AMD RX 7000 series GPU Owners' Club

[AusWolf]

My vote is for the 7900 XTX. It's a monster of a card that i'm very happy with.

Even if the rumours about RDNA 4 offering 7900 XTX-like performance for a much lower price and power consumption are true?

Thanks for answering, by the way.

 

[Deleted member 237813]

Ive managed to settle into the 2600mhz region but its not fully stable yet. Vcore is at 1025.

no way this is stable in very demanding games. avatar melts you core clocks and will show instabilitys very well.

there are games where 950 would be stable but avatar needs 1070mv for 2600. But since its only an offset it will only cause core clocks anyway. rdna 3 can only save power via core clock reduction and powerlimit.

i set it with chill per game to game basis. its awesome simply set profile for a game. rdr 2 i only use 2200mhz since its at 70-80fps with chill lock to 80 fps as i dont need more its at 240-250 watts native 4k .

Even if the rumours about RDNA 4 offering 7900 XTX-like performance for a much lower price and power consumption are true?

Thanks for answering, by the way.

that should be the case nonetheless if a new generation releases. we dont need rumors for that. the rumor is no high end sku. but getting the performance from last gen for less money is normal..

 

[Ferrum Master]

Even if the rumours about RDNA 4 offering 7900 XTX-like performance for a much lower price and power consumption are true?

Thanks for answering, by the way.

It came from MLID... it is a hoax at best. Front page writers had a slow news day and didn't have anything better to write about... yellow press or its like in local news firefighters took down some cat from a tree.

 

[AusWolf]

It came from MLID... it is a hoax at best. Front page writers had a slow news day and didn't have anything better to write about... yellow press or its like in local news firefighters took down some cat from a tree.

Good point.

So how is the 7900 XTX on idle power consumption? The last I checked, it was terrible, but back then, the 7800 XT wasn't good either, and since then, it got lowered by quite a lot by the newest drivers.

 

[Deleted member 237813]

lower than nvidia if we talk real idle. also depends on the model of course.

 

[Zach_01]

Good point.

So how is the 7900 XTX on idle power consumption? The last I checked, it was terrible, but back then, the 7800 XT wasn't good either, and since then, it got lowered by quite a lot by the newest drivers.

I could believe RDNA4 at 7900XT+ level at 600-700 US MSRP but 7900XTX at 7800XT (550?) price is too far fetched...

As for (TBP) power the Nitro+ XTX with latest official 24.1.1 is hovering between <10W (rarely I've seen 5W min) and < 30W depending whats on the screen and if you move pointer, scrolling, task bar showing/hiding and so.
When something is moving like scrolling a page power shoots briefly to 45+W.
On screen saver goes around 45-50W and with monitor off <20W (I've seen avg about 15W)
Whatching movies (full screen or not) depends on where and what it is. From 45 to 65 maybe.

This is from last 20min while loading a few vids, typing here but no true idling, like nothing moves, and no screen saver.
True idling in my case is 10~25W. So avg should be like monitor off.

 

[Deleted member 237813]

it will depend on how nvidia will price their cards and how fast they are. if a 5070 is 7900xtx for 699. Amd has no reason to give it for 599 away with more vramand performance as usual.

 

[Ferrum Master]

Good point.

So how is the 7900 XTX on idle power consumption? The last I checked, it was terrible, but back then, the 7800 XT wasn't good either, and since then, it got lowered by quite a lot by the newest drivers.

I had an argument with W1z about it, as I don't believe his dedicated device setup measurements are very accurate here. The point is, whatever what - look at frequencies. If your core clocks and RAM clocks sleep at their lowest while you do casual stuff, everything is fixed and fine, it is very HW setup dependent. And latest drivers drivers have fixed it even more, you are right. W1z also doesn't use latest drivers that did more fixes in recent super review tables too, I understand it is a pain to re-benches for basically usual nothing, so that's not a complaint, it is something for us to understand and take into account. Just look at your particular setup.

I am living on RADV and Linux lately, I dual boot into windows only for the darn Lightroom. In Linux using dual monitors setup everything hovers as low it can for my setup. Vram 96MHz and core at 4-50MHz.(5-20W peaks, avg 7W). There are OC bugs for 79xx series that are still being worked on in kernel 6.8. It is connected with the DRM firmware value bits, the commands to alter values have changed and everyone is confused a bit. But using at stock everything works fine. Gaming wise Steam has done so much, for those titles I play currently I don't see problems. I am not fond of the RT IQ we are presented or the PR machinery telling us that upsacaling is better than native... yeah sure... just disable built in crap TAA. So upsacling is used only for rare emulation cases for me. The vaseline like IQ in recent years amuses me. I rather play jaggy no AA, but sharp. People say DLSS/DLAA does wonders, it is better than native, sure it is better if a smarter sharpening algo is cleaning after the current AA/Shader bullshit and created a problem that should not be there in the first place. RT wise the path tracing is better, sure... but at what price? 2K$ to play at ~30-60FPS with a 4090? No thanks.

