Does 7900XT still has high idle consumption on multiple monitors? And should I wait for the Supers even when I'm leaning towards AMD cards?

[Avro Arrow]

Not sure shouting "You unfortunate AMD users AMD DRIVER SUCKS lol BUY NVIDIA HAHA" while showing off a 3090 and an Arc GPU of all things after used 3090 is deemed not an option and I ordered a 7900XTX is anything constructive, but it is what it is.
I will immediate concede if I'm really dragged into a dick contest (I have no interest in spending 3090 or 4090 money, watercooling and shunt mods; I have a prejudice / psychological limit? that any single component costing USD>1000 equivalent is not a good idea that was why I really hesitated on buying 7900XTX in the first place, and that limit was established as USD500 equivalent back when Geforce FX5800 was the top end card and I was a kid), but I will try to summarize the thinking process on the buying decision here (most of the points is already said somewhere in this thread, and by some other members in other "should I buy" thread ) :

Spoiler: My kinda simple but long thought process. YMMV. Walls of text warning

TLDR: I moved to a new place -> got a little more space -> gotta go 4K -> 3070 is overwhelmed -> need to buy new card -> both sides driver okay, feature no care -> Green pricing sucks -> Red lets goooooo

-Timing
The furthest I can wait is the Super launch. I'm completely fine with a consistently "smooth" V-Synced 30FPS like in a console (yes srsly), but the current setup fails to reach 30FPS consistently due to bad optimization / VRAM usage and I ain't wait until 2025.
I live in Macau, which means Taobao in Chinese is the best option in price (not necessarily for warranty; Local brick-and-mortar stores are better for warranty, but prices are always meh especially for AMD cards. Also, apparently the demand for current-gen AMD cards are very low to the point they don't stock the cards). And Taobao is running the 11.11 deals (think Black Friday, except it actually starts on 11.2 and lasts until 11.11 probably).
That leads to some implications:
- The 4090 ban has already turned the 4090 into unobtainium, and ruins the pricing of used 3090 and new 4080 and 4070Ti to an extent. (I didn't think of this when I made the buying decision, but the 4090 ban will also hurt 4080Super and 4070TiSuper because 4080S will be the best thing NVIDIA can legally ship to China)
- Even after 4080S launched, 4080 is not likely to price better than 7900XTX
- Pressure to cut price on 7900XTX is low
- See below; I will pick 7900XTX over 4080 if they are priced at the same level
- Judging on the deals I'm getting now, the pricing of the 7800/7900 series should be no worse than after Super launch
--> BUY NOW !
There are likely flaws on such an overdose of guesswork on this part, but it is what it is.

Brand preferences
-Driver
It may be a suprise to the guy trying to run dick contest, but I have said in 1st post that I have some good and bad experience on driver stability on the current card (3070) and last card (5700XT) before I added the 4K monitor. Granted I skipped the 1st year of the lifecycle on both cards. And I pretty much skipped the 1st year of 7900XT(X).
I know a lot of guys here says AMD DRIVER BAD, but I don't know if it is a outdated take or not; The experience on my 5700XT is not bad enough to say "I will never buy AMD". In terms of driver stability it is ... about the same as the 3070 on my current setup (before 4K monitor upgrade). The card before (1070) was better, but I was not seriously gaming back then.
This can be a usage issue anyway: I got some trivial stuff that I would prefer not to shutdown the PC even when I'm out of home. That leads to the PC being virtually 24/7, which might have screwed up some (V)RAM, and kicks up some graphic driver related gremlins when I want to start a game after a long day. I'm probably spewing nonsense here. Also, if I don't count the kind of crashes after several days of idling, there are virtually no crashes on the 3070 , 5700XT and 1070.
The day 1 review of 7900XT(X) listed those ridiculous idle power consumption numbers, and that's what started this thread.

