AMD RX 7000 series GPU Owners' Club

[tabascosauz]

in case anyone's wondering, no, there is no fine wine to be had in 23.1.1

Stock with 2 monitors (don't mind the CPU score, typical anomaly):

Stock with 1 monitor:

Undervolt target 2750 @ 1050mV:

 

[AusWolf]

in case anyone's wondering, no, there is no fine wine to be had in 23.1.1

Stock with 2 monitors (don't mind the CPU score, typical anomaly):
View attachment 278722

Stock with 1 monitor:
View attachment 278724

Undervolt target 2750 @ 1050mV:
View attachment 278723

Wait a second... in your 1 monitor setup, your CPU score is higher? How does that work?

 

[kurosagi01]

AMD Software: Adrenalin Edition 23.1.1 for AMD Radeon™ RX 7900 Series Graphics Release Notes | AMD

New 23.1.1 driver, anyone on it yet? From the release notes:



So, in the past 3 weeks, AMD fixed absolutely NOTHING

I just tried switching the second monitor over to HDMI 2.0 and disabling Freesync, card still completely incapable of handling a second monitor on 22.12.2. Gonna try 23.1.1 now

The funny thing is that while I was gathering Genshin metrics today with CapframeX, multi monitor and single monitor repeatedly come out the exact same on frametimes. Frametimes and FPS, almost identical data. Yet, it's blatantly obvious to anyone with a set of eyes that the multi monitor setup is CONSTANTLY stuttering. Maybe it's a Freesync problem that shit just hits the fan when there's a second monitor connected? As my vertical monitor, the S2721DGF already has its Freesync permanently disabled just to try and minimize any compatibility issues.

I'll be sticking with the previous driver, unless the future drivers has a game that requires it.

Just got my 7900 XTX Red DEVIL limited, and is a great card, buuut the OC side is a bit werid? need more TDP for this card, the card can "run" at 3100-3200Mhz core but will lose alot of fps instead of 2900/3000 cause of hitting TDP limit.

Will make a video about it soon!?

This is something Jay has pointed out with overclocking the 7900s, you can push the power limit, voltage and core clock. You can boost it but get less performance, but the memory clock drops dramatically.

For me the best thing to do is, leave the target clock speed, set the power limit to 15% and see what Adrenalin "auto" assign the clock speed and then manually set that clock speed and then undervolt until you can run it at a stable voltage.

 

[tabascosauz]

Wait a second... in your 1 monitor setup, your CPU score is higher? How does that work?

12800-12900 is my normal score at -25. CPU just decided to run a little slow that run. Happens.

Also, I'm still figuring out the full extent of it, but latest HWInfo is causing a fair bit of stuttering in games. I've stopped using it for now and just using the Radeon overlay

 

[AusWolf]

12800-12900 is my normal score at -25. CPU just decided to run a little slow that run. Happens.

Also, I'm still figuring out the full extent of it, but latest HWInfo is causing a fair bit of stuttering in games. I've stopped using it for now and just using the Radeon overlay

Yeah, but your GPU score is higher with 2 monitors (within margin of error, I'd say), so that can't be the cause of the difference in total score.

 

[tabascosauz]

@kapone32 have you set a custom fan curve yet? These stock fan curves are batshit insane. Basically until my GPU hit 70C edge temp the fans wouldn't spin past 500-1000rpm. Which was also visually annoying because two of the fans wobble below 700rpm.

I kid you not, light games pulling about 100W ran almost as warm as 250-300W loads because the fans were stuck at 500rpm.

I set a new curve that has the fans start around 800rpm at the fan stop threshold, continuing gradually up to about 72% PWM at 70C, and afterwards a straight ramp to 100% (which I will basically never hit when undervolted). Dunno about the fans on your Pulse but the reference fans are pleasantly quiet up until about I'd say 1500rpm.

Yeah, but your GPU score is higher with 2 monitors (within margin of error, I'd say), so that can't be the cause of the difference in total score.

It's only a 200 point difference total, pretty safe to say the CPU score is responsible for that one. Doesn't matter anyway because the undervolt score is what I'm after. To get past 13k CPU score all I have to do is dial my CO back up to the usual -28; I was back down at -25 just for troubleshooting.

I know the recommendation right now is +15% power limit and leave the clock target, but I'm not really interested in running XTX power (355W) through the XT reference cooler. It handles it fine but it's kinda antithetical to what I'm after with undervolting.

Here's a stock power limit run (308W), at 1050mV and runs a few degrees cooler while scoring ever so slightly better than stock. Gonna try keep reducing volts, but if this is a stable profile then I think I'll just leave it. And wait patiently for the finewine LOL

So far Genshin is happy, MW19 is happy, MW2 is happy (about 5C lower), Sniper Elite 5 is happy, next up are Arma and DCS.