 

[AusWolf]

I could believe RDNA4 at 7900XT+ level at 600-700 US MSRP but 7900XTX at 7800XT (550?) price is too far fetched...

As for (TBP) power the Nitro+ XTX with latest official 24.1.1 is hovering between <10W (rarely I've seen 5W min) and < 30W depending whats on the screen and if you move pointer, scrolling, task bar showing/hiding and so.
When something is moving like scrolling a page power shoots briefly to 45+W.
On screen saver goes around 45-50W and with monitor off <20W (I've seen avg about 15W)
Whatching movies (full screen or not) depends on where and what it is. From 45 to 65 maybe.

This is from last 20min while loading a few vids, typing here but no true idling, like nothing moves, and no screen saver.
True idling in my case is 10~25W. So avg should be like monitor off.

View attachment 332994

That looks very much like my 7800 XT at idle / low load. Thanks.

I had an argument with W1z about it, as I don't believe his dedicated device setup measurements are very accurate here. The point is, whatever what - look at frequencies. If your core clocks and RAM clocks sleep at their lowest while you do casual stuff, everything is fixed and fine, it is very HW setup dependent. And latest drivers drivers have fixed it even more, you are right. W1z also doesn't use latest drivers that did more fixes in recent super review tables too, I understand it is a pain to re-benches for basically usual nothing, so that's not a complaint, it is something for us to understand and take into account. Just look at your particular setup.

I am living on RADV and Linux lately, I dual boot into windows only for the darn Lightroom. In Linux using dual monitors setup everything hovers as low it can for my setup. Vram 96MHz and core at 4-50MHz.(5-20W peaks, avg 7W). There are OC bugs for 79xx series that are still being worked on in kernel 6.8. It is connected with the DRM firmware value bits, the commands to alter values have changed and everyone is confused a bit. But using at stock everything works fine. Gaming wise Steam has done so much, for those titles I play currently I don't see problems. I am not fond of the RT IQ we are presented or the PR machinery telling us that upsacaling is better than native... yeah sure... just disable built in crap TAA. So upsacling is used only for rare emulation cases for me. The vaseline like IQ in recent years amuses me. I rather play jaggy no AA, but sharp. People say DLSS/DLAA does wonders, it is better than native, sure it is better if a smarter sharpening algo is cleaning after the current AA/Shader bullshit and created a problem that should not be there in the first place. RT wise the path tracing is better, sure... but at what price? 2K$ to play at ~30-60FPS with a 4090? No thanks.

I couldn't agree more! Upscaling is a tool to help when your GPU can't handle native rendering. Sure, DLAA is great, I guess, but we have RSR which does the same thing just without the overhyped green AI magic. Not that I'd ever want more than 3440x1440 native anyway.

Paying south of $2K for a graphics card is a definite no for me as well. Heck, I could get a Sapphire pulse 7900 XTX for 950 GBP, and even this feels like pulling teeth. I'm not sure if it's worth forking up the cash by selling the 7800 XT and some random stuff, and then watching its market value deteriorate when RDNA 4 is out. Maybe just waiting for RDNA 4 and using lower settings in the meantime would be a more sensible plan. I'm really in conflict here (first world problems, I know).

 

[Ferrum Master]

Paying south of $2K for a graphics card is a definite no for me as well. Heck, I could get a Sapphire pulse 7900 XTX for 950 GBP, and even this feels like pulling teeth. I'm not sure if it's worth forking up the cash by selling the 7800 XT and some random stuff, and then watching its market value deteriorate when RDNA 4 is out. Maybe just waiting for RDNA 4 and using lower settings in the meantime would be a more sensible plan. I'm really in conflict here (first world problems, I know).

I justified to purchase because of Lightroom AI denoise speed, not gaming whatsoever. It is a must these days, as it allows me to shoot my A7m4 camera at much higher ISO, my previous card could do it in 50s on one image, this card does it it 7-10s multiple that by few hundred pics you take in one weekend trip if you refrain using burst shots. So I save up a lot of my almost non existent free time for hobbies and time is worth the money keeping my usual workflow, leaving overnight batch jobs ended up with much things to be redone or simply crashed, because it is Adobe.

