Radeon RX 6700, 6700 XT & 6750 XT users club

[Nordic]

https://www.3dmark.com/3dm/106787299?
ok It's 267 watts peak power use. And a new score.

That makes more sense. Your getting a little more power out of your device. Mine won't use that much power weather or not I give it more or less power. I think I am temperature limited. Your cooling 15c cooler despite using more watts. When i get my PTM7950 I should be able to push a bit further but still not as far as you. Better heat transfer won't make up for your heatsink being significantly bigger.

 

[amd64skater]

That makes more sense. Your getting a little more power out of your device. Mine won't use that much power weather or not I give it more or less power. I think I am temperature limited. Your cooling 15c cooler despite using more watts. When i get my PTM7950 I should be able to push a bit further but still not as far as you. Better heat transfer won't make up for your heatsink being significantly bigger.

Ok I'm done for now going to let it cool down for now. But not bad with undervolting and a bit of overclock on the cpu. Thank you @Nordic @Beginner Micro Device @Lew Zealand for the info and help.

3dmark.com

www.3dmark.com

 

[Macro Device]

Thank you @Nordic @Beginner Micro Device @Lew Zealand for the info and help.

And it's almost 5:30 am where I live and I didn't sleep this night. Can't understand what you are thanking me for but you're welcome!

 

[amd64skater]

And it's almost 5:30 am where I live and I didn't sleep this night. Can't understand what you are thanking me for but you're welcome!

You know now looking back I was glancing at the past posts didn't keep track smh. But you have helped in the past so thanks again lol. Tired as well 9:45pm here just started my weekend. Now on that last pass I think my card was hitting it's threshold because my results are the best yet incomplete. I will have to rerun again tomorrow.

That makes more sense. Your getting a little more power out of your device. Mine won't use that much power weather or not I give it more or less power. I think I am temperature limited. Your cooling 15c cooler despite using more watts. When i get my PTM7950 I should be able to push a bit further but still not as far as you. Better heat transfer won't make up for your heatsink being significantly bigger.

I forgot which 6750xt you have. Also yes pretty impressive on the stock cooler with a bit of undervolting made that much of a difference in temperature. I never go past 70 on the gpu.

 

[Nordic]

I forgot which 6750xt you have. Also yes pretty impressive on the stock cooler with a bit of undervolting made that much of a difference in temperature. I never go past 70 on the gpu.

I have the MSI RX 6750 XT MECH 2X OC which is a low end cooler. Without overclocking it stays pretty quiet but it is certainly not an amazing cooler like the Xfx merc.

 

[amd64skater]

I have the MSI RX 6750 XT MECH 2X OC which is a low end cooler. Without overclocking it stays pretty quiet but it is certainly not an amazing cooler like the Xfx merc.

Yes they did pretty good with this card on the cooler and it's a huge dust magnet have to use compressed air once a month. Now for settings min was 2800 max 2925 mv 1180 then max everything else. Also 267 watts ok I'm guessing that is good or ok? With that being said I kind of what to see how far I can go or should I just leave it and call it a day?

 

[Nordic]

Yes they did pretty good with this card on the cooler and it's a huge dust magnet have to use compressed air once a month. Now for settings min was 2800 max 2925 mv 1180 then max everything else. Also 267 watts ok I'm guessing that is good or ok? With that being said I kind of what to see how far I can go or should I just leave it and call it a day?

"good is a relative term here. The more power you use the less efficient your gpu is.
My GPU is most efficient at 2200mhz as measured by time spy score divided by power used.
To get the maximum performance you need more power used. Poor Beginner Micro Devices can't even get more than 215w out of his.

 

[amd64skater]

"good is a relative term here. The more power you use the less efficient your gpu is.
My GPU is most efficient at 2200mhz as measured by time spy score divided by power used.
To get the maximum performance you need more power used. Poor Beginner Micro Devices can't even get more than 215w out of his.

Oh not sure I set mine at 2700mhz 1145mv the rest is the same. I play MW and most games around those settings power 175 watt avg. Temps no higher than 51 degrees. Now before I did this whole undervolting adventure my temps were around 65 or so with no tuning just plug and play. Only thing I did mess with was the fan profile.

 

[Lew Zealand]

How much does your 7700 XT get in 3DMark at full OC?

Sapphire Pulse 7700 XT
Cores minimum 2800 MHz, maximum 3400 MHz
2700-2800 actual core speed, dependent on load
Memory 2414 MHz, actual 2401
+15% Power (263W total board power, not just GPU Chip)
1000 mV setting, actual is variable

I scored 15 719 in Time Spy

AMD Ryzen 5 5600, AMD Radeon RX 7700 XT x 1, 32768 MB, 64-bit Windows 10}

www.3dmark.com

Graphics Score 18510

However I find this comparison more informative:

Result

www.3dmark.com

The second run is similar settings to above but instead -10% Power (208W total board power)

Graphics score is -4.9% with -25% power. But that still doesn't tell the whole story as this is the very best case for the overclock. In games like CP2077, Horizon ZD, Hogwarts Legacy using a mix of high and ultra settings @1440p, the difference is 1-3%. Barely more than run-to-run variation for 25% more power. No thanks, so I run at -10% 208W pretty much all the time. The only time I run the power up to ~+5% is if it's cold in the room and yes I do that on occasion, lol.