 

[kapone32]

@kapone32 have you set a custom fan curve yet? These stock fan curves are batshit insane. Basically until my GPU hit 70C edge temp the fans wouldn't spin past 500-1000rpm. Which was also visually annoying because two of the fans wobble below 700rpm.

I kid you not, light games pulling about 100W ran almost as warm as 250-300W loads because the fans were stuck at 500rpm.

I set a new curve that has the fans start around 800rpm at the fan stop threshold, continuing gradually up to about 72% PWM at 70C, and afterwards a straight ramp to 100% (which I will basically never hit when undervolted). Dunno about the fans on your Pulse but the reference fans are pleasantly quiet up until about I'd say 1500rpm.



It's only a 200 point difference total, pretty safe to say the CPU score is responsible for that one. Doesn't matter anyway because the undervolt score is what I'm after. To get past 13k CPU score all I have to do is dial my CO back up to the usual -28; I was back down at -25 just for troubleshooting.

I know the recommendation right now is +15% power limit and leave the clock target, but I'm not really interested in running XTX power (355W) through the XT reference cooler. It handles it fine but it's kinda antithetical to what I'm after with undervolting.

Here's a stock power limit run (308W), at 1050mV and runs a few degrees cooler while scoring ever so slightly better than stock. Gonna try keep reducing volts, but if this is a stable profile then I think I'll just leave it. And wait patiently for the finewine LOL

View attachment 278732

So far Genshin is happy, MW19 is happy, MW2 is happy (about 5C lower), Sniper Elite 5 is happy, next up are Arma and DCS.

Yes I did set a fan profile but I would have to re do it every time I restarted or had a Windows Update. Since the latest driver that seems to be working fine though. The Memory goes to a maximum of 86 C so that is better and I see the highest RPM at 3264

Yeah my block is out for Delivery!! Daughter is at school and Wife is at work so no distractions. I can't wait to get a look at the bare PCB.

 

[MrPerforations]

hello Club,
im having no luck getting anything out of the AMD™ Sapphire™ Radeon RX7900XTX nitro+.
i am having lots of no display output problems.
are you having the same issue please?
using display port 1.4 lead to cheapest 4k monitor (28"@ £180)

thought of a plan for next time i do a thermal paste job.
going to put the thermal paste in a plastic bag, dump it in some hot water and then the thermal paste will have a lower viscosity, using less, making it easier to spread and and thinning out the layer.
might want to try it Kapone32.

 

[GamerGuy]

I'd suggest leaving the card untouched first, get it running first before messing around with repasting and such. I assume the driver was a clean install in new OS, or that you've done your due diligence by running DDU to remove all previous driver IF the card's going into a pre-existing system.

 

[MrPerforations]

i water blocked it yesterday and only had the idea today.

 

[kapone32]

i water blocked it yesterday and only had the idea today.

So since you put the block on you have no Display?

 

[tabascosauz]

Yes I did set a fan profile but I would have to re do it every time I restarted or had a Windows Update. Since the latest driver that seems to be working fine though. The Memory goes to a maximum of 86 C so that is better and I see the highest RPM at 3264

Yeah my block is out for Delivery!! Daughter is at school and Wife is at work so no distractions. I can't wait to get a look at the bare PCB.

3200rpm?? holy moly

I didn't notice any temp improvements from the driver alone, but my memory temp was down at 70C on the dot with the core undervolt profile in MW2. Granted, it was only running about 250W in that game, but I don't think an extra 58W is enough to negate a 12-20C decrease in memory junction. So it might be worth chopping off some volts, if you haven't already.

I wish Radeon software could read mem temps so I wouldn't always have to fire up hwinfo and introduce stutters just to see it

thought of a plan for next time i do a thermal paste job.
going to put the thermal paste in a plastic bag, dump it in some hot water and then the thermal paste will have a lower viscosity, using less, making it easier to spread and and thinning out the layer.
might want to try it Kapone32.

In order for that to work water's gotta be quite warm, long soak, and you need to immediately apply after removing it from the water.

 

[kapone32]

3200rpm?? holy moly

I didn't notice any temp improvements from the driver alone, but my memory temp was down at 70C on the dot with the core undervolt profile in MW2. Granted, it was only running about 250W in that game, but I don't think an extra 58W is enough to negate a 12-20C decrease in memory junction. So it might be worth chopping off some volts, if you haven't already.

