The RX 6000 series Owners' Club

[GerKNG]

Vulkan was better on my Vega, DX12 is WAAAAAAAAYYY better on my 6900XT.

RDNA2 seems to not like vulkan at all.

even in Rainbow Six Siege i get less FPS (my 3090 gets like 150 FPS more compared to DX)
and it stutters.

 

[moproblems99]

I have noticed that my 3900X would boost to about 4679mhz across multiple cores but now the highest I have seen is about 4550mhz. Not all that particularly worried about it but the cpu temp didn't get past 67C.

 

[Raendor]

Ok, you’re putting me to a rest a bit as vulkan was good on my previous 1080, but on 6800xt DX12 works normal, while vilkan drops usage heavily. Although vulkan was always better around loading game zones. Dx12 always stutters while you move into next map section.

 

[Felix123BU]

RDNA2 seems to not like vulkan at all.

even in Rainbow Six Siege i get less FPS (my 3090 gets like 150 FPS more compared to DX)
and it stutters.

Vulkan needs a lot more code optimization to reach peak performance, and since RDNA 2 is new and different architecturally than lets say Vega (who loved Vulkan), and those games being rater "old", don't think the developers put too much time into updating for RDNA 2, both Vulkan and DX12 being a low level API's there's so much you can do with drivers

I am a bit surprised though why RDNA 2 works so well with basically any DX12 game, but not so good with Vulkan.

 

[Raendor]

Vulkan needs a lot more code optimization to reach peak performance, and since RDNA 2 is new and different architecturally than lets say Vega (who loved Vulkan), and those games being rater "old", don't think the developers put too much time into updating for RDNA 2, both Vulkan and DX12 being a low level API's there's so much you can do with drivers

I am a bit surprised though why RDNA 2 works so well with basically any DX12 game, but not so good with Vulkan.

I hope it’s still more on the driver side. Technically api shouldn’t affect graphics series due to the way game was developed.

 

[Felix123BU]

I hope it’s still more on the driver side. Technically api shouldn’t affect graphics series due to the way game was developed.

Api's can effect a specific series, has happened before, see the Maxwell series on Nvidia's side, they where and still are "awful" at DX12 and Vulkan comparatively, and on AMD's side DX11 was worse, always below the competition no matter how much driver work was done. That does not mean that the 6000 series does not support Vulkan properly, its just that with Vulkan and DX12 the biggest gains come from the developer and the code.

This happened when Vulkan and DX12 came out, the GCN architecture was much better suited architecturally for these api's than Maxwel or Pascal, but Nvidia caught up in hardware functions and support with future gens, and with both being equally potent theoretically, its the developers who have the biggest impact on performance for a specific architecture, since even though both vendors support the api's, the architectures work slightly differently.

I am not worried by default, but if I would be the worrying guy, I would be cool as long as the game supports both Vulkan and DX12. And anyway, never games will have RDNA 2 at hand, so in future the issues should be less prevalent. I also hoped Vulkan gets more traction, but as it looks right now, its a very niche thing.

 

[GamerGuy]

Good to know, I usually run games on Vulkan (when possible) as I've gotten quite used to thinking it's better based on my experience with my Vega64 rig. Now, will always default to DX12 for my RX 6900 XT in every game which has both..at least until RDNA2 cards reach the level of performance like it has with the more mature GCN cards when running Vulkan games.

Gonna fire up Metro Exodus PC EE now, near the end game as I'm now playing the Dead City level....

 

[dragospetre88]

Guys, I saw that many of you have the power limit GPU (W) setting at a lower value than tdc limit GFX (A) in more power tool. Why is that? I thought TDC limit has to be higher... What am I missing here?

 

[xenosys]

I am on air, yes. I cannot go any higher for timespy, else it crashes. What do you mean it's on the low side? I haven't seen many rx 6900 xt posting 21k scores in Timespy on air...

In games I can push to 2625-2725, which results in 2,66-2.67 ghz constant clock speeds. But that's with 325-350W in mpt, which makes my card suck up to 375W. I am afraid to go any higher on voltage, although I have a solid super flower 1000W psu...