 

[Macro Device]

Poor Beginner Micro Devices can't even get more than 215w out of his.

Probably because 6750 XT is not as locked down as 6700 XT. I own the latter, not the former.

Graphics score is -4.9% with -25% power.

Very good.

In games like CP2077

Can you run the in-game benchmark on Ultra preset @ 1440p but with FSR turned off?

 

[Nordic]

Probably because 6750 XT is not as locked down as 6700 XT. I own the latter, not the former.

I've seen people push their 6700xt further than my 6750xt, despite the 6750xt having a memory speed advantage.

 

[Lew Zealand]

Can you run the in-game benchmark on Ultra preset @ 1440p but with FSR turned off?

Sure thing.

Ultra settings but all upscaling off (lol I did the test again when I saw it set to Custom, forgetting that turning off FSR makes it a "custom" run)

Sapphire Pulse 7700 XT
Cores OC to 2800 MHz min & 3400 max, actuals reported below
Memory OC to 2414, 2401 as reported in GPU-Z
UV to 1025mV in Adrenalin, actuals reported below

+15% power, @262W TBP
Cores 2770-2810 MHz
1036-1070 mV
80.76 fps

And bonus because I've made note of this card's preference for low power usage:

-10% power, @208W TBP
Cores 2650-2700 MHz
920-957 mV
79.11 fps

So 2% faster for 25% more power. This is typically what I see in games so I just run at 208W and I'll eat the missing 1-2 fps, lol.

FWIW, I run CP2077 at 1440p using HUB's quality settings from a vid they did a few years ago and here's the difference with the same settings:

@+15% power, 262W: 114.93 fps
@-10% power, 208W: 113.41 fps

Yeah, I'll eat that single percent, too

And for reference, with the HUB settings at stock GPU settings (229W) it gets 108.00 fps so the UV/OC delivers the vast majority of the OC potential with increasing power limit being minimal.

 

[Macro Device]

despite the 6750xt having a memory speed advantage.

This advantage is miniscule to say the least. RX 6700 series GPUs are not VRAM bandwidth limited, they could be perfectly balanced at ~3.5 GHz core clock without any VRAM tuning (7700 XT (12 GB VRAM, 432 GBps) only loses to 6800 non-XT (16 GB VRAM, 512 GBps) by a few percent in a very limited list of games and it's only happening at 4K/heavy ray tracing where both GPUs are far from ideal regardless). Alas, that's too hot.

The main reason why these 6700 XTs are faster is their core clock is higher and more stable. Considering identical voltage and different wattage, it all comes to current. I don't know what exactly ruins my party but I'd love to resolder a strap or flash some cheeky BIOS to have all the potential unlocked.

But at this point, everything I play is either running fast enough already or needs triple/quadruple digit percent worth of upgrade which is impossible by pure OC.

-10% power, @208W TBP
Cores 2650-2700 MHz
920-957 mV
79.11 fps

This is 41% faster than my 6700 XT... despite same wattage. Nice.

Holy cow, performance mods are something.

I genuinely did not change in-game settings, @Lew Zealand.

The GPU was running at ~2715 MHz core, ~2135 MHz VRAM, ~175 W core-only power draw.

 

[Lew Zealand]

Holy cow, performance mods are something.
View attachment 333125
View attachment 333126

I genuinely did not change in-game settings, @Lew Zealand.

The GPU was running at ~2715 MHz core, ~2135 MHz VRAM, ~175 W core-only power draw.

Wow that's an amazing mod improvement! Which mod(s)?

This is 41% faster than my 6700 XT... despite same wattage. Nice.

I'll note that I paid 43% more for my 7700 XT than my 6700 XT which contains all sorts of caveats like:

Got 6700 XT at the model's minimum price of $300 which is the best deal for a reasonable-cost good performing (raster) new card recently.
...meaning this 7700 XT at $430 also slots into a similar category of best deal for 1-tier-up performance in spite of the 7700 XT being slagged for poor price/performance*.
My 6700 XT is the opposite of a golden sample, more of a brown one so my personal performance delta is a little better for the 7700 XT.

* but only in DX12 and Vulkan games, which is both the future and more GPU-demanding so need this performance more. The 6xxx series is more well-balanced in DX11 and DX12 performance than the 7xxx series, which "underperforms" in DX11 by maybe 10-15% or so.

 

[Macro Device]

Which mod(s)?

https://www.nexusmods.com/cyberpunk2077/mods/10490.
I picked this one:

Used the 8 GB file despite having 12 GB VRAM (helps with stuttering) and also made some corrections to the core mod file itself. I'm still investigating this mod because I think I didn't achieve the full potential.
Didn't notice any image quality degradation.

I can send you my version in private but the original mod also delivers per se. I just made some corrections because this mod is meant for NVIDIA users so I had to comment out some NV-exclusive features + AMD-unfriendly features.