I wish Radeon software could read mem temps so I wouldn't always have to fire up hwinfo and introduce stutters just to see it

Well I just got an email from UPS that my block is almost here. I am going to see if Watercooling is the way to go. No more fan profiles and no more memory temp issues (hopefully). That might be the solution. We will see

 

[tabascosauz]

"Nvidia drivers are just as bad as AMD drivers"

I'm not sure what exactly this is yet, because the first and second fields in Latencymon don't show the offending process name. All I know is that 20ms stutter makes games borderline unplayable.

 

[TheoneandonlyMrK]

Sign me up, I have a reference 7900XT from powercolor, still the first hour up and I might refresh the OS but I'm impressed, a decent step on from a Vega64.

 

[tabascosauz]

Moral of the story, check your DP cables and make sure you have enough PSU for the job, because AMD's 750W recommendation is apparently horseshit? Either that or AMD is about to inform me that an almost-new SF750 is a low quality power supply

Undervolt is finally working now, stutter-free across the board in everything, even in Timespy. Default 2870MHz target, 1050mV, +0 power limit. Even running Timespy with FRTC on is now at 25k graphics score and approaching previous stock score (25.5k).........have never before met a GPU so picky about hardware

Not even one crash since new Club3D cable and HX1000.

 

[kurosagi01]

Moral of the story, check your DP cables and make sure you have enough PSU for the job, because AMD's 750W recommendation is apparently horseshit? Either that or AMD is about to inform me that an almost-new SF750 is a low quality power supply

Undervolt is finally working now, stutter-free across the board in everything, even in Timespy. Default 2870MHz target, 1050mV, +0 power limit. Even running Timespy with FRTC on is now at 25k graphics score and approaching previous stock score (25.5k).........have never before met a GPU so picky about hardware

Not even one crash since new Club3D cable and HX1000.

View attachment 279538

That's great you was able to achieve that, I may need to try and see mine will be Ok without the +15% power limit.
I did install 2023.1.1 to see if it will fix my issue where the driver(Adrenalin) may crash if a game was to crash with it, the only fix was restarting the PC.

 

[tabascosauz]

That's great you was able to achieve that, I may need to try and see mine will be Ok without the +15% power limit.
I did install 2023.1.1 to see if it will fix my issue where the driver(Adrenalin) may crash if a game was to crash with it, the only fix was restarting the PC.

I think every crash I've had except one was a hard freeze that required holding power button or pulling the plug. Kinda appalling since when Geforce crashes it almost always only crashes the driver or dwm.exe.

Was gonna say that you should be more than safe with a 850W Focus, but then again 5900X also games at like a consistent +100W power draw over the 5800X3D, so who knows.

 

[kurosagi01]

I think every crash I've had except one was a hard freeze that required holding power button or pulling the plug. Kinda appalling since when Geforce crashes it almost always only crashes the driver or dwm.exe.

Was gonna say that you should be more than safe with a 850W Focus, but then again 5900X also games at like a consistent +100W power draw over the 5800X3D, so who knows.

The weird behaviour happens sometimes if I was to run the overlay before loading the game. The game would go on a flicker frenzy then eventually crash and then the driver crash.
The overlay short key sometimes doesn't want to load with the game which then cause the game to crash and the driver to crash.
I've done a power curve optimiser on the 5900x which should help a little with the power draw on the 5900x. It has certainly reduced the temps on top of upgrading the cooler I had previously.
The fan control and undervolt on the XT is in place aswell.
Could just be the 2 games Im playing just needs updating to support the 7900 series.

 

[Vya Domus]

So I've noticed the fans on my card do in fact stop but the bizarre thing is that it only happens once when I first boot the PC, after that even if the temperature goes below the threshold it wont stop.

 

[tabascosauz]

So I've noticed the fans on my card do in fact stop but the bizarre thing is that it only happens once when I first boot the PC, after that even if the temperature goes below the threshold it wont stop.

Didn't you say your room ambient was like 25+? If you can't get junction temp below 40C consistently it probably will not fan stop. It doesn't trigger immediately, usually about a minute or so passes after junction temp goes below 40C + low load before fans stop. Seems to be working more reliably on 23.1.1, maybe placebo.

Once you DO hit 0rpm though, they will not spin up again until GPU edge temp hits 50C or so.

What I do is set a custom fan curve to increase fan speeds across the entire fan-on range, so that it cools down much faster to get below the fan stop range.

The weird behaviour happens sometimes if I was to run the overlay before loading the game. The game would go on a flicker frenzy then eventually crash and then the driver crash.
The overlay short key sometimes doesn't want to load with the game which then cause the game to crash and the driver to crash.
I've done a power curve optimiser on the 5900x which should help a little with the power draw on the 5900x. It has certainly reduced the temps on top of upgrading the cooler I had previously.
The fan control and undervolt on the XT is in place aswell.
Could just be the 2 games Im playing just needs updating to support the 7900 series.

