Im new to using AMD cards, (well modding them anyway) and Im curious if raising the TDP from 230w will be beneficial in more performance? And if so can someone help me with doing so? I cant find any videos or much info at all on raising or removing power limits on these cards..
Please Help!! 6700 xt Power limit removal.
Legion6700XT
21 - 40 of 48 Posts
Joined
·
4,782 Posts
the 6700XT you have isn't bad, as far as I know the QICK variant has a fewer parts compared to the reference AMD design and high end models from XFX, that may have been the limiting factor, overall if your games run fine, which I do know for a fact given this is the mid tier flagship card. I see no reason for you to upgrade so hastily on newer gen..
3D Mark is synthetic, and scores can easily scale up if you go all out on the CPU, my advise, is test with people having the same GPU with games..
Joined
·
333 Posts
12900ks and threadrippers are about the same graphics score as me, So im GPU bound now.. and yeah maybe ill wait till next gen.. see if I can grab a 6900 or 6800xt used or for less new... But yeah this card is maxed out by an i5
and as far as ive seen my QICK is one of the few which use both power coords...and is the c1 variant... 230watt...so it dont get too much better for the 6700
Joined
·
333 Posts
Joined
·
5,553 Posts
Gigabyte Aurus Elite 😭
Joined
·
3,382 Posts
Are you folks under-volting ? I had way better results under-volting, I was able to get +1000 over the stock timespy score, for 230W anyways. I never tried upping the power limit. Then I partially broke it with my copper-mod attempt. Now it's almost back to working.
Joined
·
1,135 Posts
You are beyond braindead. Never try to overclock anything ever again.
Highest I’ve had my 6750 XT Red Devil pulling is 316w in Timespy. But it doesn’t push the cards performance all that much over stock. I also scored almost 14.9K GPU with just a 12100F B660m DDR4 BCLK OC at 5.2ghz. So I do have the top 12100F/6750 XT score by a long shot. The GPU score is likely held back a bit by the CPU so maybe I’ll try it with my 12400, 12600K or 12900KS system.
Games it’s rarely pulls more than 250w at stock voltage and OC. Occasionally will go up to around 260-270w but it’s rare. Undervolted a bit in games and OC I see around 220-230w.
Games it’s rarely pulls more than 250w at stock voltage and OC. Occasionally will go up to around 260-270w but it’s rare. Undervolted a bit in games and OC I see around 220-230w.
Joined
·
333 Posts
nah never tried....@ 235watts or so... with MPT help... it went from 13.9 highest stock clock gpu score to 15,047 highest with voltage up.......timespy that is.....in fire i went from 39k to 43k...
Imma step up the voltage here and see what happens....BUT i'd like to know what MPTs throttler options are and the options on the main screen where you can turn off the deepsleep stuffm, IDK what half those check mark boxes do and none of the throttler ones in MPT.......I cant find any information on this stuff, the throttler stuff im interested in... the ppt0-1.. etc etc..i read a bit about this from the linux guys and you can limit your card to ONE frequency for clock which helps a bunch in 3dmark benches as... we all knw... the first graphics test is easy at high clock to pass, the second in timespy much more demanding and i can crash alot there...Anyone know this info on the MPT options? luckilly I found an old AMD tuner program that lets me set the core to ONE frequency....But it still jumps 100mhz up sometimes randomly from my mix/max 1 frequency setting..... others it stays perfectly to the clock i tell it to... frustrating.. but fun to mess with cards... when i went #1 with my 12600k in like every test i turned it down and just been gaming....as I was beating processors that were paired with the 6700xt (thread rippers and 12900k etc) GPU scores with my 12600k @ 5.3-5.4ghz, so im obviously GPU bound... i'd love to get a 6800xt...so expensive tho..... I've never really undervolted a card before... being long time nvidia user.. infact the last ATI card I had was a 9600 pro...loved that thing. Care to shed some light on how your undervolt went?
Joined
·
1 Posts
I have a 6700xt in a water, core class hits 2997 under full load @40c junction. I have never been able to break past the 15% pwr limit increase, not even with any tool above. Was hoping that someone had a new tool, but I guess not lol
Which brand/model?
I have a PowerColor Red Devil and have no problems pushing up to 280-300w on the stock cooler with a repaste at 60c / 80c hotspot. However, I need 1.25v for 2820-2920 clocks on the slider. In games, it stays around 2820-2830. 2950 is a driver imposed hard limit, so not sure how you would be running that high.
Yes the 15% is a driver hard limit too, but you can increase the base power limit as much as you want. With +15%, this is 281w at full load. Most games still only use 240-260w average for me.