But anyway for daily use I undervolt to 1050 mv, 2420-2520, 2138 memory. Junction temp doesn't even go past 80 degrees, while edge temp is low 60s with fan curve to 45%.

You're correct. 21K is an excellent score, so you've got a decent card on your hands. It's well above average on 3D Mark, but probably about average on this particular forum, which is generally full of overclockers. 19k is about average for a 6900XT, according to 3D Mark.

If you get that under water, you can probably push 21.5k if at any point you're inclined with load temps being a lot lower.

 

[turbogear]

You're correct. 21K is an excellent score, so you've got a decent card on your hands. It's well above average on 3D Mark, but probably about average on this particular forum, which is generally full of overclockers. 19k is about average for a 6900XT, according to 3D Mark.

If you get that under water, you can probably push 21.5k if at any point you're inclined with load temps being a lot lower.

The Hardwareluxx has collected statistics that show performance distribution of different RDNA2 cards:

[User-Review] - Statistik-Analyse: So schlägt sich Navi 21 in 3DMark Time Spy

Statistik-Analyse: So schlägt sich Navi 21 in 3DMark Time Spy Drei Enthusiasten haben Zeit und Hirnschmalz investiert, um öffentlich einsehbare Daten zur Radeon RX 6800, 6800 XT und 6900 XT im Grafik-Benchmark Time Spy unter die Lupe zu nehmen. Dies ist das Ergebnis. Inhaltsverzeichnis: Erste...

www.hardwareluxx.de

 

[Felix123BU]

The Hardwareluxx has collected statistics that show performance distribution of different RDNA2 cards:

[User-Review] - Statistik-Analyse: So schlägt sich Navi 21 in 3DMark Time Spy

Statistik-Analyse: So schlägt sich Navi 21 in 3DMark Time Spy Drei Enthusiasten haben Zeit und Hirnschmalz investiert, um öffentlich einsehbare Daten zur Radeon RX 6800, 6800 XT und 6900 XT im Grafik-Benchmark Time Spy unter die Lupe zu nehmen. Dies ist das Ergebnis. Inhaltsverzeichnis: Erste...

www.hardwareluxx.de

According to that forum, highly interesting charts, my 6800XT is godly, mine can reach 2670 Mhz average real frequency in gaming that is off the charts of that chart

This was only my second reference card after the RX 480, all other where custom and not one was such a good chip as this 6800XT 20600 in Timespy with a 6800XT is what it can do with a 100% stable OC.

Interestingly enough, driver 21.3.1 is still the highest performer for the 6800XT and 6900XT in those readings also.
Also interesting is that according to that chart, the batch of reference cards are between the best performers for the 6800XT and 6900XT. (though that might be due to those being more often water-blocked)

Guys, I saw that many of you have the power limit GPU (W) setting at a lower value than tdc limit GFX (A) in more power tool. Why is that? I thought TDC limit has to be higher... What am I missing here?

I think I am missing what you want to ask, first you ask why folks here have the GPU (W) setting lower than GFX (A)(which is TDC), then toy say TDC has to be higher, which you already asked in the first part of the question, so you are asking why TDC is higher here, when it should be higher, as you also state?

 

[dragospetre88]

8

According to that forum, highly interesting charts, my 6800XT is godly, mine can reach 2670 Mhz average real frequency in gaming that is off the charts of that chart

This was only my second reference card after the RX 480, all other where custom and not one was such a good chip as this 6800XT 20600 in Timespy with a 6800XT is what it can do with a 100% stable OC.

Interestingly enough, driver 21.3.1 is still the highest performer for the 6800XT and 6900XT in those readings also.
Also interesting is that according to that chart, the batch of reference cards are between the best performers for the 6800XT and 6900XT. (though that might be due to those being more often water-blocked)


I think I am missing what you want to ask, first you ask why folks here have the GPU (W) setting lower than GFX (A)(which is TDC), then toy say TDC has to be higher, which you already asked in the first part of the question, so you are asking why TDC is higher here, when it should be higher, as you also state?

Yeah, my mistake... Basically I wanted to know why People have power limit higher than tdc limit. I always have tdc higher when using mpt.

