RDNA2 RX6000 Series Owners Thread, Tests, Mods, BIOS & Tweaks !

Yeah I'm excited for RDNA 3, but I'm in no rush. This card is awesome and when AMD releases their spring drivers and DLSS equivalent, then you can even see Ray Tracing becoming viable.

 

You have one of the greatest cards that is almost able to approach 2.8GHz.
If I remember you were using stock voltage and giving higher power by MPT to reach those high benchmark scores.

Did you also try undervolting your card a little?
That could save some power but still could be able to reach high frequencies.
I read that 6900XT does not like that much undervolt though.

On my 6800XT I set it to 2675MHz@985mV and it reaches in games clock up to 2640MHz.

I am getting more and more interested in 6900XT Red Devil because of higher power target that one can use with MPT.
It could be that one of those replaces my 6800XT.

As a matter of fact one 6900XT Red Devil is on its way to me.
Let see how lucky I would be with silicon lottery.

I suppose selling my nice overclocker reference 6800XT on eBay would not be a problem. It was already fitted with EK waterblock by me.

Alphacool has released a water block for the Devil so that rose my attention. I think under water this card could be even better.

Here is the link to that block. It seems that it was very recently announced as delivery is 4-5 weeks.
https://www.alphacool.com/shop/graf...eon-rx-6800-xt/6900xt-red-devil-mit-backplate

 

So I experimented with PowerColor Red Devil 6900XT.
The card has been sent back as it was not that great compared to my really good 6800XT watercooled.
I could only get 4-5% higher performance compared to my 6800XT while consuming 64W more power on same frequency.

2675MHz@1065mV was highest stable setting with 6900XT set to consume 380W and 384A TDC. VRAM was set at 2120MHz with fast timing. Higher VRAM clock was unstable.
Higher voltages were unstable for this frequency and delivered less performance without going insanely high with power consumption.

Time Spy Score:
https://www.3dmark.com/3dm/59221295

Port Royal Score:
https://www.3dmark.com/3dm/59221465

I was not able to go higher than 2675MHz without crash.
I could set 2700MHz@1070mV where Time Spy was stable but not Port Royal.
https://www.3dmark.com/spy/18853323

I tried higher and lower voltages but it seems 2675MHz is maximum I could have achieved.

The 6800XT is running similar frequency to 6900XT and is working very stable since weeks with 2650MHz@985mV with total core power set to 316W and TDC 310A. That is much less power target than 6900XT Red Devil with only 4-5% missing in performance.
Time Spy Score of ~19800 on 6800XT.
This was only 4% less performance than what I could achieve with 6900XT Red Devil on same frequency.
Borderland 3 and Assassin's creed valhalla benchmark showed 5% improvement over 6800XT.

As higher performance was not possible to achieve with Red Devil 6900XT without going crazy with power consumption, I returned the card and would not sell my nice 6800XT.

I actaully expected better from 6900XT but was disappointed with my sample.

 

Yeah.... that is exactly my point --- if you have a really nice clocking 6800xt then there is no point to " "upgrade" " to 6900xt (double quotations were on purpose).
The 6900xt is just not worth the 50%+ price vs 6800xt (that is - MSRP vs MSRP comparison).

Anyway I can say that your Red Devil was not THAT bad - my 6900xt becomes unstable at anything lower than 1150mV so I'm a little bit jelly here

 

Yes after my recent experience, I cannot also recommend 6900XT over 6800XT.

Yes my Red Devil was not the worst sample but not also among the best. There are better ones out there as I have seen on forums like here and Igor's Lab.
My problem was that for small gains for 6900XT over my 6800XT it ate power like Devil's furnace.

Maybe my view would have been completely different if I did not had a great 6800XT sample.

 

So I pushed my Red Devil 6900 XT up to 500 Watts in MPT and scored top 18 of all time in graphics scores vs RTX 3090's in FireStrike Ultra on air. Ran the test again, 15,574 so within the margin of error. Average GPU clock speed: 2656 MHZ and top clock was 2758 MHZ.

Second Test Result: https://www.3dmark.com/3dm/59353460?

 

That's INSANE ! Oh man ! Congrats !

 

Thanks : D For proof here is Furmark to test the power draw. Note this is not the total board power JUST the GPU power draw lol. Add an additional 50 watts for safety and you are right up against the 525 watt limit of 3x8 pin power and the PCI -E slot.I'm right up against the thermal limits of 110 C junction temps, but this is just to show the GPU is actually drawing the wattage set in MPT for testing purposes.

