Overclock.net banner
2521 - 2540 of 3437 Posts

·
Registered
Joined
·
693 Posts
I've got the same card under water, and get 300 points less than you do with the same clocks. You've got yourself a great 6900XT Red Devil card, sir.
Thx. I think that's just variance in the run. Looks to be the same as yours, except my CPU score is much higher (12 cores vs 6 cores).
 

·
Registered
Joined
·
79 Posts
Thx. I think that's just variance in the run. Looks to be the same as yours, except my CPU score is much higher (12 cores vs 6 cores).
I tried to find info online about how the timespy graphic scores change based on CPU, but with same GPU. Didn't find any comparisons, but I would presume your 5900X renders a couple of extra frames at 1440p at that high fps vs my 5600X, and that in turn could reflect positively on your Graphics score.

Edit: I realized I'm the 4th in the world in Timespy with 6900xt + 5600X combo right now, this is without XTXH and no overclock on CPU. Hmm...
 

·
PC Evangelist
Joined
·
47,664 Posts
I tried to find info online about how the timespy graphic scores change based on CPU, but with same GPU. Didn't find any comparisons, but I would presume your 5900X renders a couple of extra frames at 1440p at that high fps vs my 5600X, and that in turn could reflect positively on your Graphics score.

Edit: I realized I'm the 4th in the world in Timespy with 6900xt + 5600X combo right now, this is without XTXH and no overclock on CPU. Hmm...
Nobody combos 5600x + 6900 xt.
 

·
Registered
Joined
·
457 Posts
So I've been messing the past few days with MPT betas and general overclocking using Time Spy as a benchmark on my 6900XT Toxic XTX.
There are some obvious stuff everyone knows and then there's new things that might or might not do anything for you but you're free to try.

I'll try to cover everything in this small post for everyone interested in trying to get better benchmarking scores or better gaming performance.
For this, we're going to use More Power Tool and AMD driver tunning utility.

First of all, I assume everyone knows how to get More Power Tool set up and running but if not, here's a super fast and easy guide I've made that covers the basic install+power limit mod.


So once you have MPT set up here's the settings I've messed with that might come handy for those trying to squeeze a bit more out of the cards.

On the first menu, you can uncheck the following to disable deep sleep on the card. This trick was used before 5.1 drivers hotfixed the suttering issue some people experienced on CPU heavy scenarios where the card would go to sleep.

Again this can be hypothetically helpful during benchmarks because it prevents the card from downclocking during transitions and therefore supposedly help with frametime and framerate consistency.

2518496


Next up we have Overdrive Limits section, here you can change Memory Timing Control from 1 to 2. This enables a second timing option on the driver but it's unstable for everyone however, rumour has it that enabling this and leaving the card on Fast Timing 1 on the driver actually improves the timings further. I cannot 100% attest to this but I felt like it does indeed help.

2518500


Next up we have the Power and Voltage section. This is where the magic happens and this is where most of the gains will come from since these cards are well capped.
Most of the 6900XTs I've seen and my own, scale up to 400/410W. They can obviously pull much higher (I've had mine pull at least 480W and at 500W I get OCP shut down from my PSU which is only a 750W unit). However, to pull more than this it's either have to be a XTXH die (higher Vcore under load) or you have to manually mess with the Vcore using something like an EVC2SX as I have.

Some people also reported higher stability on their VRAM and Core by messing with min VSOC and min VCore on this page. I personally haven't noticed anything from this eventhough I run min VSOC at 0.9V.

For the limits themselves, I haven't noticed any particular scalling with them on my own card. My opinion is that, if you don't hit those limits you're going to be fine. You should only go higher if your card is throttleing. SOC limit is nowhere near the limit at stock so you can even leave it alone. The other two can be messed with because if you're pushing over 2600 Mhz you're going to be power thorttled during heavy load scenarios such as a benchmark of high resolution gaming.

2518502


Next up we have the frequency, this one has more settings exposed. I personally have messed with only 4 of these settings. I personally increased vCLKmax by 20 Mhz, dCLKmax by 20 Mhz, fCLKmin/max to 2000 Mhz and fCLKboostfreq to 2000 Mhz too. Due to some variance from run to run I cannot attest if these settings do anything positively to the scores or not but since fCLK is related to SAM and since supposedly, SAM helps with performance on benchmarks and some games, running a higher fCLK if stable should be benefitial. Again, feel free to test and report back.

