Radeon RX 6700, 6700 XT & 6750 XT users club

[Kyan]

Can anyone tell me their unigine superposition score with Rx 6750xt ?I scored 7284 with average of 65 FPS.Is that normal?Im worried

Here's my result in 1080p extreme on a RX6750XT Reference Design with those custom settings.



And here are the result with stock settings, but it run at 200W instead of 150W

 

[AusWolf]

Here's my result in 1080p extreme on a RX6750XT Reference Design with those custom settings.

View attachment 309961View attachment 309962

And here are the result with stock settings, but it run at 200W instead of 150W

View attachment 309964

Something doesn't seem right there. I get around 7200 without changing a single setting, just like @cheermate.

 

[Lew Zealand]

Here's my result in 1080p extreme on a RX6750XT Reference Design with those custom settings.

View attachment 309961View attachment 309962

And here are the result with stock settings, but it run at 200W instead of 150W

View attachment 309964

Try OCing the cores and not the memory at 200W, see if your score goes up. Your memory is max OC'd and could be getting into error-correction which will reduce overall speed.

 

[Noci]

Something doesn't seem right there. I get around 7200 without changing a single setting, just like @cheermate.

Bon soir Kyan,
AusWolf & Lew Zealand are right, my 6700 XT is doing even better on 1080p extreme (but I don't have to worry about temps).
Your card should be able to beat this, however I'm not sure how much impact the CPU has.

 

[RJARRRPCGP]

I'm new to this thread and browising through these 22 pages I thought; why the hell, lets give this benchmark thing a go on my XFX Speedster QICK 319 RX 6700XT Black a go.

Didn't play around much with the Radeon software yet, but never too late to start. From previous attemts, I noticed the VRAM didn't like user interferrence at all, so I've played around with the sliders a bit except for the VRAM. (sorry the language is in Ducth, but the layout is the same)
View attachment 309239

When I slide them full to the right, 3DMark Time Spy crashes and no results, except for a message: an error occured. But with the settings in the pic above it pulls through. PPT limit for this setting is 242.650 W, but it peaked at 221.607 W, GPU Core Voltage peaked at 1.120 V and the current peaked at 12.676 A. Frequency peaked @ 2,671 MHz, that's below the 2,753 of the slider.
Temps, GPU peak 43 C, hotspot 74 C peak, nothing to worry about, got a full cover waterblock.

Time Spy Score with this, see below:
View attachment 309240

Any suggestions how to spice these scores up? For the CPU, it's no problem atm it is undervolted and not running on steriods .

By the way, I noticed somewhere in this thread somebody mentioned his card had Micron IC's for VRAM, GPU-Z tells me I have Samsung. Does this mayby explain why I cannot tickle my VRAM?

View attachment 309243

And before commenting, I know I'm runnig @ PCIe x16 3.0 instead of PCIe x16 4.0, it's defenitly due to the riser cable that came with the Thermaltake Core P90. Have to look for a replacement, but the A-brands are still f*cking expensive and I'm reluctant to try an AliBaba one, unless somebody can advise a really good one with a reasonable pricetag.

Additional the Superposition BM with the settings above v@ 1080P extreme

View attachment 309253

Which sliders when 3D Mark failed? Was that for core or VRAM? Is that what you mean when you talked about having the sliders full to the right? Was it the core minimum, core maximum and GPU Vcore?
Yeah, for unstable core, there's often an error about failing to initialize Direct3D, even though it was initialized already.

 

[Noci]

Which sliders when 3D Mark failed? Was that for core or VRAM? Is that what you mean when you talked about having the sliders full to the right? Was it the core minimum, core maximum and GPU Vcore?
Yeah, for unstable core, there's often an error about failing to initialize Direct3D, even though it was initialized already.

I was impatient and slid them all to max .

But this afternoon I've been tweaking, max on my GPU frequency is now 2850 MHz and the VRAM 2100 MHz. VRAM also on quick timings and the power limit + 15%

Forgot, Voltage also to the max 1200 mV

 

[RJARRRPCGP]

I was impatient and slid them all to max .

