The RX 6000 series Owners' Club

[Kodehawa]

So RDNA 3 owners will now have to OC through the driver instead of downloading an extra program that does the same? Poor souls!

I guess it's mostly the lack of FCLK overclocking and max PPT modifications (and everything else that can be modified there but not on driver). I wouldn't mind it much if it wasn't there, but since it exists...

 

[RJARRRPCGP]

SCLK is GPU core. Classic symptoms of unstable SCLK are:

1: Hard-lock crashes.

2: Squares popping up on the screen.

3: Random driver-timeout error message from AMD's software.

4: Error message from 3D program about failing to initialize the graphics hardware.

5: Error message from 3D program about failing to initialize the graphics hardware, even when it was already initialized.

 

[Kodehawa]

SCLK is GPU core.

Fixed, oversight on my thought process lol

 

[3x0]

So RDNA 3 owners will now have to OC through the driver instead of downloading an extra program that does the same? Poor souls!

MPT has alot more options than the driver

 

[haps]

MPT has alot more options than the driver

And it dosent change voltage curve like AMD drivers do. No other way for proper undervolting.

 

[tabascosauz]

Now, who has more video OC'ing restrictions? Nvidia or AMD?

I don't see anything out of the ordinary here. What's all the fuss about? Isn't this what RDNA 3 owners see?

No access to MPT for RDNA3 hurts a lot. Aside from being limited to the length of the sliders in Wattman, you don't get the finer fan control tuning in MPT (fan stop temperature limits and the target temp/clocks for fan curve). The AIB cards are fine but the reference cards' fan control is pretty F'ed up, MPT could've helped a great deal.

iirc RDNA2 also had access to stuff like editing VRAM P-states, which I imagine would really help with the aggressive VRAM downclocking on RDNA3. And editing the V-F curve, VSOC, etc. I don't think any of that functionality was actually removed in the RDNA3 design (which functions fundamentally the same), just locked away. Undervolting via Wattman is really weird and abstract.

I originally was patiently waiting for the MPT creators to bring out an RDNA3 update, until I saw the articles about how AMD locked everything down and it'll either be impossible or a long time before RDNA3 can be tweaked via powerplay tables. Not a good look for a GPU family that already doesn't have much going for it over its predecessor.

 

[mb194dc]

The 5600 XT was a carefully tailored card to a price point. One to take care of the 1660 Super/Ti and later to challenge the 2060, but not eat into the 5700's market share. Letting us freely overclock it would have decimated the 5700's sales as it has the identical core structure. An overclockable 5600 XT would been especially competitive in non-memory bandwidth-restricted games.

From a business perspective I appreciate what they had to do at $280 versus $350 for the 5700. Though from the Pulse 5600 XT owner's perspective of course it's annoying.

I looked into jailbreaking it but I didn't want to potentially fail during the GPU shortage, though more recently I don't depend on it so that's tempting now…

Yes my unlocked 5600xt red dragon does 2,1ghz on the core and performs in-between stock 5700 and 5700xt...

I think the unlock bios leaked from MSI is for any 5600xt with Micron ram, could be wrong though.

I also have a 6800xt, no reason to do anything fancy with it. Just tweaks it Wattman.

AMD probably restrict rdna 3 to stop people bricking and RMA cards. The unlocked 5600xt bios tricks the drivers in to thinking you've got a 5700xt I think. Maybe something similar will come rdna 3, eventually.

.

 

[Kodehawa]

Undervolting via Wattman is really weird and abstract.

idk why they regressed in this, on my RX 480 you could edit the entire undervolt curve in Wattman, per P-state.

 

[RJARRRPCGP]

Yes my unlocked 5600xt red dragon does 2,1ghz on the core and performs in-between stock 5700 and 5700xt...

I think the unlock bios leaked from MSI is for any 5600xt with Micron ram, could be wrong though.

I also have a 6800xt, no reason to do anything fancy with it. Just tweaks it Wattman.

AMD probably restrict rdna 3 to stop people bricking and RMA cards. The unlocked 5600xt bios tricks the drivers in to thinking you've got a 5700xt I think. Maybe something similar will come rdna 3, eventually.

.

