop would travel somewhere else, it would not cause any damage, since its non conductive and easy to apply and cover, and also easy to remove, plus being a heat conductor

For the 6800XT, 2.7ish is the max it can go, I read somewhere there was a good rason for AMD to put 2.8 max, since the logic inside the chip could not handle more than 2.8. The only reason for applying LM is to have great temps, and maybe to help stabilize your max overclock. For me it was 100% worth it, just for the temps, I am also getting 2.7, 2725mhz at best.

Temps are around 44-45c on chip and 56-57 max on hotspot with a 350w max limit, usually using 310-320w.

Seeing guys here with such low temps on their cards; I am just thinking how much my reference card is being help up by temps. My current settings are 2665Mhz (max), 2150MHz Mem ,~430W, 1175mv (can't get stability on anything lower) which are stress tests stable. Clocks always remain above 2600MHz but junction temp can reach 110C for certain extreme situations which forces a downclock. Any Ideas on how I can increase my clocks and get to sweet that 2.7 GHz game stable settings while on air?

Also, are people here not afraid of voiding their warranty when messing around with cards?

"Also, are people here not afraid of voiding their warranty when messing around with cards?" yes and no
Yes because GPUs are such a rare commodity these days, and no because I have done this things countless times in the past and not afraid of doing it again, also if I would be scared of tuning any piece of hardware, I would buy a console and live happily ever after Plus its no fun if you don't take any risks

 

[Godhand007]

"Also, are people here not afraid of voiding their warranty when messing around with cards?" yes and no
Yes because GPUs are such a rare commodity these days, and no because I have done this things countless times in the past and not afraid of doing it again, also if I would be scared of tuning any piece of hardware, I would buy a console and live happily ever after Plus its no fun if you don't take any risks

I am not opposed to tweaking hardware or else I wouldn't be pushing my card so hard on Air. The point about the warranty is that sticker removal or evidence of change is introduced while doing thermal paste change etc., which is almost a guaranteed warranty rejection with few exceptions. It's fun to do bungee jumping but pays to keep the cord attached.

These are my settings for reference RX 6900 XT. What are your thoughts? Am I overdoing it or being too moderate ?

 

[Felix123BU]

I am not opposed to tweaking hardware or else I wouldn't be pushing my card so hard on Air. The point about the warranty is that sticker removal or evidence of change is introduced while doing thermal paste change etc., which is almost a guaranteed warranty rejection with few exceptions. It's fun to do bungee jumping but pays to keep the cord attached.

I got the best idea ever from this forum, buying warranty stickers, there is a discussion on the topic here . The only thing that would be worrisome for me with the reference cards is that at least for the 6800XT, they come with a thermal pad, not paste, and that is hardly salvageable, that could be a reason for warranty rejection, but then again, if the stickers are intact, you can make the case that's how you got it, I know, its far-fetched. Another issue with the reference cards, and it happened with mine, I tried to put paste on it, removed the pad, had the unpleasant surprise that the cold plate was not very even and I could not get hotspot under control anymore, that made me get a waterblock. I was relatively happy with the stock cooler, but I "needed" to thinker with it

From looking at your settings, they are not overly aggressive, your problem would be the reference cooler has its limits, and if it also came with a thermal pad and not paste, that would not allow it to much OC room anyway. You could try to replace the paste, but that might not work, did not work for me, and the cooler itself is perfectly fine for stock settings, but not really for hard OC's.