2518503


I didn't personally mess with anything else inside MPT. There's a beta5 released today or something that also adds V/F curve behaviour and the hardcoded Vdroop associated with it but reports on another forum suggest it does nothing. The cards have a very droopy LLC, that's why you never see 1.175V unless you're idleing. During heavy benchmarks it will Vdroop to around 1.1 to 1.12V depending on your clockspeed and binning (better cards will mantain better voltage levels and drop less).

I want to add something else regarding how these cards scale with temperature. I already know these cards generally speaking do not care about being on water or air, as long as your hotspot/core is safe, and by safe I mean if your hotspot is below 110C, core below 100C and Memory below 100C on XTX cards and 10/20 or so degrees lower if you're on XTXH.

However, there is something to keep in mind here. These cards have a certain Vdroop related to their LLC and heat directly affects how this Vdroop relates just as how it works on AMD cpus and many other integrated circuits. So while going from air to water might not be too relevant (see how BZ is like Hall of Fame Graphic Score on air cooling) if you're not hard modding the card's Vcore then the lower the temp, hypothetically speaking the better the performance.

So we've got MPT out of the way, now on to the drivers. I won't type much more about drivers because I expect people to know the basics already. However, let me just tell you how broken the boost algorithim is and how it does not respect the limits imposed by Whattman. For this reason, messing with minimum clockspeed becomes super important. I don't have any magic numbers, however, some cards prefer a 100 Mhz difference between max and mim, others way more than this and like 200 to 400 Mhz difference.

What I would suggest for those who care is to run Hwinfo during custom time spy runs and observe clockspeed behaviour. You can easily determine what's too much and too few on min clock because at that point, effective clockspeed will drop.

Anyway I don't have anything else to add. Feel free to add stuff and your own experiences.
 

·
Professional Lurker
Joined
·
241 Posts
Happy days: as a result of the tips I got + changing flow direction as per the GPU waterblock instructions said + repasting the CPU, I've got a loop that is much much better. The 6900xt edge/junction delta is 15-20C now in TS GT2 (fresh run, probably moves to around 20-25C over time), and the max peak I saw was 67C junction, edge was at 48C (dropped from 105C junction, and 65C edge). CPU temps dropped 8C as well. I also hit a new TS record due to lower temps allowing more OC headroom (edited this post four times due to going past record scores.)

Now this is what I expected from watercooling, incredibly happy! Thank you all so much for the help!

View attachment 2518506
Yay! I am so happy for you! Hopefully you can relax a bit and start enjoying your kickass new GPU!

Here is my best TS run so far, for everyone: I scored 20 876 in Time Spy

  • Reference RX 6900 XT
  • EK Quantum Vector block
  • Gelid GC-Extreme paste
  • EK included pads
  • 420W card power limit as set in MPT; no other MPT changes
  • 16ga PCIe 8-pin power cables (buying the crimp tool for custom, homemade PSU cables was the best $40 I've ever spent)
  • 2715MHz max frequency, 500MHz min frequency (changing min freq never changed anything for me)
  • 2150MHz Memory speed; Fast timings
The settings above are what I daily drive. I don't change settings for the sake of benchmarking, I tune everything in my PC for daily driver push-button stability, and that's it.

2518519


I don't have a particularly blessed sample, but I'm not complaining, the card runs great with my LG CX 48" OLED panel.

On the first menu, you can uncheck the following to disable deep sleep on the card. This trick was used before 5.1 drivers hotfixed the suttering issue some people experienced on CPU heavy scenarios where the card would go to sleep.

Again this can be hypothetically helpful during benchmarks because it prevents the card from downclocking during transitions and therefore supposedly help with frametime and framerate consistency.
Ayyyy, it's good to see the Deep Sleep trick I discovered making the rounds. I originally posted it to Reddit (which is where I imagine it FINALLY caught the attention of the devs), then hardwareluxx dot DE, then Igor's Lab, now here! Neat!

weleh is correct though, as of 21.6.1, the stuttering that the downclocking was causing has been fixed, thank goodness.
 

·
Invalid Media
Joined
·
4,750 Posts
...have been doing some 'snooping' of H vs non-H variants, using what are supposedly identical PCB / VRMs etc by the same manufacturer. The differences in rom (69 diff) aren't earth shattering, but it would be nice to find the switches for voltage control ( >1.175) and VRAM (> 2150)...not holding my breath though; MPT 7 if/when will probably make that easier...