IIRC, it tricks the drivers into thinking it's a 5700 non-XT (a.k.a. "vanilla 5700") And with my RX 5600 XT, 2.1 GHz was an insta-crash! You have a better sample than I did, it looks like!

*/leaves RX 5600 XT in rear-view mirror*

 

[Thimblewad]

When you load an overclock from a file, they added the option to choose if you want to apply it on your CPU+GPU or GPU only which is VERY NICE.

 

[Deleted member 185158]

I scored 12 026 in Time Spy

AMD Ryzen 7 5700X, AMD Radeon RX 6700 x 1, 32768 MB, 64-bit Windows 11}

www.3dmark.com

Managed to hit 12k on the 6700 :tada:

Don't think I'll be trying any further, as overclocking can be quite a pain, but I'm pretty happy with it.

RX 6700? I just move the OC sliders all the way to max 2800.

You'd want to try and change the "Minimum frequency" slider up closer to your desired OC.
From some experimenting, just leaving the card defaults, after a period of time (short), the average clocks are roughly 2400mhz, might as well left it stock.
My stock score was 11700pnts in TimeSpy.
Easy to OC, sliders to max power limit max 15%, the scores don't seem to do a whole lot of changing.

Here's 12,222 marks Time Spy. Easy peasy, almost hit it at stock.

I scored 12 222 in Time Spy

Intel Core i7-13700K Processor, AMD Radeon RX 6700 x 1, 16384 MB, 64-bit Windows 11}

www.3dmark.com

 

[Kodehawa]

RX 6700? I just move the OC sliders all the way to max 2800.

That's instant crash for me lol

I can run Time Spy at 2740, RDR2 crashes at 2720, so I gotta run 2700 or 2670 for normal gaming. I've heard a bunch of max out slider but doesn't work for me . Can do 2150MHz memory and performance increases from 2100, so at least there's that. Also your graphic score is lower, CPU score carries that score a lot. To be honest, you end up very power limited and I messed with MPT for TS.

I get about 2610MHz in game with PL +15%

Still a very nice and underrated card.

 

[RJARRRPCGP]

At least for Adrenalin 23.2.1, the built-in stress test is broken! It will keep dying and makes the video go to idle-state! Can't be more obvious, as the VRAM frequency is at 0 MHz, too, IIRC!

 

[tabascosauz]

At least for Adrenalin 23.2.1, the built-in stress test is broken! It will keep dying and makes the video go to idle-state! Can't be more obvious, as the VRAM frequency is at 0 MHz, too, IIRC!

I'm pretty sure the built-in stress test function is just perpetually garbage. On 23.1.1 and 23.1.2 it put on an okay load but only lasts for 60 seconds regardless of what it's set to, just idle the remainder of the test

Sniper Elite 5 was pretty demanding so I just used that as a stress test

 

[RJARRRPCGP]

I'm pretty sure the built-in stress test function is just perpetually garbage. On 23.1.1 and 23.1.2 it put on an okay load but only lasts for 60 seconds regardless of what it's set to, just idle the remainder of the test

Doesn't even last 30 seconds for me, LOL! I can only confirm the drivers boosting properly, when using RT in Automation, so far. Also, it boosts correctly in Gearbox Halo CE.

 

[AusWolf]

At least for Adrenalin 23.2.1, the built-in stress test is broken! It will keep dying and makes the video go to idle-state! Can't be more obvious, as the VRAM frequency is at 0 MHz, too, IIRC!

I've never liked the built-in stress test. It never seemed to put enough load on the GPU for a proper test. I just use a 3DMark Stress Test instead, or an infinite Superposition loop.

 

[3x0]

Doesn't even last 30 seconds for me, LOL!

Heh, just like me

 

[Deleted member 185158]

That's instant crash for me lol

I can run Time Spy at 2740, RDR2 crashes at 2720, so I gotta run 2700 or 2670 for normal gaming. I've heard a bunch of max out slider but doesn't work for me . Can do 2150MHz memory and performance increases from 2100, so at least there's that. Also your graphic score is lower, CPU score carries that score a lot. To be honest, you end up very power limited and I messed with MPT for TS.

I get about 2610MHz in game with PL +15%

Still a very nice and underrated card.

Looks like yours boosts better on timespy than mine does. By a decent 100mhz. It's so weird.