-Features
RT performance: yeah, sure...?
DLSS: I still got very bad experiences with DLSS/DLAA on the games I play today (Forza Motorsport 8 and Forza Horizon 5: very awful ghosting on at least the player car). Granted FSR/XeSS might also be that bad, but screw it, I'm gonna go native as much as possible anyway. Yes I know, FM8 crashes on 7900XTX, but it is mostly fixed.
AntiLag: I despises any kind of shooter because I suck at them. AntiLag is not (likely) an issue.
Frame Generation: What is that? Can I eat that? /s
NVENC: kinda is an issue (I sometimes do video transcoding), but there are lots of workarounds.
Being a Read-Only member most of the time here on TPU, I can see the sentiment on the higher end is "If you care any of the features enough, buy NVIDIA; else buy AMD. It's pretty obvious that I don't care about the features.
When I brought up the idea of upgrading to my friends, one mentioned the dreaded 12-pin connector on 4080 and 4090 as a reason to steer away from NVIDIA. Although AFAIK it should be safe on everything below the X090 tier.
He also suggested to go for AIO models like AsRock AQUA because it takes less slots, but both water leak and clogging are things I really don't want to deal with.


-VRAM
The main issue that prompted me to upgrade is that 3070 (and its VRAM) is overwhelmed because I went 4K. So 16GB is a hard minimum.

Now I established features and driver stability is a non-factor, the minimum I have to buy on the green side is the 4080 (will be 4070TiSuper / 4070Super/16GB probably. 4060 is a joke and at best a side grade). On the red side there is the 7800XT.
4080 is firmly above USD$1000. Even if 4080 and 7900XTX is in the same price bracket I will likely pick 7900XTX. And the possibility that the green side throws a better deal even after Super launch is low. Maybe it will be outside Asia, but no chance in hell Taobao.
If I ended up going for anything lower than 7900XT, the recent news suggested that the new NVIDIA products will be very competitive on that price range.

Now the choice is 7800XT or go higher, and probably because my monitor is kinda high-end in terms of refresh rate, someone here suggested to go as high as 7900XTX, and I bought into his argument. And it ended up just below USD$1000. yay!
Part of the reasons I chose the M27U as the monitor upgrade despite M28U being cheaper is that in a Techspot / Hardware / Monitor Unboxed's feature article they said M27U is more optimized on 60~100?Hz. Granted, Size is precious, and M27U is a new model, I deliberately go lower-mid end on the 4K monitor spectrum, and I thought I was very likely to buy 7800XT as the card upgrade here, but I digress.

For reference, the pricing here for the cheapest 4070Ti and 4080 is ~10% lower and ~20% higher respectively when compared to the cheapest 7900XTX. 7900XT is ~17% cheaper than 7900XTX.

---

While I'm also suprised to see a Youtube video dedicated to a freaking driver update...That's 2023 AMD Graphics driver, both in a good and bad way.

---

Now the 7900XTX Hellhound has been delivered. This is the first top-end card I have ever owned. YAY! 5700XT doesn't count.

I'm still at work now. Can't wait to see how great (or bad) the new card turns out to be...

It's really rather simple for me. The RTX 4090 costs 70% more money than the RX 7900 XTX (even more for me because I got Newegg's open-box card that they're STILL advertising..lol) but only has 24% more performance and the same amount of VRAM. An intelligent person who is just a gamer doesn't have to think very long (or hard) to make that decision, eh?

I was thinking of ordering the Sapphire Pulse (by far the cheapest non-reference card, with a listed height of 53mm, 3-slot, it should be fine size-wise) and call it a day, but I wanted the Powercolor Hellhound (height 62mm, also listed as 3-slot, ~10% pricier) if size is not an issue.
Or , buy a cheap riser cable to put the M2 adaptor elsewhere. Although without using a riser cable ever in my life I'm probably spewing nonsense here.

The brand and model is of no real consequence when it comes to performance. I bought an ASRock Phantom Gaming model because it was on open-box sale at newegg.

 

[AusWolf]

You can also try enabling “maximum power savings” for the pci-e under the advanced power settings for the windows power profile.

3 monitors seems to be a tipping point between relatively decent idle usage and basically doubling idle power draw; especially with higher res, refresh, and bpc.

That's probably the thing. For decent idle power, you need all of your monitors to be relatively low res and low refresh rate. For example, dropping my 144 Hz display to 60 Hz drops idle power from 36-40 W to about 25.

Now I got my hands on the card, and after a cleaning and a messy cable running session...
Here are the numbers that somehow a few people a waiting for. Saaaaaaadly...
View attachment 320872
The numbers are hovering at 80~100W range. Using latest driver at the time (23.11.1) and quiet BIOS, if that matters.
Without the M27U (the 4K monitor) it will average at ~22W, with lows of 7W if I wait hard enough.
Also, the screenshot function doesn't work perfectly probably because of various monitor scaling / position, but I digress.

