The AMD Driver Thread

[tommybhoy]

so I'm left to think it's the graphics card (or GPU driver). Anyone else having a similar issue?

It's more likely an OS issue imo, I've disabled system/monitor sleep/standby for years regardless of vendor because of black screens and manually turning off the monitor.

 

[henson0115]

also getting fan settings reset after every reboot on swapping from a 6800(dying) to 7800xt, driver is 23.11.1, fastboot already disabled.
edit:23h2 seems to have somehow fixed it

 

[GhostDog1981]

Anyone got any tips on how to improve my undervolt overclock?

I have a 7800xt MBA and it does everything I need it to do at 1440p.

Very quiet and cool max hotspot is 75C undervolted.

These are my stable everyday settings:

Core clock 2900mhz max 500 min

Vram clock: fast timings 2600mhz

Fan curve max 75% after 75C

Voltage 1040mv (less than this got direct x errors on MW3- as although Timespy was stable at 985mv!)

This pulls 290W

3D Mark Timespy 21900-22000 on few runs.

Getting 160-180FPS on Max everything settings at 1440p on MW3

 

[GhostDog1981]

Anyone got any settings they care to share? Re above.

Thanks.

 

[ICDP]

The 7x00 range do not undervolt the way people think they do. Setting lower mV does not lower the power used, it allows for more headroom on the core clocks as the core is running cooler. The power slider dictates the power budget, not the undervolt.

To use less power while getting more (or the same as stock) performance, you need to lower .mV and also set the power slider into negative.

This gives more core clocks at a lower power budget. Simply moving the power slider to max tells the GPU to use more power regardless of your undervolt.

Edit: In my testing of overclocking and undervolting a 7900 XT.

1. Adjust the power slider until the GPU is using your preferred watts. The AMD overlay will show your GPU power usage.

2. Adjust core clocks, VRAM timings and VRAM clocks to your desired setting.

2. Reduce (undervolt) your .mV until it becomes unstable. Then bump it up a bit until stable and test using actual graphical intensive games rather than synthetic benchmarks.

2. Keep testing for stability in actual games NOT canned benchmarks.

Using the above methods I can overlock with actual 10% extra performance but about 50w extra power requirements. Or my preference is under 300w power used for stock performance.

 

[GhostDog1981]

The 7x00 range do not undervolt the way people think they do. Setting lower mV does not lower the power used, it allows for more headroom on the core clocks as the core is running cooler. The power slider dictates the power budget, not the undervolt.

To use less power while getting more (or the same as stock) performance, you need to lower .mV and also set the power slider into negative.

This gives more core clocks at a lower power budget. Simply moving the power slider to max tells the GPU to use more power regardless of your undervolt.

Thanks I will try this (reduce power limit to 0%) tonight.

 

[ICDP]

Thanks I will try this (reduce power limit to 0%) tonight.

It is different than how the 6x00 range worked. What you have done previously is a balls to the wall overclock with max power allocation.

Just remember the power slider dictates power usage, which is far more intuitive.

 

[henson0115]

both my 6800 and 7800xt seems to work the same way in terms of undervolting, both give more headroom for core clocks via less voltage and temperature. both seem to go past what the power slider allows in my case too.
edit: @ICDP are you sure you are just not seeing the difference in chip sample quality?

 

[ICDP]

both my 6800 and 7800xt seems to work the same way in terms of undervolting, both give more headroom for core clocks via less voltage and temperature. both seem to go past what the power slider allows in my case too.
edit: @ICDP are you sure you are just not seeing the difference in chip sample quality?

My 6700 XT and 6800 would drop voltage with an undervolt. The 7900 XT seems to always use the same power regardless of undervolt. Though it is based on a low sample size

 

[MadMatty]

I recently picked up the 7800Xt to upgrade from a 290 and am still getting to know the different undervolt/overlock mechanism. No real issues encounters but an oddity in Timespy CPU score which is notably down compared to the 290. I cant work out why that may have happened, my cpu oc profiles haven't changed and it was the exact same testing environment for both. Uninstalled old drivers, ddu clean and then latest drivers. Power profile set to ultimate. While I benchmark in Win10 due to ease of reducing background apps, the results are more or less the same with win11. Checked rebar setting to no effect.