2518520


Also, with MPT, be aware that TDC (Amps) refers more to the physical VRM structure (including but not limited to Amp stages, for example re. the 90 Amp stages of the ASRock Formula oc vs 70 Amp stages of most other).

As to 'how hot', I think it is prudent to look at max temps shown in the vbios. Water-cooling can and will help with that, but only to a point. As others already observed, for Hotspot, you want to make sure that you have decent thermal paste (I prefer Gelid or other 'thicker' to avoid issues over time). Liquid Metal of course also works, but it is tricky re. protection of other components (conformal or LET coating) and also depends to some extent on mounting orientation.

2518521
 

·
Registered
Joined
·
457 Posts
All the cards have +1000A VRMs so it's fine.

I'm more concerned about my own PSU and that's why I stopped testing today. Was doing a 2800 Mhz run at 1.2Vcore and tripped OCP on my PSU at over 500W reported (which in reality would be much higher). Actually I had PL locked to 480W and the card pulled much more which is why it probably shut off. Had to power cycle the PSU to boot.

So yea, until I upgrade I'm not sure I'll be pushing the limits much more.
 

·
Professional Lurker
Joined
·
241 Posts
...have been doing some 'snooping' of H vs non-H variants, using what are supposedly identical PCB / VRMs etc by the same manufacturer. The differences in rom (69 diff) aren't earth shattering, but it would be nice to find the switches for voltage control ( >1.175) and VRAM (> 2150)...not holding my breath though; MPT 7 if/when will probably make that easier...
Nice work investigating the BIOS files themselves! I wish it were as easy as finding some switches, but there is a lot of work to be done for RDNA2 before we have a flash-able bios mod available.

For context, it was "so easy" to do with the RX 5600 XT because of a few things:
  • To make the bios flash-able, I believe all RedPanda did was find a way to say "Yep, we're good!" to the built in checks for bios legitimacy. If you can't crack the security, go around it, right?
  • Once that was done, to make parameters on the RX 5600 XT modifiable, the bios had to be modified in such a way that it told the drivers "Hey, I'm an RX 5700" when the driver set the power limits, but still told the driver "Treat me like an RX 5600 XT" in regards to the rest of the driver package, so that drivers didn't write to the missing 2GB of VRAM, etc.

I sadly don't think it will be as easy for RDNA2 cards in regards to circumventing both the GPU's onboard checks for bios legitimacy (& card model type), AND the driver's checks for the same.

Hellm is still working hard over on Igor's Lab, as am I with whatever I can help him with. We're still a ways off from seeing a bios mod for RDNA2 cards I think, but Hellm is back to work with MorePowerTool, and for that I am thankful!

Once someone convinces Hellm to set up a link for donations to Hellm himself, I'll post a link.

Link goes to the ongoing thread for the BETA versions of MPT and RBE on Igor's Lab
 
  • Rep+
Reactions: ptt1982 and J7SC

·
Professional Lurker
Joined
·
241 Posts
All the cards have +1000A VRMs so it's fine.

I'm more concerned about my own PSU and that's why I stopped testing today. Was doing a 2800 Mhz run at 1.2Vcore and tripped OCP on my PSU at over 500W reported (which in reality would be much higher). Actually I had PL locked to 480W and the card pulled much more which is why it probably shut off. Had to power cycle the PSU to boot.

So yea, until I upgrade I'm not sure I'll be pushing the limits much more.
That is a SPICY test! I love the Seasonic PRIME series, myself; especially the Titanium model. Over-engineering at its finest 💗
 

·
Registered
Joined
·
457 Posts
You can flash XTX or XTXH vbios. I've done it via Linux and it sort of works.

There's something else on the card preventing it, maybe there's actually hardware changes since die stepping is different and supposedly theres an improved clock gen as well.
 

·
Invalid Media
Joined
·
4,750 Posts
Nice work investigating the BIOS files themselves! I wish it were as easy as finding some switches, but there is a lot of work to be done for RDNA2 before we have a flash-able bios mod available.

For context, it was "so easy" to do with the RX 5600 XT because of a few things:
  • To make the bios flash-able, I believe all RedPanda did was find a way to say "Yep, we're good!" to the built in checks for bios legitimacy. If you can't crack the security, go around it, right?
  • Once that was done, to make parameters on the RX 5600 XT modifiable, the bios had to be modified in such a way that it told the drivers "Hey, I'm an RX 5700" when the driver set the power limits, but still told the driver "Treat me like an RX 5600 XT" in regards to the rest of the driver package, so that drivers didn't write to the missing 2GB of VRAM, etc.