I don't believe 7900XTX will fall below 900 Quid soon as the Supers are still expensive in EU and brang nothing really. It will sooner fall short on supply than being cheaper, they had their lesseon with 6xxx series overstock that hurt 7xxx series. That's why I picked up my card last summer, and I was right, the price didn't drop down. I got mine around 970Euro. Selling 7800XT is a great idea, no matter how you look at it. It will age terribly and loose value faster.

I don't have hopes for RDNA4 being faster in general. It will have much faster RT performance, for sure, there will be backports from their CDNA branch not that AMD cares for RT, it will be a byproduct from the development where the real cash flows in for them. Other than that... it will be cheaper for them. The power envelope is unknown, they will decide it as the last bit with the last beta silicon they will get in their labs and measure optimal settings, it has never been the other way around. If it clocks high using more power, they won't cripple it down for the sake of being less TDP, nobody knows really how the tech node will act with their final design, it is all lies we read now. Their money now was on the upcoming console refresh with the ~ RDNA3.0-3.5, and make that right. Nvidia users will sulk on that again. Somehow Sony and Microsoft also doesn't see any point in doing mainstream RT, I wonder why.

Well... considering the fact that photography is my hobby also... also film... one roll of E100 costs around 30Euro's, but if you like convert it into craft beer values, that's only like 3-4 beers... everything is expensive these days.

 

[Lew Zealand]

Well, I don't think i'll bother with any sort of undervolting this time around. I'm seeing effective clocks of nearly 3.2Ghz at stock setup (apart from the fan curve tweak) which is just mind blowing to me:
View attachment 332428

The card is whisper quiet, cool and plays everything I can think to throw at it. Really happy with the purchase

FWIW when testing my Pulse 7700 XT (first mention in this thread??) it can boost to the same but as some have mentioned, it's load and power dependent. Ya also gotta set the Min and Max (OK and voltage) in Adrenalin properly too.

This is multiple runs of the Forza Horizon 4 benchmark, 3174 MHz is during the last run and it still doesn't hit max power during most of the run and is above 3100 for the whole thing. The 4th from last one there is also over 3000, I was just ramping up at that point before testing some lower power settings.

On the other hand, CP2077 and Hogwarts Legacy only get to ~2800 with the same settings, pegged at max 262W power use. So very load-dependent. Seems like all the chiplet 7xxx GPUs perform similarly. I wonder if AMD was targetting 3000-3200 MHz and ended up surprised by excessive power usage in DX12 or higher loads so dialed things back.

 

[AusWolf]

FWIW when testing my Pulse 7700 XT (first mention in this thread??) it can boost to the same but as some have mentioned, it's load and power dependent. Ya also gotta set the Min and Max (OK and voltage) in Adrenalin properly too.

View attachment 333377

This is multiple runs of the Forza Horizon 4 benchmark, 3174 MHz is during the last run and it still doesn't hit max power during most of the run and is above 3100 for the whole thing. The 4th from last one there is also over 3000, I was just ramping up at that point before testing some lower power settings.

On the other hand, CP2077 and Hogwarts Legacy only get to ~2800 with the same settings, pegged at max 262W power use. So very load-dependent. Seems like all the chiplet 7xxx GPUs perform similarly. I wonder if AMD was targetting 3000-3200 MHz and ended up surprised by excessive power usage in DX12 or higher loads so dialed things back.

I don't think AMD "got surprised" by anything. It's just that the card's boost algorithm can detect what's best in every scenario. Hogwarts and CP are complex DX12 titles that also use some RT if enabled. I guess that uses more parts of the GPU than Forza does. With more parts active, there's less headroom for extra clock speed, just like your CPU boosts higher when unpacking files than in some complex AVX work. I'm pretty sure it's normal and accounted for.

 

[Lew Zealand]

I don't think AMD "got surprised" by anything. It's just that the card's boost algorithm can detect what's best in every scenario. Hogwarts and CP are complex DX12 titles that also use some RT if enabled. I guess that uses more parts of the GPU than Forza does. With more parts active, there's less headroom for extra clock speed, just like your CPU boosts higher when unpacking files than in some complex AVX work. I'm pretty sure it's normal and accounted for.