But this afternoon I've been tweaking, max on my GPU frequency is now 2850 MHz and the VRAM 2100 MHz. VRAM also on quick timings and the power limit + 15%

Forgot, Voltage also to the max 1200 mV

LOL AMD, why does it look like my RX 6750 XT is a recycled 6700 XT?! At least when GPU-Z is up, it looks like that's the case! I suspect that, because GPU-Z says "C0" for mine!

I just hope that it's not also a fake! (in one form or another)

 

[AusWolf]

Bon soir Kyan,
AusWolf & Lew Zealand are right, my 6700 XT is doing even better on 1080p extreme (but I don't have to worry about temps).
Your card should be able to beat this, however I'm not sure how much impact the CPU has.


View attachment 309978

The CPU has practically zero impact on your Superposition score.

 

[cheermate]

Bit of a silly question but would it be okay to leave the protective sticker/films on the middle of the GPU fans?Forgot to take them off and i don't really want to unscrew the GPU again

 

[AusWolf]

Bit of a silly question but would it be okay to leave the protective sticker/films on the middle of the GPU fans?Forgot to take them off and i don't really want to unscrew the GPU again

Yeah, should be fine (unless it comes off on its own and gets stuck in something, which has a minimal chance of happening).

 

[Kyan]

Try OCing the cores and not the memory at 200W, see if your score goes up. Your memory is max OC'd and could be getting into error-correction which will reduce overall speed.

I really want to stay away of 200W, that's why i've underclock and undervolt it, i don't need that extra power for my usage, maybe later or if i feel like it during the winter.
One problem i have is that at 1125mV instead of 1150mV at 2500MHz, it seems to crash randomly in game. I don't know if it's related but i don't have problem with those settings so it may be that.

So you are advising me to don't OC memory if i don't overclock ? Never heard about the error-correction, how much does it impact performance ?

Something doesn't seem right there. I get around 7200 without changing a single setting, just like @cheermate.

I don't really know what i should change, maybe disabling SAM ? I remember seeing that it will not always improve performance with Zen 2.

 

[AusWolf]

I don't really know what i should change, maybe disabling SAM ? I remember seeing that it will not always improve performance with Zen 2.

Good question. Like I said, I've got everything at bone stock - no OC, or any other setting changed whatsoever (SAM enabled).

Edit: Zen 2 shouldn't really limit you in Superposition. It is not a CPU-bound test at all.

 

[Kyan]

Bon soir Kyan,
AusWolf & Lew Zealand are right, my 6700 XT is doing even better on 1080p extreme (but I don't have to worry about temps).
Your card should be able to beat this, however I'm not sure how much impact the CPU has.


View attachment 309978

When i was doing research before buying the 6750XT, i remember that my R5 3600 is bottlenecking but less then 3% if i remember correctly

Good question. Like I said, I've got everything at bone stock - no OC, or any other setting changed whatsoever (SAM enabled).

Edit: Zen 2 shouldn't really limit you in Superposition. It is not a CPU-bound test at all.

Yeah, when i was running Superposition, the CPU was around 10% usage.

 

[cheermate]

Yeah, should be fine (unless it comes off on its own and gets stuck in something, which has a minimal chance of happening).

Thanks for the help.I was worrying about this quite a bit.

 

[AusWolf]

Thanks for the help.I was worrying about this quite a bit.

No worries. I tend to like preserving my hardware as much in its original state as possible, including not removing non-essential film.

 

[Macro Device]

Any suggestions how to spice these scores up?

Set your VRAM speed exactamente at 2120 MHz. 99% chance it will run just perfectly fine, go stock otherwise.
Add ~15 MHz on the core step by step till it stops running stable. 2753 to 2768; test; 2768 to 2783; test; et cetera, et cetera. Don't expect this GPU running OK beyond 2800 though, 6700 XTs are way worse in clock speeds than 6750 XTs.
Disabling SAM vastly improves VRAM overclocking stability but may be pointless because some games run way worse with SAM disabled.
You might also try MPTing your GPU but it makes minimal or even zero sense in GPUs below RX 6800.

Frankly, there's not much you can do and you're already running at very high speed. Best advice is forgetting about this and using the GPU as it is.

 

[Kyan]

LOL AMD, why does it look like my RX 6750 XT is a recycled 6700 XT?! At least when GPU-Z is up, it looks like that's the case! I suspect that, because GPU-Z says "C0" for mine!