Other things I tried that didn't help the situation:
Changing color depths of M27U from 10bpc to 8bpc
Changing resolution of M27U 4K to 1080p (I can guess from the Windows setting screen that signal output is still 4K)
Changing refresh rate of M27U (default 150Hz, tried 144Hz, 60Hz, 59.94Hz)
EDIT: Advanced Power Settings-> PCIE -> Changing from Moderate? to Maximum
Flipping the switch for Eyefinity for the second time (see below)

Things that helped the situation in the wrong way:
Flipping the switch for Eyefinity for the first time
(because of the wacky positioning, of course that's gonna be a horror show. Maybe because I touched some other monitor settings for the monitors, the second time I flipped the switch the power draw stayed high.
The physical space constraint means that I can't put 3 monitors horizontally at the same time. But moar monitors are moar monitors, hence the wacky positioning / config.)

---

I really didn't think this through huh? What an idiot I am. : /
Now if Windows always remember to turn off the monitors...
( I guess maybe it's the trivial things I run that took away the attention of Windows. )

---

I swear the SG really isn't 61mm tall after I measured and put both cards next to each other for comparison.
Also, I really hate alcohol beverages (allergy I guess, but also general hate), but I appreciate the thoughts out there...

What's your idle power when the monitors are off?

You'll need to be careful with programs that use the GPU in the background. The Epic games launcher is notoriously bad in this regard during an install or update.

 

[Vya Domus]

Yeah with three monitors it's really unlikely you'll see low power figures at idle with any card, AMD or Nvidia.

 

[lawood]

Not necessarily true...

I have 1080 280hz, 1444p 165hz and 1080p 75hz, my idle is only 17w on nvidia, i've heard that's impossible on amd.

 

[Todestrieb]

It's really rather simple for me. The RTX 4090 costs 70% more money than the RX 7900 XTX (even more for me because I got Newegg's open-box card that they're STILL advertising..lol) but only has 24% more performance and the same amount of VRAM. An intelligent person who is just a gamer doesn't have to think very long (or hard) to make that decision, eh?


The brand and model is of no real consequence when it comes to performance. I bought an ASRock Phantom Gaming model because it was on open-box sale at newegg.

I feel like the main enemy of 7900XTX is always 4070Ti and 4080, and the target audience is very high-end but budget conscious enthusiasts, which makes the choice slightly (but not very) difficult.
I should have made my thought process on choices on partner models in another spoiler box in that post, but here we go:

Spoiler: Thought process on which brand partner to go

- Physical constraints. Anything bigger than 3-slot is a no go because of my PCIE cards situation. I could have bought a riser and hope for the best that it doesn't cockblock the new chonky card if I really have to go 4-slot.
- Brand reputation. On my side of the world, and for red cards, it is always Powercolor > Sapphire > everything else, with TUF being a wildcard of "Do you believe in TUF tax?"
- Cooler performance. I know Hellhond is more than good enough, but... Not sure if Hellhound is the outright best for 3-slot, or this is a super high-end card thing, but pulling 350W, maxing at hot spot 85C while being completely inaudible is quite ridonculous for someone who have never owned a super high-end card before.
- Price.

Yeah with three monitors it's really unlikely you'll see low power figures at idle with any card, AMD or Nvidia.

Maybe if it is triple 1080p I will get luckier. And the number I wanted to shoot for is a high but much more acceptable 40~50W anyway.

That's probably the thing. For decent idle power, you need all of your monitors to be relatively low res and low refresh rate. For example, dropping my 144 Hz display to 60 Hz drops idle power from 36-40 W to about 25.

I would like to try dropping to 60Hz for all three monitors, but for whatever reason both 1080p don't want to change refresh rate in Windows settings (144Hz and 59.94Hz respectively).

What's your idle power when the monitors are off?

You'll need to be careful with programs that use the GPU in the background. The Epic games launcher is notoriously bad in this regard during an install or update.