Yeah FH4 is a light load on all GPUs, I tested AC:Ody with the Vulkan rendering mod and it also hit 3000+ with maxed out power which I hadn't seen on the card before, just around 2800 Max on my typical heavier DX12 games. W1zz mentioned in the review that he saw a large OC on the 7700 XTs to over 3000 MHz but that was also in an older, light benchmark: Unigine Heaven. This was something I hadn't seen at all until today so was a bit of a surprise.

In other cards with older architectures (Turing, Pascal, RDNA2) the card would go unstable when overclocked too far on a light load before the power limit was hit, I was surprised to see a different behavior here, to maintain stability up to almost 3200 MHz. Possibly higher, I never got it to crash as I was done at 3170.

 

[AusWolf]

Yeah FH4 is a light load on all GPUs, I tested AC:Ody with the Vulkan rendering mod and it also hit 3000+ with maxed out power which I hadn't seen on the card before, just around 2800 Max on my typical heavier DX12 games. W1zz mentioned in the review that he saw a large OC on the 7700 XTs to over 3000 MHz but that was also in an older, light benchmark: Unigine Heaven. This was something I hadn't seen at all until today so was a bit of a surprise.

In other cards with older architectures (Turing, Pascal, RDNA2) the card would go unstable when overclocked too far on a light load before the power limit was hit, I was surprised to see a different behavior here, to maintain stability up to almost 3200 MHz. Possibly higher, I never got it to crash as I was done at 3170.

Those are super high clocks!

Maybe I should start tinkering with my 7800 XT. The only thing I've done so far is push the power limit to the max, which makes it more stable at 2400 MHz in every game instead of fluctuating based on different loads (2200 MHz in some games, 2300-2350 in others).

I wish it was as easy to overclock RDNA 3 as it is the 6500 XT. With that card, I can drag every slider to the max, and the card will hit a hard wall at 2950 MHz. Not that it matters much, though. An extra 10% on nothing is still nothing.

 

[Zach_01]

Some undervolt can raise clocks within the same power level.
Voltage slider in drivers, at least for RDNA3, is a kind of an offset curve setting close to what CurveOptimizer is for Ryzen.
And what (UV) can work with 1 game doesn't mean will work on another with different type load, just keep in mind

 

[tpa-pr]

FWIW when testing my Pulse 7700 XT (first mention in this thread??) it can boost to the same but as some have mentioned, it's load and power dependent. Ya also gotta set the Min and Max (OK and voltage) in Adrenalin properly too.

View attachment 333377

This is multiple runs of the Forza Horizon 4 benchmark, 3174 MHz is during the last run and it still doesn't hit max power during most of the run and is above 3100 for the whole thing. The 4th from last one there is also over 3000, I was just ramping up at that point before testing some lower power settings.

On the other hand, CP2077 and Hogwarts Legacy only get to ~2800 with the same settings, pegged at max 262W power use. So very load-dependent. Seems like all the chiplet 7xxx GPUs perform similarly. I wonder if AMD was targetting 3000-3200 MHz and ended up surprised by excessive power usage in DX12 or higher loads so dialed things back.

Interesting because again, Reddit informed me that 3Ghz was a pipedream. At this point I have to assume they were either talking about non-AIB cards or (more likely) they were just dead wrong.

 

[AusWolf]

Some undervolt can raise clocks within the same power level.
Voltage slider in drivers, at least for RDNA3, is a kind of an offset curve setting close to what CurveOptimizer is for Ryzen.
And what (UV) can work with 1 game doesn't mean will work on another with different type load, just keep in mind

Oh no, that's way too much rocket science for me!

To be honest, I'm only curious what the card has left in it without too much messing around. If it gives me some extra stability in Avatar: FoP, that'll be a nice bonus.

 

[remekra]

Interesting because again, Reddit informed me that 3Ghz was a pipedream. At this point I have to assume they were either talking about non-AIB cards or (more likely) they were just dead wrong.

I mean it depends. If we are talking about stock so limited to 360W then yeah reaching that 3GHz might be harder or impossible. Even on my Nitro+ with PL raised to 460W and UV of 1085mv and max clocks of 3200MHz it rarely reaches 3GHz. In most games sits between 2700-2900.

 

[Zach_01]

Done some simple tests with Superposition benchmark regarding UV and PowerLimit.
I cant test the card with max power level (460+) because of my 750W PSU so I did it at 406W max (default PL on Primary BIOS)

3 runs, 10+min each, with 2 different voltage levels and 2 different power levels.