 

[turbogear]

According to that forum, highly interesting charts, my 6800XT is godly, mine can reach 2670 Mhz average real frequency in gaming that is off the charts of that chart

This was only my second reference card after the RX 480, all other where custom and not one was such a good chip as this 6800XT 20600 in Timespy with a 6800XT is what it can do with a 100% stable OC.

Interestingly enough, driver 21.3.1 is still the highest performer for the 6800XT and 6900XT in those readings also.
Also interesting is that according to that chart, the batch of reference cards are between the best performers for the 6800XT and 6900XT. (though that might be due to those being more often water-blocked)


I think I am missing what you want to ask, first you ask why folks here have the GPU (W) setting lower than GFX (A)(which is TDC), then toy say TDC has to be higher, which you already asked in the first part of the question, so you are asking why TDC is higher here, when it should be higher, as you also state?

Yes mine 6800XT (R.I.P. edition) was also Godly.
Mine was running all time at 2700MHz in games.

Maybe our theory of first batches of reference cards being all Gold samples is true.

 

[Felix123BU]

8
Yeah, my mistake... Basically I wanted to know why People have power limit higher than tdc limit. I always have tdc higher when using mpt.

Basically to have more wattage available for OC's. though that can be dangerous for the chip and the power circuitry and not recommended (except if maybe you only increase Power Limit GPU and not TDC). Generally if you increase limits you should do it linearly, that would always leave your max TDC higher than max GPU power limit for Navi 2.

Yes mine 6800XT (R.I.P. edition) was also Godly.
Mine was running all time at 2700MHz in games.

Maybe our theory of first batches of reference cards being all Gold samples is true.

"Maybe our theory of first batches of reference cards being all Gold samples is true . " --- could very well be true, first batch of GPUs where all great OC-er, have not heard about crappy first batch 6000 series. It would be a new marketing style for the new Lisa Su AMD

 

[HaggtGT]

Basically to have more wattage available for OC's. though that can be dangerous for the chip and the power circuitry and not recommended (except if maybe you only increase Power Limit GPU and not TDC). Generally if you increase limits you should do it linearly, that would always leave your max TDC higher than max GPU power limit for Navi 2.


"Maybe our theory of first batches of reference cards being all Gold samples is true . " --- could very well be true, first batch of GPUs where all great OC-er, have not heard about crappy first batch 6000 series. It would be a new marketing style for the new Lisa Su AMD

My 6800 XT Red Devil stock bios has a really low default TDC? You said to leave it higher than your limit but this card is way below it never mind way above it. Do you know why?

 

[xenosys]

Yes mine 6800XT (R.I.P. edition) was also Godly.
Mine was running all time at 2700MHz in games.

Maybe our theory of first batches of reference cards being all Gold samples is true.

Well I also have a Powercolor Reference 6800XT and was hitting 20950 in Timespy a couple of months ago, so you may be right.

 

[Felix123BU]

My 6800 XT Red Devil stock bios has a really low default TDC? You said to leave it higher than your limit but this card is way below it never mind way above it. Do you know why?

Not sure (maybe the power delivery cooling is not the best and they toned it down). Here is my default stock MPT readings

Looking at this (1) GPU PCB Breakdown: RX 6800XT Red Devil - YouTube
I can say your TDC is very low, my reference board has a smidge weaker PCB design as far as power delivery goes, and since your cards can theoretically handle much more, if I where you, I would put 300 A (if it does not blow up, up to 330) TDC Limit GFX and 300 W Power Limit GPU, that should be in theory safe enough, of course if the card does not get out of hand with temperatures. But again, I am not telling you to do this, its only what I would do if I had that card, and if you do it, I take no responsibility for what might happen if you do it

 

[turbogear]

My 6800 XT Red Devil stock bios has a really low default TDC? You said to leave it higher than your limit but this card is way below it never mind way above it. Do you know why?

Yes on Red Devil it is opposite to reference card.
If I think on reference it was 255W and 300A TDC by default.

I did not also understand why the Devil is doing opposite to reference design.
Maybe it depends on chip quality?