I just hope that it's not also a fake! (in one form or another)

I really think the 6750XT is just a 6700XT which have win the silicon lotery. It make it able to handle higher clock speed and amd have charge more money for it. I'm not an expert it's just what i think.
Edit: I'm probably wrong, but it feels like it.

 

[cheermate]

Quick question for Asus Rx 6750xt users.Does hwinfo show only one GPU fan speed for you as well?Used to have another card and hwinfo listed both speeds of the fans.

 

[RJARRRPCGP]

I really want to stay away of 200W, that's why i've underclock and undervolt it, i don't need that extra power for my usage, maybe later or if i feel like it during the winter.
One problem i have is that at 1125mV instead of 1150mV at 2500MHz, it seems to crash randomly in game. I don't know if it's related but i don't have problem with those settings so it may be that.


I don't really know what i should change, maybe disabling SAM ?

Not recommended to disable SAM for your goals.

 

[AusWolf]

I really think the 6750XT is just a 6700XT which have win the silicon lotery. It make it able to handle higher clock speed and amd have charge more money for it. I'm not an expert it's just what i think.
Edit: I'm probably wrong, but it feels like it.

It's not your feeling. It's fact.

Quick question for Asus Rx 6750xt users.Does hwinfo show only one GPU fan speed for you as well?Used to have another card and hwinfo listed both speeds of the fans.

The reference card shows only one fan speed as well (all fans spin at the same speed).

 

[Macro Device]

Clocks&Watts chart (6700 XT).

Underlined italics is sweet spot energy wise (further downclocking ruins speed more than it saves electricity).
And bold is sweet spot performance wise (further overclocking devastates energy efficiency).

Tested in Cyberpunk 2077 at 1080p Ultra during absolutely no action whatsoever in a random location on the map.
Voltage: -65 mV offset.
VRAM: 7% OC (less than +1 W).

2745—2845: unstable.
2645—2745: 170 W. 95 FPS.
[stock speed] 2545—2645: 151 W. 94 FPS.
2445—2545: 127 W. 94 FPS.
2395—2495: must be 120 W and perfectly balanced, not tested yet. 93 FPS?
2345—2445: 115 W. 93 FPS.
2245—2345: 106 W. 91 FPS.
2145—2245: 98 W. 89 FPS.
2045—2145: 95 W. 87 FPS.
1945—2045: 93 W. 85 FPS.

Basically 115 W is 4060's official TDP, yet the only way 4060 can be competitive is RT On + Frame Generation. 127 W is less than 4060 Ti's official TDP (by 33 W), yet 4060 Ti costs about a half fortune more than 6700 XT. RDNA2 is worse than Ada but as you can see, not by too much.

This 127 W setting, paired with a semi-custom cooler on my card, provides me with low 70s in hotspot temperature with no possibility to hear my GPU fans since CPU fans (950 to 1050 RPM 120 mm ones) are louder. 6700 XT is a killer 1080p GPU after all xDDDDDDDD

 

[Lew Zealand]

Clocks&Watts chart (6700 XT).

Underlined italics is sweet spot energy wise (further downclocking ruins speed more than it saves electricity).
And bold is sweet spot performance wise (further overclocking devastates energy efficiency).

Tested in Cyberpunk 2077 at 1080p Ultra during absolutely no action whatsoever in a random location on the map.
Voltage: -65 mV offset.
VRAM: 7% OC (less than +1 W).

2745—2845: unstable.
2645—2745: 170 W. 95 FPS.
[stock speed] 2545—2645: 151 W. 94 FPS.
2445—2545: 127 W. 94 FPS.
2395—2495: must be 120 W and perfectly balanced, not tested yet. 93 FPS?
2345—2445: 115 W. 93 FPS.
2245—2345: 106 W. 91 FPS.
2145—2245: 98 W. 89 FPS.
2045—2145: 95 W. 87 FPS.
1945—2045: 93 W. 85 FPS.

Basically 115 W is 4060's official TDP, yet the only way 4060 can be competitive is RT On + Frame Generation. 127 W is less than 4060 Ti's official TDP (by 33 W), yet 4060 Ti costs about a half fortune more than 6700 XT. RDNA2 is worse than Ada but as you can see, not by too much.