29W for Steam + Epic Game Store minimized on system tray + Firefox + a Chromium thing.
Single-digits for real idle. (things I listed above is all closed)

 

[AusWolf]

29W for Steam + Epic Game Store minimized on system tray + Firefox + a Chromium thing.
Single-digits for real idle. (things I listed above is all closed)

That's fine, then. Just set a really low monitor switch-off time and call it a day.

 

[Todestrieb]

That's fine, then. Just set a really low monitor switch-off time and call it a day.

Yes, b-but...
I would still hope for better numbers for general usage...and Windows doesn't forget to turn off the monitors...
hides in a corner

 

[AusWolf]

Yes, b-but...
I would still hope for better numbers for general usage...and Windows doesn't forget to turn off the monitors...
hides in a corner

I agree, but that's the best we can do, I guess.

 

[tabascosauz]

Now I got my hands on the card, and after a cleaning and a messy cable running session...
Here are the numbers that somehow a few people a waiting for. Saaaaaaadly...

The numbers are hovering at 80~100W range. Using latest driver at the time (23.11.1) and quiet BIOS, if that matters.
Without the M27U (the 4K monitor) it will average at ~22W, with lows of 7W if I wait hard enough.
Also, the screenshot function doesn't work perfectly probably because of various monitor scaling / position, but I digress.

Other things I tried that didn't help the situation:
Changing color depths of M27U from 10bpc to 8bpc
Changing resolution of M27U 4K to 1080p (I can guess from the Windows setting screen that signal output is still 4K)
Changing refresh rate of M27U (default 150Hz, tried 144Hz, 60Hz, 59.94Hz)
EDIT: Advanced Power Settings-> PCIE -> Changing from Moderate? to Maximum
Flipping the switch for Eyefinity for the second time (see below)

Things that helped the situation in the wrong way:
Flipping the switch for Eyefinity for the first time
(because of the wacky positioning, of course that's gonna be a horror show. Maybe because I touched some other monitor settings for the monitors, the second time I flipped the switch the power draw stayed high.
The physical space constraint means that I can't put 3 monitors horizontally at the same time. But moar monitors are moar monitors, hence the wacky positioning / config.)

---

I really didn't think this through huh? What an idiot I am. : /
Now if Windows always remember to turn off the monitors...
( I guess maybe it's the trivial things I run that took away the attention of Windows. )

---

I swear the SG really isn't 61mm tall after I measured and put both cards next to each other for comparison.
Also, I really hate alcohol beverages (allergy I guess, but also general hate), but I appreciate the thoughts out there...

That's unfortunate. But not unexpected, I guess. I think 90W is a Navi31 thing, but I would probably be expecting to double or triple my idle draw with your monitor setup too.

The only other thing potentially able to help your power draw is to fiddle with blanking settings in CRU. Didn't work for me on 7900XT, but some people were successful. Windows power settings and bit depth are not very helpful in this issue.

Don't be hard on yourself for it, multimonitor is always a guessing game. Reviews can't always give you relevant and accurate information, and I found out the hard way that talking bad about anything AMD is essentially taboo online. Makes it hard to discuss issues, share data and explore solutions.

4070 Ti is a big step down from what you have, and not really much of a 4K card. The downgrade is not something that's worth it for most people - I jumped ship because the various issues that Navi31 had at the time were not acceptable on a day to day basis, and 1440p is less of a practical difference to me. If you can eliminate 1 monitor, you should be able to reduce power.

Better to sit on the card for a bit before you decide. Hellhound and Pulse are both great cards. Remember that every time you change GPUs, you play the lottery for coil whine and fan quality. It really doesn't matter what brand the card is from - vast majority of 7900XT and XTX cards are reference PCB or heavily based on the same reference PCB. Sit tight if you've got a good sample.

 

[Todestrieb]

Blanking settings in CRU... That's a setting I've never heard of, and a program I didn't thought it would be any useful here.
Will try it later after some research.

Somewhere in this discussion I pretty much decided to forget the power draw numbers and yolo'd into 7900XTX. So of course I have accepted there are "risks" that the numbers will go high.

If I really really can't stand the numbers I probably would have second guessed, and think again about 7800XT (still, arguably not big enough of an upgrade IMO) or 4070TiSuper or 4080 (I will have to debate in my head: Do I hate the 8GB VRAM enough that I have to pull the trigger ASAP? Do I hate the power draw numbers enough to swallow the bullet and stay with the overpriced team green? ) as an upgrade.