I suppose I should try the wattman software stuff, but it seems to give me issues honestly. I was using Afterburner instead.

If you'd like to compare some more, tag me. In the meantime, I'll work on some other benchmark scores. I've only run a few, the card is pretty new yet.

 

[Kodehawa]

Looks like yours boosts better on timespy than mine does. By a decent 100mhz. It's so weird.

I suppose I should try the wattman software stuff, but it seems to give me issues honestly. I was using Afterburner instead.

If you'd like to compare some more, tag me. In the meantime, I'll work on some other benchmark scores. I've only run a few, the card is pretty new yet.

Increase your power limit. This card is INCREDIBLY starved for power at default settings.

If your cooler can handle it, that being, XFX decided slapping a 6700XT cooler on the 6700 was the way to go, so I can cool it fine.

 

[Deleted member 185158]

Increase your power limit. This card is INCREDIBLY starved for power at default settings.

If your cooler can handle it, that being, XFX decided slapping a 6700XT cooler on the 6700 was the way to go, so I can cool it fine.

I did. +15%. I think it's more an installation problem actually. I have a lot of legacy benchmarks which seem to clash with newer benchmarks.

CloudGate was boosting well though. Looked like it was hitting 2740mhz during the benchmark run.

I scored 80 666 in Cloud Gate

Intel Core i7-13700K Processor, AMD Radeon RX 6700 x 1, 16384 MB, 64-bit Windows 11}

www.3dmark.com

I'd like to add that I'm not sure 3DMark calculates the average frequency well either.

Was using TechPowerUp Gpu-Z sensors open during the benchmark run.
Hovering the mouse over the clock speed, I can go back and see where the GPU was during the run.
That's where I got the 2740mhz, which was actually 2741mhz. It was sustained during the run while 3DMark claims average of 2500mhz and some change.

 

[Kodehawa]

I did. +15%. I think it's more an installation problem actually. I have a lot of legacy benchmarks which seem to clash with newer benchmarks.

CloudGate was boosting well though. Looked like it was hitting 2740mhz during the benchmark run.

I scored 80 666 in Cloud Gate

Intel Core i7-13700K Processor, AMD Radeon RX 6700 x 1, 16384 MB, 64-bit Windows 11}

www.3dmark.com

I'd like to add that I'm not sure 3DMark calculates the average frequency well either.

Was using TechPowerUp Gpu-Z sensors open during the benchmark run.
Hovering the mouse over the clock speed, I can go back and see where the GPU was during the run.
That's where I got the 2740mhz, which was actually 2741mhz. It was sustained during the run while 3DMark claims average of 2500mhz and some change.

Time Spy is just very hard on RDNA2 power usage, so its power limited as heck.

Make sure to test your OC on games. I can pass Time Spy infinitely at 2740, but RDR2 crashes after 10 or 20 minutes.

 

[Deleted member 185158]

Time Spy is just very hard on RDNA2 power usage, so its power limited as heck.

Make sure to test your OC on games. I can pass Time Spy infinitely at 2740, but RDR2 crashes after 10 or 20 minutes.

I don't game..... Strictly competitive benching.

Thinking the card is hard locked PPT.

 

[Veseleil]

Hello. As I don't have enough time to read trough the entire thread ATM, I'd like to ask a simple question, if you don't mind.

Which would be the fastest and/or the most reliable way to check a 6800xt, 6900xt for a possible failure/instability?
Consider a card that is used for cryptomining, that would be used primarily for gaming, probably undervolted and underclocked. Thanks

 

[Kodehawa]

I don't game..... Strictly competitive benching.

Thinking the card is hard locked PPT.

No. I ran 195W PPT and it drew 195W. Got 12111 graphics on 3DMark TS once.

What's your model?

 

[3x0]

Hello. As I don't have enough time to read trough the entire thread ATM, I'd like to ask a simple question, if you don't mind.

Which would be the fastest and/or the most reliable way to check a 6800xt, 6900xt for a possible failure/instability?
Consider a card that is used for cryptomining, that would be used primarily for gaming, probably undervolted and underclocked. Thanks

Unigine Heaven and this tool for VRAM testing: https://github.com/GpuZelenograd/memtest_vulkan
At least 30 min each