Joined
·
6 Posts
Hi, i have exact same model 6700 XT. Currently i set 2915 MHz on Radeon Software, in Superposition frequency around 2780-2810, temps are 47-50 GPU 65-71 Hotspot and consumption around 220-210W. Now im trying to get higher consumptions via MPT. Is there anything i should do on MPT? I set your GPU, GFX and SoC limits.
Wrote as SPPT, i think i dont need save and flash the bios..?
Also these are Voltage and Watt consumptions at 2915 MHz
Correct, you do not need to save and flash the bios, but you do need to restart the graphics driver or reboot for settings to take effect.
That is normal power usage for 1.20v in Superposition. If you are looking for the last 35 MHz, you can try enabling TEMP_DEPENDENT_VMIN in "Features" tab > "PPTable Features" > "Feature Control". Then on the "Power" tab, set Temperature dependent Vmin:
The following configuration hits 2836-2860 MHz in Superposition 1080p Extreme and ~60c / 78c hotspot with full fan speed. Power consumption is 240-255w. Time Spy / Time Spy Extreme are not stable at these frequencies, I need to lower to 2820-2920 range and ~2800 clock in Time Spy with ~280-290w maximum power consumption with +15% in Radeon Software.
The total gains from SOCCLK + PHYCLK (VRAM controller) both at 1075 MHz and 2150 MHz FCLK (Infinity Fabric) + MALL (Infinity Cache) are ~1% in Superposition. TimeSpy gains ~2.5%.
Joined
·
6 Posts
So i've tried to increasing PL, GFX, SoC and Vmin/high values by 290-260-45 and 1250-1250 only for Vmin/high GFX levels and maxing out all settings in Radeon Software: 2950-2150 MHz. +15 Power Limit.
At %100 fan speed, in Superposition around 55~/70~ Degrees. But the problem is no matter what i do, (i tried increasing power limits) frequency newer goes 2900 MHz or closer. Just staying around between 2830-2860 MHz. Is it because FCLK, PHYCLK? If it is i would like to try, and what is FCLK the limit for 6700 XT?
This is normal behavior of the boost algorithm, it will only hit 2900+ MHz at a very light load.
You can play with lowering (yes, lowering) the lower boost clock range in Radeon Software, sometimes for example a 150-400 MHz total range actually results in 10-30 MHz effective clocks in some workloads (ie. 2550-2950 instead of 2850-2950).
You have effectively maximized your GFXCLK. I would suggest testing in Time Spy, lower voltage slightly (~5mV) until unstable then add 5mV, it should be stable in gaming from my experience. Lowering voltage and temperature may result in slightly higher boost, but for me it is only ~10 MHz, maybe 10-15 points in Superposition.
FCLK should do at least 2000 or so with stock voltage, not sure about any limit, but you can try up to 2200. I would start with 2000 and move up in 20 MHz increments. Near the limit, you may see a regression in performance, when unstable, it will cause black screen and system reboots - the driver will not recover. It will be unstable before you hit any driver imposed limit, if one even exists.
PHYCLK performance gains are extremely minimal. I can run up to 1200 PHYCLK but scores are worse. My scores are slightly better with SOCCLK = PHYCLK at 1:2 FCLK. None of these other clocks will affect GFXCLK but score efficiency / FPS will improve.
Joined
·
6 Posts
Hi again, I recently bought R5 5600 and B550-F Gaming. After overclocking CPU, and RAM i tried to do Time Spy with my current MTP and Radeon settings. I set up, in Graphics Test 1 after 15 seconds i got error: test cancelled by user, also this happened in Fire Strike too. Now im trying to understand what is problem, you have any idea why is this happening? Since i upgrading CPU and mobo i cannot use my settings, im thinking because of voltage.
Current MTP settings:
These are my previous/stable settings before upgrade. No heat issue, no FPS-freq drop.
BTW i am i set Vmin Low/High SoC limit for now. Currently trying, before that it was 0.
This seems to me like the graphics driver is crashing due to too high frequency for the GFX CLK. Lower the boost limit (2950 to 2940, 2930, etc.) and test. If you only tested with Superposition before, note that Superposition is easier to pass than Time Spy or even Fire Strike.
Also make sure TEMP_DEPENDENT_VMIN is enabled in "Features" tab > "PPTable Features" > "Feature Control" or the voltage override will not work. If using this, you should set TVmin to 50 and hysterisis to 20 in that section for both GFX and SoC.