But now the 7900XTX has done well in the short gaming sessions I had while staying ridiculously quiet, and until its driver turns out crashing too hard too frequently or I hit some really serious driver issues, I'm gonna keep this 7900XTX and hope for better drivers or workarounds.

2080Ti could be marketed as a 4K card, so 4070Ti is gonna be fine.. alright, not very much for 2023.

 

[AusWolf]

Blanking settings in CRU... That's a setting I've never heard of, and a program I didn't thought it would be any useful here.
Will try it later after some research.

Somewhere in this discussion I pretty much decided to forget the power draw numbers and yolo'd into 7900XTX. So of course I have accepted there are "risks" that the numbers will go high.

If I really really can't stand the numbers I probably would have second guessed, and think again about 7800XT (still, arguably not big enough of an upgrade IMO) or 4070TiSuper or 4080 (I will have to debate in my head: Do I hate the 8GB VRAM enough that I have to pull the trigger ASAP? Do I hate the power draw numbers enough to swallow the bullet and stay with the overpriced team green? ) as an upgrade.

But now the 7900XTX has done well in the short gaming sessions I had while staying ridiculously quiet, and until its driver turns out crashing too hard too frequently or I hit some really serious driver issues, I'm gonna keep this 7900XTX and hope for better drivers or workarounds.

2080Ti could be marketed as a 4K card, so 4070Ti is gonna be fine.. alright, not very much for 2023.

Same here, I decided to swallow the power draw and go with a shorter monitor switch-off time as a compromise. The card is absolutely brilliant in all other fronts.

The only thing that sits in the back of my head is the GPU fan turning on for a couple of seconds every minute or so when I'm browsing the web. I can't hear it, but I see it in GPU-Z and makes me think about fan lifespan. I guess I could turn off idle fan-stop in the driver, but... nah...

 

[tabascosauz]

Blanking settings in CRU... That's a setting I've never heard of, and a program I didn't thought it would be any useful here.
Will try it later after some research.

It's been a good couple months since I've done it so I'm kinda rusty. CRU has some built in "presets" some of which involve reduced blanking already. Could try those, and there's probably someone with a guide on how to do it. Anyway, it's far from guaranteed - it's like setting a lower refresh rate, just hoping that you can appease the VRAM gods, just something to try.

Like NVCP, AMD Software also has the option to create custom resolutions and achieve the same but it's more limited in what it can do. More annoyingly, for the longest time setting a custom resolution would immediately lock you into 6-bit color depth (which is NOT okay), it's a pretty well-known bug. Dunno if it's finally been fixed in the past few months, but I still just wouldn't use it since CRU exists.

Recently CRU has also been helping me out with adjusting VRR range on my monitors to try to get around Freesync flickering. Another reason to choose CRU, nothing else can do this much.

But now the 7900XTX has done well in the short gaming sessions I had while staying ridiculously quiet, and until its driver turns out crashing too hard too frequently or I hit some really serious driver issues, I'm gonna keep this 7900XTX and hope for better drivers or workarounds.
2080Ti could be marketed as a 4K card, so 4070Ti is gonna be fine.. alright, not very much for 2023.

sounds like a keeper then

I guess at the end of the day, any card is a 4K card if it works for what you need. I'm getting close to pulling the trigger on either Reverb or Quest 3, so about to put 4070 Ti to the test for "4K"

 

[Assimilator]

Yeah with three monitors it's really unlikely you'll see low power figures at idle with any card, AMD or Nvidia.

Monitors 2 through 4 are 2560x1440 @ 75Hz on an RTX A2000 connected over Thunderbolt 4 (#1 is the laptop's built-in screen), GPU is at minimum idle clocks and barely drawing double-digit amounts of power. No custom resolutions or anything, the only fiddling I had to do was tell Windows to drive the monitors at 75Hz because it defaults to 60Hz, because it's coded by idiots.

So no, it's not unlikely unless you don't use an NVIDIA GPU, because NVIDIA has had this shit figured out forever.