Which 2 games?

Curve optimizer doesn't reduce MT power usage on 5900X. Big temperature reductions to be had, for sure, if you run a higher offset across the board, but no real power decrease unless you intentionally reduce power limits. I used to have one before X3D. Lighter games range from 60-90W, heavier games well above 100W. X3D runs light games as low as 30W, most games around 50W, rarely in the 60s or 70s.

I haven't had the overlay crash the system yet. There are certainly games that it doesn't want to load in on, but that's all that happens. How much stuff do you have enabled in the overlay? I only monitor GPU and VRAM, not keen on increasing system load needlessly. Maybe it improves if you slim down the overlay a bit?

I basically stopped running HWInfo because it was causing absolutely horrendous stuttering just being on in the background. I didn't even enable Asus EC monitoring. HWInfo is back to auto start now, but only monitoring RAM temps and VRAM junction temp, everything else disabled. No more stutering for the most part.

Clearly, the "improved" power consumption graphs from the XT review aren't telling the whole story, if a SF750 => HX1000 upgrade solved all my crashing problems.

 

[kurosagi01]

Which 2 games?

Curve optimizer doesn't reduce MT power usage on 5900X. Big temperature reductions to be had, for sure, if you run a higher offset across the board, but no real power decrease unless you intentionally reduce power limits. I used to have one before X3D. Lighter games range from 60-90W, heavier games well above 100W. X3D runs light games as low as 30W, most games around 50W, rarely in the 60s or 70s.

I haven't had the overlay crash the system yet. There are certainly games that it doesn't want to load in on, but that's all that happens. How much stuff do you have enabled in the overlay? I only monitor GPU and VRAM, not keen on increasing system load needlessly. Maybe it improves if you slim down the overlay a bit?

I basically stopped running HWInfo because it was causing absolutely horrendous stuttering just being on in the background. I didn't even enable Asus EC monitoring. HWInfo is back to auto start now, but only monitoring RAM temps and VRAM junction temp, everything else disabled. No more stutering for the most part.

Clearly, the "improved" power consumption graphs from the XT review aren't telling the whole story, if a SF750 => HX1000 upgrade solved all my crashing problems.

I think the CPU W reported on the overlay was averaging around 70-90w for me and the 2 games are COD MW2 and NFS Unbound.
I may try and reduce the number of overlay shown as I just left it as the default overlay setup and also not have hwinfo running in the background as well.

 

[tabascosauz]

I think the CPU W reported on the overlay was averaging around 70-90w for me and the 2 games are COD MW2 and NFS Unbound.
I may try and reduce the number of overlay shown as I just left it as the default overlay setup and also not have hwinfo running in the background as well.

No problems here with MW2. In fact it's one of the only games that actually did work properly on the previous 22.12.2 driver LMAO. But I also just run it Vsync with Anti-Lag and 162fps global FRTC, not chasing fps with "competitive" settings.

idk if DLSS2 adds latency, because input lag on this setup even with Vsync on is better than 3070 Ti DLSS Vsync off

I'm not sure I would rely on the overlay for any reporting related to CPU, I would only trust HWInfo with Snapshot Polling for temps/power/clocks there. MW19 and MW2 run similarly on CPU load, and MW19 was up around 100W consistently for MW19.

 

[Chomiq]

Clearly, the "improved" power consumption graphs from the XT review aren't telling the whole story, if a SF750 => HX1000 upgrade solved all my crashing problems.

Didn't GN cover this in their transients video, the thing that SFX PSU's sometimes have to make compromises in terms of circutry due to size constraints and that's why often people that report black screen issues were running some sort of SFF build?

 

[Vya Domus]

I have a 750W PSU as well and no crashing for me even when GPU-Z records 450W+ power spikes.

Didn't you say your room ambient was like 25+? If you can't get junction temp below 40C consistently it probably will not fan stop. It doesn't trigger immediately, usually about a minute or so passes after junction temp goes below 40C + low load before fans stop. Seems to be working more reliably on 23.1.1, maybe placebo.

Once you DO hit 0rpm though, they will not spin up again until GPU edge temp hits 50C or so.

What I do is set a custom fan curve to increase fan speeds across the entire fan-on range, so that it cools down much faster to get below the fan stop range.

I'll look if it makes any difference but I can't see how the junction temp can ever realistically get below 40C, considering that it's about 5-7C over the GPU temp, so for the junction to be below 40C I'd need like 35-33C GPU temp, my water cooled 1080 barely ran that cold at idle though the temperatures aren't necessarily comparable.