 

Sorry, I must have missed the post you made. My voltages are at the "stock" limit of 1175 mV and yes using MPT to increase the power limit. I did try to undervolt, yet I ran into the card using the highest voltages anyways, and clocks became unstable. So I decided to lower the memory bandwidth and retain the higher voltages because Wattman doesn't really let you do a "true" undervolt. Only MPT does that for now. That 6800 XT is a monster by the sounds of it on water. I'd love to try a custom water block on this Red Devil for fun. Cost though, pretty expensive.

 

Thanks for info.
I gave back the Red Devil 6900XT as my sample was not as good as yours.
For me it was unstable if I push frequency slider higher than 2675MHz without decreasing the voltage slider to a range of 1060mV.
Time Spy score of 20600 was limit as it did no go higher with any further increase of power in MPT beyond 400W and 384A TDC for above mentioned frequency setting.
All higher than 2675MHz setting were unstable regardless of which voltage I tried.
The VRAM was unstable above 2120MHz on fast timing.

At the end with the stable setting I was only getting 4-5% higher performance than my 6800XT watercooled card but with much higher power consumption so I decided that it does not make much sense to keep the card and replace my 6800XT.

Maybe as you mentioned memory bandwidth is the key. How did you lower the memory bandwidth? With MPT?

 

20600 is a very respectable number in Time Spy. I meant the same thing you stated earlier, the VRAM was unstable so I lowered it, same as you. The 6800 XT is an amazing card under water. Loving this RDNA 2 architecture.

 

W1zzard has released a new version of GPU-Z v2.38.0.
The problem with GPU-Z saving only 512kB of bios file size for RDNA2 6000-series cards has been fixed.
RDNA2 cards have 1024kB bios size.

Flashing those wrong bioses from TPU to your cards would have led to bricking your card.

I reported this problem to W1zzard yesterday and he fixed it on the same day and released new GPU-Z already.

He deleted all Radeon 6000-seriee bioses from TPU bios database so nobody can uses these.
Here is the link:
https://www.techpowerup.com/279635/techpowerup-gpu-z-v2-38-0-released

I think we community should support now by submitting our bioses through GPU-Z so the database can be built again now with correct bios size.

 

Just submited my
RX 6800 XT Gaming X TRIO

Finally we can update bios on the cards. Waiting for the 6900xt bios on mine

 

Whaaa ? This issue was known since the release of 6800xt and 6900xt..... It was reported to W1zzard countless times...
For examples that it was known since the release check Igor`s Lab Forums.... We were testing 6xxx bios modding with the RBE Team and hellm ....since forever there....
It is actually good that he FINALLY fixed it.

 

Now someone just needs to test 6900xt bios on a 6800xt. So at least we can have more voltage to give a bit more clocks. thats what I need on mine

 

The tests have been done on the Igors Labs..... I am participating since nearly the start... For now everything becomes unstable and there are even problems when flashing the stock bios back.... so... BEWARE.
Once a "stable" solution will pop up for the 68xx owners you can be sure I will post it here.

 

Interesting, good to see it fixed but I assumed W1zzard wasn't aware of it since GPU-Z has been updated once or twice already since some of this discussion around the 6000 series GPU bios size started.
Maybe getting a good bios file or adding the newer software from what some of the GPU vendors providing bios updates for the 6000 series were using just took a while.

That little lock on the driver and bios still makes soft modding via more power tool a bit of a better solution but if the bios lock could be be bypassed it's good that there's a way to dump bioses now through GPU-Z without risking bricking should these older ones have been used on GPU's that do not provide dual-bios switches.

Wonder how well the GPU's actually scale it reads like the 6700's aren't that far behind the 6800 despite the hardware (cores and such.) difference so maybe the cards just boost a lot better from higher speeds though optimized code and efficiency could see the 6800XT and even 6900 really scale once RDNA2 is utilized more thoroughly.
(A earlier post showed some good gains from focusing on the Wave32 format and benefits to ray tracing performance already over Wave64 so it's not impossible some tailored optimizations could shift the GPU scaling more towards these additional CU's and all that.)


AMD's cutting the cores down to close to the 5700XT level yet it's another 10 - 15% difference, still a bit into the week before the new 6700's actually launch and that will show a more complete pictures as reviews come up but I find it interesting.
Well utilization was already a thing with Vega but RDNA made improvements over GCN though it's not too surprising I guess that high clock speeds count for more although each GPU being a "up-to" 10% performance diff and not more for each step is still kinda unique.

 

Well when I informed him about it on Friday, he told me that he was not aware of this issue.

I gave him a link to the forum on Igor's website where you have pasted your correct 6900XT 1024kB bios.
I would have sent him my 6800XT 1024kB bios but I was not near my computer to upload that but I saw your post at Igore's lab and gave him that link.
He downloaded your bios and check the difference to GPU-Z and he released the fix on the same day.

 

Oh....ok.... I thought hellm already reported it to him once he has found it out with with the 1024kB size...