 

[tabascosauz]

Monitors 2 through 4 are 2560x1440 @ 75Hz on an RTX A2000 connected over Thunderbolt 4 (#1 is the laptop's built-in screen), GPU is at minimum idle clocks and barely drawing double-digit amounts of power. No custom resolutions or anything, the only fiddling I had to do was tell Windows to drive the monitors at 75Hz because it defaults to 60Hz, because it's coded by idiots.

So no, it's not unlikely unless you don't use an NVIDIA GPU, because NVIDIA has had this shit figured out forever.

To be fair, it's high refresh panels that Navi31 is allergic to. It doesn't like it when there are multiple monitors and among them are any number of 144Hz+ screens. 4 monitors is a lot but 1440p75 is for all intents and purposes basically 1440p60.

12W for 4 screens is pretty nice. Mine is at 6W just with single 1080p TV. But on 2 1440p165 screens I have seen regular 20W and 40W before on Turing and Ampere, so I wouldn't say it's not possible - just not as high as 80W.

 

[AusWolf]

Monitors 2 through 4 are 2560x1440 @ 75Hz on an RTX A2000 connected over Thunderbolt 4 (#1 is the laptop's built-in screen), GPU is at minimum idle clocks and barely drawing double-digit amounts of power. No custom resolutions or anything, the only fiddling I had to do was tell Windows to drive the monitors at 75Hz because it defaults to 60Hz, because it's coded by idiots.

So no, it's not unlikely unless you don't use an NVIDIA GPU, because NVIDIA has had this shit figured out forever.
View attachment 320966
View attachment 320970

I don't think it's an AMD vs Nvidia thing. I think it's more of a chiplet vs non-chiplet thing. Chiplet GPUs keep the VRAM in a higher power state with high refresh screens for some reason.

 

[Avro Arrow]

I feel like the main enemy of 7900XTX is always 4070Ti and 4080, and the target audience is very high-end but budget conscious enthusiasts, which makes the choice slightly (but not very) difficult.
I should have made my thought process on choices on partner models in another spoiler box in that post, but here we go:

- Physical constraints. Anything bigger than 3-slot is a no go because of my PCIE cards situation. I could have bought a riser and hope for the best that it doesn't cockblock the new chonky card if I really have to go 4-slot.

Well, all I can say is that my ASRock 7900 XTX PG isn't more than 3 slots wide:

- Brand reputation. On my side of the world, and for red cards, it is always Powercolor > Sapphire > everything else, with TUF being a wildcard of "Do you believe in TUF tax?"

That makes me laugh because usually I hear that Sapphire is #1. I have personally owned cards by Albatron, ASRock, Gigabyte, Palit, Powercolor, PNY, Sapphire and XFX. I've been more than satisfied with all of those brands. The only brands that I tend to shy away from are ASUS (I'm not paying for a name) and MSi (because I hate that company). I used to work for Tiger Direct and I remember that there was no significant difference in the number of defective returns for any of the brands. Every company that has branded video cards has been making them for more than a decade. Several of the card makers also produce motherboards (which are FAR more complex than video cards) so it's not like you need to worry about their level of technical expertise.

- Cooler performance.

With the exception of the reference card, they're all well within spec and you wouldn't notice any real difference between them. Having said that, my RX 6800 XT is a reference card and I never had any problems with heat or noise when using it so I don't think that matters either. When you see these reviews, they often have to split hairs because they consider it their job to point out differences, however small and insignificant those differences may be.

I know Hellhound is more than good enough, but... Not sure if Hellhound is the outright best for 3-slot, or this is a super high-end card thing, but pulling 350W, maxing at hot spot 85C while being completely inaudible is quite ridonculous for someone who have never owned a super high-end card before.

The Hellhound isn't even Powercolor's top-level card, the Red Devil is. Having said that, the Hellhound would definitely be the smarter purchase because any difference in performance would be imperceptible while the difference in cost is glaringly obvious.

- Price.

That's the second-biggest factor for me when choosing a video card (making sure that the card isn't MSi is the biggest factor for me). I don't care about RGB lighting or what the card looks like because it's inside my case and they all look the same to me in there. The only time that appearance would ever matter to me is if I'm looking at two cards priced the same and I have to choose one.

Maybe if it is triple 1080p I will get luckier. And the number I wanted to shoot for is a high but much more acceptable 40~50W anyway.