I sadly don't think it will be as easy for RDNA2 cards in regards to circumventing both the GPU's onboard checks for bios legitimacy (& card model type), AND the driver's checks for the same.

Hellm is still working hard over on Igor's Lab, as am I with whatever I can help him with. We're still a ways off from bios mods for RDNA2 cards, but Hellm is back to work with MorePowerTool, and for that I am thankful!

Link goes to the ongoing thread for the BETA versions of MPT and RBE on Igor's Lab
...good to hear about the ongoing MPT work ! And yeah, it is far more tricky than just throwing a few switches - that said, I chose the Gigabyte 3 pin and corresponding Aorus 3 pin as they apparently use the same PCB, so it limits the number of potential suspects in the room, so to speak.

...I did manage to load the Aorus 3x8 bios on my Gigabyte 3x8, and it did 'show' 1.2V but there are plenty of other checks and balances which spoil the party :mad:
 

·
Registered
Joined
·
162 Posts
My air cooled 6900 xt reaches 115 degrees hotspot at 410 watts and shuts down during timespy to protect itself is this normal?
 

·
Registered
Joined
·
729 Posts
My air cooled 6900 xt reaches 115 degrees hotspot at 410 watts and shuts down during timespy to protect itself is this normal?
Yep, from memory i think shutdown temp is 118c for MBA.

The stock cooler is great for stock operation, quietness and a small overclock. But for extreme overclocks it gets overwhelmed once you go past 325-350W.
 

·
Registered
Joined
·
162 Posts
Yep, from memory i think shutdown temp is 118c for MBA.

The stock cooler is great for stop operation, lack of noise and a small overclock. But for extreme overclocks it gets overwhelmed once you go past 325-350W.
I think an air cooled card shouldn’t reach 115 degrees on 400 watts though. I don’t understand whats wrong its not the airflow of the case I know that for sure.
 

·
Registered
Joined
·
133 Posts
I think an air cooled card shouldn’t reach 115 degrees on 400 watts though. I don’t understand whats wrong its not the airflow of the case I know that for sure.
It will. I have an MSI 6900XT Gaming X and suffer from the same problem. Very high hotspot temp after I use MPT to increase power limits.
Unfortunately I seem to have one of the few cards that have manufacturers have no plans to make a waterblock for :(
 

·
Registered
Joined
·
162 Posts
It will. I have an MSI 6900XT Gaming X and suffer from the same problem. Very high hotspot temp after I use MPT to increase power limits.
Unfortunately I seem to have one of the few cards that have manufacturers have no plans to make a waterblock for :(
My card has a waterblock but I need to get it from aliexpress which bothers me, also building a custom loop is really expensive in total. I am not really sure but I think I am going to build a loop. This is the first time I had a GPU shutdown automatically because of temperatures :D
 

·
Registered
Joined
·
133 Posts
My card has a waterblock but I need to get it from aliexpress which bothers me, also building a custom loop is really expensive in total. I am not really sure but I think I am going to build a loop. This is the first time I had a GPU shutdown automatically because of temperatures :D
Also a first for me. I have found the limit now, if I push max frequency in wattman 30 MHz it will shut down. So for daily use I have set it 100 MHz lower just to be safe :)
 

·
Registered
Joined
·
457 Posts
Well it's just bad air cooler really.

This is my old Nitro+SE 6800XT doing 410W at 86C hotspot.
This is at 20C ambient.

Sure the die is the same with less CU's but thermal shutdown on a 6900XT at 400W shouldn't really happen.
Look at BZ's 6900XT, he's pulling way over 500W on an air cooler nowhere near shutdown hotspot.
Sure his ambient might be low like 17-18C or something but still.

2518609
 

·
Registered
Joined
·
133 Posts
Well it's just bad air cooler really.

This is my old Nitro+SE 6800XT doing 410W at 86C hotspot.
This is at 20C ambient.

Sure the die is the same with less CU's but thermal shutdown on a 6900XT at 400W shouldn't really happen.
Look at BZ's 6900XT, he's pulling way over 500W on an air cooler nowhere near shutdown hotspot.
Sure his ambient might be low like 17-18C or something but still.
His card is very modified, so not really a fair comparison to a stock card.

I think thermals would be improved with a re-paste, but I don't have the balls to do that (yet)
 
2521 - 2540 of 3437 Posts
Top