 

[Nordic]

I have not been using fast timings on my 6750xt yet. I've read a lot of people were struggling to have fast timings stable. Fast timings are stable for me at 2300mhz. None of the benchmarks I have tried show any measurable performance difference but theoretically it could improve performance somewhere.

 

[Macro Device]

I have not been using fast timings on my 6750xt yet. I've read a lot of people were struggling to have fast timings stable. Fast timings are stable for me at 2300mhz. None of the benchmarks I have tried show any measurable performance difference but theoretically it could improve performance somewhere.

Fast timings are a gimmick on RDNA2. Probably, on RDNA3 too, yet I'm not so sure about 7900 GRE with its weakest relative VRAM bandwidth amongst all this series. Stay on default and don't bother, especially with 6750 XT which has more VRAM bandwidth than it needs.

 

[Nordic]

Fast timings are a gimmick on RDNA2. Probably, on RDNA3 too, yet I'm not so sure about 7900 GRE with its weakest relative VRAM bandwidth amongst all this series. Stay on default and don't bother, especially with 6750 XT which has more VRAM bandwidth than it needs.

I don't see why I would not turn on fast timings when it doesn't effect power usage or stability.

 

[Noci]

With a RX 6700XT @ 3440 x 1440 without a performance mod (just optimized settings in Adrenalin) and an average of 72.8 FPS I can't complain looking at the numbers posted above.

 

[Macro Device]

With a RX 6700XT @ 3440 x 1440 without a performance mod (just optimized settings in Adrenalin) and an average of 72.8 FPS I can't complain looking at the numbers posted above.

View attachment 333214

Virtually the same resolution (3072x1728 is even ~7.1% more pixels than 3440x1440), significantly better average framerate.

What's causing lower minimum framerate, I don't know. It happens in vanilla as well. Maybe i5-12400F needs a replacement (just kidding, I'm gaming at 4K and the only reasonable thing to replace is the GPU).

 

[Lew Zealand]

Virtually the same resolution (3072x1728 is even ~7.1% more pixels than 3440x1440), significantly better average framerate.

What's causing lower minimum framerate, I don't know. It happens in vanilla as well. Maybe i5-12400F needs a replacement (just kidding, I'm gaming at 4K and the only reasonable thing to replace is the GPU).

View attachment 333216

I would love to see the effect of 6c12t vs 8c16t of the same gen (AMD and Intel w/no E-cores) over 10-30 runs to see if there's some "congestion" at certain points in the benchmark which somehow 8c16t alleviates. Or maybe it's just as variable with 8c16t. My 6c12t exhibits quite variable minimums on CP2077's benchmark as well as H:ZD's 1% lows, though it's a more consistent in H:ZD. That said, what is min in CP:2077? 1%, 0.1%? H:ZD has minimum and 1% output and it seems that minimum might be the single longest frametime as that is very variable.

 

[amd64skater]

I have not been using fast timings on my 6750xt yet. I've read a lot of people were struggling to have fast timings stable. Fast timings are stable for me at 2300mhz. None of the benchmarks I have tried show any measurable performance difference but theoretically it could improve performance somewhere.

Yeah I have tried both and fast timings I get a better score for benchmarking. I get stuttering if fast timing is on when I game.

 

[Macro Device]

I would love to see the effect of 6c12t vs 8c16t of the same gen

I tried that with my 8c16t 11th gen engineering sample (locked at 3800 MHz) last year. Difference wasn't major but it was real. About 5 to 8 percent better min FPS with 8 cores. Something along the lines of 50 VS 53. It was consistent.

In real game, though, no difference whatsoever, even in the most clogged Dogtown areas.

 

[bobalazs]

6700 XT Stock efficiency tested Cyberpunk2077 (second tab tested RDR2 -but based on undervolted 1.09V)
(real voltage shown, by hwinfo64, is apparently different from that reported from radeon software/driver)
CPU: Ryzen 5600
GPU: Sapphire Nitro+ RX 6700 XT
PSU: Seasonic B12 BC-550W
MB: Asus PRIME B450M-K II
2x16 Gb Patriot 3600 Mhz ddr4 CL 18
radeon driver 23.12.1



Link:
googleSheets

 

[Noci]

6700 XT Stock efficiency tested Cyberpunk2077 (second tab tested RDR2 -but based on undervolted 1.09V)
(real voltage shown, by hwinfo64, is apparently different from that reported from radeon software/driver)
CPU: Ryzen 5600
GPU: Sapphire Nitro+ RX 6700 XT
PSU: Seasonic B12 BC-550W
MB: Asus PRIME B450M-K II
2x16 Gb Patriot 3600 Mhz ddr4 CL 18
radeon driver 23.12.1



Link:
googleSheets


View attachment 333422

My XFX Speedster QICK runs the CP2077 benchmark @ app. 2800mHz and 1200mV.
With that it uses about 184W, not sure what this says about efficiency.

Have to say it's water cooled, GPU max 44c and hotspot max 65c. Most likely that has something to do with it.