The Radeon software suite has an automatic undervolter which would probably go a long way to lowering that number for you while only sacrificing a few percentage points of performance. I often tell people that the best way to reduce power use in a PC is to make sure that you don't have an Intel CPU because in addition to sucking tonnes of juice, they also require fancy liquid cooling that increases power draw even more to run the pump.

I would like to try dropping to 60Hz for all three monitors, but for whatever reason both 1080p don't want to change refresh rate in Windows settings (144Hz and 59.94Hz respectively).

I can't see that making a huge difference so I wouldn't worry about it.

29W for Steam + Epic Game Store minimized on system tray + Firefox + a Chromium thing.
Single-digits for real idle. (things I listed above is all closed)

That's not too bad considering that it's a high-end video card. Remember that video cards have always sucked a tonne of juice and people still managed just fine. I remember the Powercolor Devil13 (which had twin R9 290 Hawaii GPUs on-board) had a TDP of 580W about 9½ years ago. Despite having the highest TDP that I've ever seen for a consumer-grade video card, it was only three slots wide:

Don't worry too much about it, you'll be just fine.

 

[Todestrieb]

I see it in GPU-Z and makes me think about fan lifespan. I guess I could turn off idle fan-stop in the driver, but... nah...

Sure if you can forget about the power draw you can also forget about GPU-Z for a while, right...?
......right? : p

sounds like a keeper then

I guess at the end of the day, any card is a 4K card if it works for what you need. I'm getting close to pulling the trigger on either Reverb or Quest 3, so about to put 4070 Ti to the test for "4K"

\Cheers/ (with a can of soft drink)
Can't wait to see upscaling artifacts up close in the VR goggles! /s

---

@Avro Arrow
->brands : Well, I know a few people who believe in the TUF/STRIX religion, so...
->Cooler : I have gone out my way to buy lots of bequiet! and Noctua stuff for the sake of noise level. And my current home have a pretty low noise floor, so a really quiet cooler is kinda important. But hey, preferences.
I think Hellhound probably had a slightly better cooler (not sure!) with a few added features (BIOS switch, LED switch to change color / turn off RGB entirely) which Pulse doesn't have. Hellhound ended up only ~3% pricier than Pulse instead of ~10%, which is worth it IMO. And hey, every card is cool, but Hellhound is cooler.
-> Undervolt : I don't see undervolt being useful in helping idle power draw. Single digit power draw is only seen when all the monitors are turned off by Windows). General non-gaming usage for my 7900XTX started at 80W (before all the tinkering below), so yeah.

---

After some 4 hrs of investigating and fiddling with CRU presets and numbers and finding compromises...
The compromises I ended up with is lowering the AOC 24G2 to 1080p100Hz. VRAM stays at 909MHz, and the real idle power draw (monitor on) is now ~49W. Hurray!

The exact settings I changed is:
M27U: Extension blocks -> Display ID 1.3 ->Detailed resolutions -> Both 4K150Hz and 4K144Hz -> use Auto(PC) preset
24G2: Detailed resolutions -> Both 60Hz and 144Hz -> use Exact preset
Further changes to blanking values doesn't help in my case.

I came across a video from the driver update guy before. He suggested to create a custom resolution in driver side with slightly lower than max refresh rate and CVT preset timings. All CVT presets in driver and CRU ended up incompatible with my both high refresh monitors, but YMMV.

Spoiler: A few other findings

For the sake of simplicity and my sanity, in this spoiler box, VRAM goes low = <200MHz, VRAM goes mid = 909MHz, VRAM goes high = 2485MHz.
In previous posts I hadn't dig a step deep enough to change the refresh rate for 24G2 and VP228H. Now I found it without asking CRU to do that.
The 24G2, rather than the M27U, is more likely the reason that VRAM goes high.

Before touching any settings, using the previous settings as the baseline (4K150 for M27U, 1080p144 for 24G2, 1080p60 for VP228H)
removing M27U -> VRAM goes low.
removing 24G2 -> VRAM goes low.
removing VP228H -> VRAM stays high.
Any 1080p monitor only -> VRAM goes low.
( I forgot to see what happens with M27U only. )
Changing M27U to 1080p or 1440p doesn't change these behaviours. (probably because the signal is still 4K.)