I've only noticed the oddity in Timespy normal, but I cant seem to change anything to restore cpu score. Maybe Radeon driver is not optimal. Might uninstall ryzen master to see if its conflicting in some way.

 

[henson0115]

I've only noticed the oddity in Timespy normal, but I cant seem to change anything to restore cpu score. Maybe Radeon driver is not optimal. Might uninstall ryzen master to see if its conflicting in some way.

done any bios updates latelty, theres been a few security flaw fixes which can decrease performance on x570, but that seems a lot in your case

 

[MadMatty]

done any bios updates latelty, theres been a few security flaw fixes which can decrease performance on x570, but that seems a lot in your case

No bios updates recently. All i've really done is remove old drivers (Final official legacy for the 290) DDU, re-installed latest chipset drivers in case ddu touched them ... then installed latest drivers for the 7800.

 

[henson0115]

No bios updates recently. All i've really done is remove old drivers (Final official legacy for the 290) DDU, re-installed latest chipset drivers in case ddu touched them ... then installed latest drivers for the 7800.

ah 1 thing i noticed in my firestrike/timespy runs was background tasks using the cpu, kill all of your background task and see how it affects your cpu scores. maybe its something as simple as that? you had less task running on the first run to the second perhaps?

 

[MadMatty]

ah 1 thing i noticed in my firestrike/timespy runs was background tasks using the cpu, kill all of your background task and see how it affects your cpu scores. maybe its something as simple as that? you had less task running on the first run to the second perhaps?

I do check before starting the benchmark run, as well as waiting until the cpu temp has dropped (temp via debug display on mobo is helpful fo knowing when cpu has dropped to idling.) All my past 290 runs are high 15k right up to 17k with high oc but I'm barely able to reach 14k with the 7800 in any run with low and high cpu oc. I just wonder if any other 7800 owners have noticed such a pattern or maybe any of the 7xxx cards.

 

[henson0115]

I do check before starting the benchmark run, as well as waiting until the cpu temp has dropped (temp via debug display on mobo is helpful fo knowing when cpu has dropped to idling.) All my past 290 runs are high 15k right up to 17k with high oc but I'm barely able to reach 14k with the 7800 in any run with low and high cpu oc. I just wonder if any other 7800 owners have noticed such a pattern or maybe any of the 7xxx cards.

had a similar result, but put it down to run variance and more task running my end on the 7800xt run, my 6800 is overclocked compared to the 7800xt which is running stock in that run

SCORE17 638 with AMD Radeon RX 7800 XT(1x) and AMD Ryzen 9 5900X​

Graphics Score19 548
CPU Score11 355

SCORE16 916 with AMD Radeon RX 6800(1x) and AMD Ryzen 9 5900X​

Graphics Score17 988
CPU Score12 646

same cpu curve but ive also changed bios since the 6800 run

 

[MadMatty]

Interesting. There is still a difference that is larger than the margin for error. I've made a post on 3d mark steam forum as well to see if anyone there has seen a similar issue.

 

[welshrat]

had a similar result, but put it down to run variance and more task running my end on the 7800xt run, my 6800 is overclocked compared to the 7800xt which is running stock in that run

SCORE17 638 with AMD Radeon RX 7800 XT(1x) and AMD Ryzen 9 5900X​

Graphics Score19 548
CPU Score11 355

SCORE16 916 with AMD Radeon RX 6800(1x) and AMD Ryzen 9 5900X​

Graphics Score17 988
CPU Score12 646

same cpu curve but ive also changed bios since the 6800 run

Is there really only that much difference between those two cards. I have a 6800 and and very happy with it still but was getting a little jealous of the 7800xt, seems there really isn't a huge amount in it

 

[GhostDog1981]

Is there really only that much difference between those two cards. I have a 6800 and and very happy with it still but was getting a little jealous of the 7800xt, seems there really isn't a huge amount in it

I get nearly 22000 in Timespy for my 7800xt mba once overclocked. So about a 6950XT and almost a 4070ti level.

There’s only 4000 difference between that and a stock 7900XT (26000)

 

[welshrat]

Ah ok that seems more like it. Still going to hold out for a cheaper 7900xt next year sometime I think or maybe the refreshes.

 

[GhostDog1981]

Ah ok that seems more like it. Still going to hold out for a cheaper 7900xt next year sometime I think or maybe the refreshes.

Or RX 8000?