This 127 W setting, paired with a semi-custom cooler on my card, provides me with low 70s in hotspot temperature with no possibility to hear my GPU fans since CPU fans (950 to 1050 RPM 120 mm ones) are louder. 6700 XT is a killer 1080p GPU after all xDDDDDDDD

I should test in CP2077 but it shows how much better a sample 6700 XT you have than I do. Tested looping Time Spy's second GPU test at 1080p I max out at:

2550 MHz (reported) and above: random crashes in the first part of the scene as it pans along the bottom row.
2540 MHz: 214W pegged and frequently clocks down to maintain 214W
2400 MHz: 170-180W
2100 MHz: 140-150W

This is with +15% (Max) power and voltage at -0.075mV (1125mV as listed in Adrenaline), if I raise that a couple bins at 2550 to attempt stability, it's already pegged at max Wattage so there are no gains to be had as it clocks down anyway. In a very low demand game, maybe I'd see 2600 but there's no point as the FPS is high enough already.

That's from memory, only the 2400 MHz number may be off a bit (could be 2350MHz) but the extremes are correct. Below 2100 MHz you're at this card's minimum load voltage of 0.881V (though it can drop to 0.856 when not loaded 100%), so there's no reason to test lower than that as it'll be mostly linear scaling. The GPU's minimum wattage seems to be ~65W in Rocket League or any low demand 3D game, and it's rummaging at 800 MHz at 1440p/144 fps locked.

At 2445-2545 MHz you're already beating my minimum 100% load wattage at 2100 MHz. I wonder if your sample runs lower voltage at all speeds and maybe even can run a lower minimum voltage (like 0.76v or similar). My 6800 XT also doesn't like lower than about 0.856V but my 6600 XT is happy with about 0.762v minimum. I dunno if this is an individual card thing or a GPU chip series (Navi 21, 22, 23) thing or both.

I only have one card duplicate, Pulse 5600 XT and Pulse BE 5600 XT and one runs at 0.956v stably and the other at the same speeds runs at 0.9v stably and I haven't tried to lower it from there yet. So there seems to be considerable run to run variation.

BUT! And this is a LOL. They use almost the same power even though those voltages are different enough to see some good power savings in the 0.9v one so is this all just bad voltage reporting? I just dunno and haven't compared the two right next to each other yet to see.

 

[Macro Device]

it shows how much better a sample 6700 XT you have than I do.

It doesn't.

Tested looping Time Spy's second GPU test at 1080p I max out at:

Because this test wants way more watts than the Cyberpunk game. In Time Spy I get ~215 W on the same settings that get me ~175 W in Cyberpunk at 1080p Ultra.

I wonder if your sample runs lower voltage at all speeds

I highly doubt that. -65 mV is the best UV I could humanly get out of it, yet it only works fine if I don't enable RT. For RT scenarios, I make it -45 mV.

Let's test our GPUs in TimeSpy at these settings and see who's who? Disregard fan tuning, my settings might be too little/too much for your cooling solution.

Will post in a while from now, I have to re-download 3DMark suite.

 

[HD64G]

Let me offer my summer testing results testing profiles for efficiency (the core voltage in catalyst differ much from the ones applied in practice):

RX 6750 XT Catalyst profiles	Max clock (MHz)	Max voltage	Core Temp	Hotspot Temp (27C room temp)	Time Spy GPU score	Score analogy vs stock profile	Max system power from wall (W)	max GPU power draw
2GHz@1V	1995​	0,825​	61​	68​	10968​	81,5​	231​	113,4​
2,3GHz@1,2V	2285​	0,95​	65​	80​	12264​	91,1​	266​	144,9​
2,6GHz@1,2V & VRAM fast	2578​	1,075​	68​	93​	13465​	100​	308​	182,7​
UV@1,175V	2632​	1,162​	69​	93​	13538​	100,6​	323​	196,2​
Stock (2,7GHz@1,2V)	2666​	1,2​	69​	93​	13461​	100​	327​	199,8​

 

[Macro Device]

@Lew Zealand

Sorry for bad English, it's my 3rd lannich...