The first tinkering in CRU got a lot of things screwed up, so I have to do a reset all.
After that, I noticed some weirdness on VRAM clock, so, with all three monitors on 59.94Hz (VRAM is low here) as the baseline in this test:
M27U to 60Hz -> VRAM goes mid (!)
24G2 to 60Hz -> VRAM goes high (!!)
VP228H to 60Hz -> VRAM stays low
All three to 60Hz -> VRAM goes high
At this point I tried to flip the switch for Freesync for 24G2 on both driver and monitor side. Nothing changes.
After a few fiddlings on presets, I found a preset that will fix the above behaviour. (change to 60Hz VRAM stays low)
Changes for this behaviour for M27U is not listed above: Detailed resolutions -> 4K60Hz -> use Native preset
A reset all somehow doesn't reset this behaviour for 24G2, which made me worried that I might have changed something permanently. Maybe a setting wipe / DDU can deal with this, I guess.

Then I fiddled some intermediate refresh rate and found that with 24G2 at 100Hz VRAM will go mid.
As it looks impossible for VRAM to go low while keeping the max refresh rate, and I can't find a way to push 24G2 back to 120Hz or 144Hz without VRAM going high again no matter what the M27U does, about this point I decided to call it a loooong day.

At this result config (4K150, 1080p100, 1080p60) the VRAM is mid, but if I change M27U to 119 or 120Hz VRAM will go high.

 

[AusWolf]

Sure if you can forget about the power draw you can also forget about GPU-Z for a while, right...?
......right? : p

That's what I tend to do, yes.

 

[pavle]

I don't think it's an AMD vs Nvidia thing. I think it's more of a chiplet vs non-chiplet thing. Chiplet GPUs keep the VRAM in a higher power state with high refresh screens for some reason.

I believe it's more of an AMD thing because of how ATi handles 2D - it's all through shaders, has been since R300 and looks like they're using VRAM much more than nvidia (don't know their method).
Also in 3D I constantly see lower VRAM utilisation on nvidia than on DAAMiT.
As for your idle fan turning off - you could try setting higher temperature condition for turning on.

 

[AusWolf]

I believe it's more of an AMD thing because of how ATi handles 2D - it's all through shaders, has been since R300 and looks like they're using VRAM much more than nvidia (don't know their method).
Also in 3D I constantly see lower VRAM utilisation on nvidia than on DAAMiT.
As for your idle fan turning off - you could try setting higher temperature condition for turning on.

That sounds like a plausible theory - just doesn't explain why the card keeps the VRAM alive, but not the shaders.

 

[pavle]

That sounds like a plausible theory - just doesn't explain why the card keeps the VRAM alive, but not the shaders.

With so many shaders (so powerful GPU) I would guess it doesn't need to speed up, just RAM because resolution and bandwidth requirements.

 

[AusWolf]

I just got a solution that may work:

If your CPU has an iGPU, then connect your displays to your motherboard. The idle power consumption on my 7800 XT went down to 18 Watts, even when playing a Youtube video!

On the downside, either load power consumption, or power reporting is now all over the place. It fluctuates between 130 and 320 W, which is very strange considering that the card comes with a 246 W default power limit. Interestingly, clocks are stable, and so is game performance, so I have no idea what's going on.

 

[Vya Domus]

I believe it's more of an AMD thing because of how ATi handles 2D - it's all through shaders, has been since R300 and looks like they're using VRAM much more than nvidia (don't know their method).
Also in 3D I constantly see lower VRAM utilisation on nvidia than on DAAMiT.
As for your idle fan turning off - you could try setting higher temperature condition for turning on.

I don't know what you mean, every GPU will use it's shaders for 2D.

AMD may have more idle power consumption because of more memory modules and chiplet architecture.

 

[TumbleGeorge]

~5%* for Infinity Fabric ~2%* for Infinity cache. The remaining consumption is for the other elements of the architecture. Number for Infinity Fabric I found here.

*For Navi 31

 

[AusWolf]

One more observation:

Plugging only the primary display into the GPU (3440x1440, 144 Hz in my case), and the secondary (1024x600, 43 Hz) into the motherboard sets idle power consumption right: 18 W sitting on the windows desktop, which is less than half of what it used to be. No more need for an ultra low display switch-off time!

Vdeo playback power is still high at roughly 50 W, but at least GPU behaviour in gaming is now correct.