For only 2000 FCLK I would lower SoC Voltage to 1150mV (stock voltage). My sample, which isn't great, only needs 1168mV for 2100 FCLK and 1181mV for 2150 FCLK. 1181mV may create unnecessary heat and power consumption and may lower the lifespan of the part. I am unsure about long term safety.
Assuming 1250mV on GFX is actually being set, it may be unnecessary. Some GPUs will overclock to maximum boost at stock voltage, many RX 6700 XT cannot however. You can try to find your maximum at 1225mV or 1237mV, 1250mV can cause excessive power consumption and temperatures in some scenarios. Usually +12mV (~+15-20w power, extra heat) is not worth it for only 20-30 MHz extra clock speed, unless you are benchmarking only.
For example 24/7 setting now is 2820-2920 MHz boost (GFX CLK) set in driver at 1237mV GFX, with 1075 MHz SOCCLK and PHYCLK, 2150 FCLK at 1181mV SoC and 2150 MHz memory w/ Fast Timing set. The extra voltage needed for the last 20-30 MHz GFX CLK is not worth it to me for 0.3 FPS more in a 90 FPS game, however I did max out the GPU to set a couple of Unigine Superposition and 3DMark Speed Way top scores.
Joined
·
6 Posts
2950 Max frequency works in Superposition, Heaven and GPUPI, as you said they are not a reliable test for stability as Time Spy and Fire Strike. However lowering frequency to 2900 and 2915 MHz, in Time Spy didnt get any crash or issue, test completed succesfully.
To stabilize the max frequency I tried to go further to increasing voltage by 005 mV, lastly tried 1262 mV which I think a bit high but didnt solve the problem in Time Spy, also pass the Superposition.
Currently set Vmin/max values by 1257 mV @2915 MHz (didnt check 2920, 2930 I know it crashes at 2950). I will try to lowering mV and find maximum boost clock.
Also I am doing it for benchmarking and playing few games, not using it 24/7. Now trying to 2150 FCLK settings with 1181 mV SoC value, if its woudnt be stable I will lower it.
Current settings as you suggested:
Just got back from vacation and saw you submitted some scores at HWBOT.
I see poor scaling with GFX > ~1250mV, but for Time Spy and Speed Way it does continue to scale minimally up to 1275mV at the expense of extreme power consumption and heat.
A few extra points can be gained by tweaking SOCCLK (Soc) and PHYCLK but it is extremely minimal. Too high PHYCLK will lower Time Spy graphics score up to 200-300 points, even if the driver doesn't crash and there are no visible clues. My assumption based on the behavior is that there is an error correction mechanism.
Your Time Spy graphics is higher than I can achieve by ~150 points. I also have lower efficiency in Superposition 1080p Extreme compared to the #1 result, maybe because my motherboard is limited to PCIE 3.0. Tried 15 runs with various optimizations. The 8K optimized test is great, though.
I scored 14 204 in Time Spy
AMD Ryzen 9 5900X, AMD Radeon RX 6700 XT x 1, 32768 MB, 64-bit Windows 11}
www.3dmark.com
Joined
·
6 Posts
Welcome backJust got back from vacation and saw you submitted some scores at HWBOT.
I see poor scaling with GFX > ~1250mV, but for Time Spy and Speed Way it does continue to scale minimally up to 1275mV at the expense of extreme power consumption and heat.
A few extra points can be gained by tweaking SOCCLK (Soc) and PHYCLK but it is extremely minimal. Too high PHYCLK will lower Time Spy graphics score up to 200-300 points, even if the driver doesn't crash and there are no visible clues. My assumption based on the behavior is that there is an error correction mechanism.
Your Time Spy graphics is higher than I can achieve by ~150 points. I also have lower efficiency in Superposition 1080p Extreme compared to the #1 result, maybe because my motherboard is limited to PCIE 3.0. Tried 15 runs with various optimizations. The 8K optimized test is great, though.
I scored 14 204 in Time Spy
AMD Ryzen 9 5900X, AMD Radeon RX 6700 XT x 1, 32768 MB, 64-bit Windows 11}
If you increase GFX values, frequency slightly improves 10-30 MHz just for increasing 10 mV (usually). But as you said there is too much power consumption. My last test was with 1262 mV but I did not go further.
You should try to get 1st place with your SOCCLK, PHYCLK and other settings. Just try, set GFX 1257 Vmin/max and increase Radeon frequency by 2950. I am sure you would get more score.
Also I dont think its because of PCIE 3.0, I got my previous score (7831) with B460 mobo, you have seriously good settings on MTP, as i said you have to try 2950 and high voltages at full speed. Your both 1080p and 8K scores would be crazy.
21 - 40 of 48 Posts
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