Basically if I understand correctly one of the above mentioned values is the Core Power target and the other is Maximum Current target of core.
Power is related to voltage and current:
Power=Voltage x Current

So if we have same power target on reference and Devil, then on reference lower voltage and higher current will produce this power target.
On Devil it would be opposite higher voltage and lower current give the power target.

Why is it set like this on some of non reference cards?
That is a question I don't know the answer for.

 

[Felix123BU]

Well I also have a Powercolor Reference 6800XT and was hitting 20950 in Timespy a couple of months ago, so you may be right.

Yup, mine is also a Powercolor Reference 6800XT

 

[dragospetre88]

Yes on Red Devil it is opposite to reference card.
If I think on reference it was 255W and 300A TDC by default.

I did not also understand why the Devil is doing opposite to reference design.
Maybe it depends on chip quality?

Basically if I understand correctly one of the above mentioned values is the Core Power target and the other is Maximum Current target of core.
Power is related to voltage and current:
Power=Voltage x Current

So if we have same power target on reference and Devil, then on reference lower voltage and higher current will produce this power target.
On Devil it would be opposite higher voltage and lower current give the power target.

Why is it set like this on some of non reference cards?
That is a question I don't know the answer for.

My default vbios for red devil rx 6900 xt has 281 W power limit and 320 tdc limit gfx. When i use more power I tend to go for 325 W power limit and 345-350 tdc limit gfx. Is this ok? From what I understand from you it is not...

 

[turbogear]

My default vbios for red devil rx 6900 xt has 281 W power limit and 320 tdc limit gfx. When i use more power I tend to go for 325 W power limit and 345-350 tdc limit gfx. Is this ok? From what I understand from you it is not...

I did not say if it is not okay.
It was only my observation that the reference and Red Devil 6800XT are configured differently.
I was trying to find the reason behind it.
I am also trying to understand why some cards are configured differently.

Below is the default setting on my 6900XTU:
The total Power Limit is 382W for me with 15% from Radeon Software.
i.e. 332x1.15=382W.

TDC set by PowerColor is 320A. Again in this case TDC is set with value of 62 below the max Power limit.

I have currently TDC at 340A in MPT. I will try increasing it to 360A.
This would still be below the 382W Power Limit.

 

[Felix123BU]

My default vbios for red devil rx 6900 xt has 281 W power limit and 320 tdc limit gfx. When i use more power I tend to go for 325 W power limit and 345-350 tdc limit gfx. Is this ok? From what I understand from you it is not...

As long as you do not get stability issues and to high temps, its perfect so to say. As turbogear said, seems some cards are configured a bit differently (red devils in particular), for reasons unknown to us
But as long as you get the extra performance, and are not more than +15% in MPT vs stock settings (10% plus in MPT I would consider conservative), that would give you an additional 30% extra power overall, and as long as you don't notice instability or abnormally high temps on any of the CPU, RAM, VRM etc, you are golden

 

[dragospetre88]

As long as you do not get stability issues and to high temps, its perfect so to say. As turbogear said, seems some cards are configured a bit differently (red devils in particular), for reasons unknown to us
But as long as you get the extra performance, and are not more than +15% in MPT vs stock settings (10% plus in MPT I would consider conservative), that would give you an additional 30% extra power overall, and as long as you don't notice instability or abnormally high temps on any of the CPU, RAM, VRM etc, you are golden

Junction temp does not GO above 95 degrees C and the performance is up roughly by 8-10% compare to stock settings. Pretty good I would say.

 

[Felix123BU]

Junction temp does not GO above 95 degrees C and the performance is up roughly by 8-10% compare to stock settings. Pretty good I would say.

Yup, a 10% overclock is not bad at all, quite good actually Foarte bine

 

[dragospetre88]

Yup, a 10% overclock is not bad at all, quite good actually Foarte bine

Just did some fresh testing at 4k - default preset vs oc (SAM enabled with both). Horizon Zero Dawn hit 84 fps on average with my oc, up from 78. Valhalla reached 80 fps up from 72, and metro exodus enhanced reached 78 from 71 (just 1440p, high Ray tracing, ultra preset - The game lags too much at 4k). 1% lows and 0.1% lows were significantly higher too.