1st run


Result:
Avg voltage: 0.964v
Avg power: 402W
Avg FPS: 108

Highlighted some interesting points



-------------------------------------------------

2nd run
Reduced voltage level from 1150 >> 1050
Same PL

Result:
Avg voltage: 0.940v (-24mv)
Avg power: 402W
Avg FPS: 114 (+5.5%)



-------------------------------------------------

3rd run
Reduced voltage level from 1150 >> 1050
Reduced power and trying to much FPS from 1st run

Result against 1st run
Avg voltage: 0.897v (-67mv)
Avg power: 362W (-10%)
Avg FPS: 111.5 (+3%)



--------------------------------------------------

This is an indication of how much the GPU is pushed to the edge and over its efficiency curve "just" for a small percentage of performance.
With -100mv on the votage slider GPU sustained +200MHz on clocks and 5+% on FPS
Additionally with -10% power limit GPU still sustained +100MHz on clocks and 3+% on FPS, let alone the slightly better thermals and acoustics.

I should point out also that this will not necessarily reflect exact game performance and conditions in all games.
This was a "steady environment" with no movement at all and not in the super high load variation that games have.

 

[QuietBob]

In my testing there seems to be a big discrepancy in the maximum reported GPU clock between different tools. For instance, my stock MBA 7900XTX regularly goes over 3 GHz and even approaches 3200 MHz in some games, according to HWinfo's Effective Front End Clock.

But in these same games, tested with the same settings, max recorded clock rarely exceeds 2900 MHz when viewing CapFrameX logs

 

[Vya Domus]

But in these same games, tested with the same settings, max recorded clock rarely exceeds 2900 MHz

Because that's the stock frequency limit on a 7900XTX.

AMD's frequency limit doesn't work like people think it does, it's literally the same as a power limit, momentarily allowing the GPU to exceed it and then paring back the clocks. The GPU always goes up to the highest possible frequency until it hits some imposed limit. This is not the same as Nvidia, where they use a fixed frequency/voltage curve and depending on the limits the core falls somewhere along that curve, that's also why there is no frequency limit with those cards because the clocks are always bounded by the curve.

As far as I can tell most programs including AMD's own software display the effective shader clock which usually reports the lowest clocks of all and I think that's also what the "max frequency limit" keeps track of. I think I posted on this thread a screenshot of my 7900XT hitting 3.2 Ghz on the shader clock with an unstable undervolt, the only limit in practice these cards have is the power limit.

 

[Zach_01]

In my testing there seems to be a big discrepancy in the maximum reported GPU clock between different tools. For instance, my stock MBA 7900XTX regularly goes over 3 GHz and even approaches 3200 MHz in some games, according to HWinfo's Effective Front End Clock.

But in these same games, tested with the same settings, max recorded clock rarely exceeds 2900 MHz when viewing CapFrameX logs

Are you seeing the same front end clock?
If you are seeing a "GPU clock" then its the shader clock and not the front end.
What you see in drivers also is shader clock.

Is shader clock on HWiNFO matching you other software reading?

 

[kapone32]

Because that's the stock frequency limit on a 7900XTX.

AMD's frequency limit doesn't work like people think it does, it's literally the same as a power limit, momentarily allowing the GPU to exceed it and then paring back the clocks. The GPU always goes up to the highest possible frequency until it hits some imposed limit. This is not the same as Nvidia, where they use a fixed frequency/voltage curve and depending on the limits the core falls somewhere along that curve, that's also why there is no frequency limit with those cards because the clocks are always bounded by the curve.

As far as I can tell most programs including AMD's own software display the effective shader clock which usually reports the lowest clocks of all and I think that's also what the "max frequency limit" keeps track of. I think I posted on this thread a screenshot of my 7900XT hitting 3.2 Ghz on the shader clock with an unstable undervolt, the only limit in practice these cards have is the power limit.

I have noticed that Console ports run at a consistent clock speed but lower than some older Games. I have seen high boost clocks on older optimized Games like Sleeping Dogs. The thing is these cards are more powerful than we give them credit for. I when I started playing AMS2 I noticed that the clock was like 1580 as I had Vsync on and 144 Hz was the same FPS.

 

[GamerGuy]

Oops, wrong thread, sorry 'bout that!

 

[AusWolf]

Because that's the stock frequency limit on a 7900XTX.

AMD's frequency limit doesn't work like people think it does, it's literally the same as a power limit, momentarily allowing the GPU to exceed it and then paring back the clocks. The GPU always goes up to the highest possible frequency until it hits some imposed limit. This is not the same as Nvidia, where they use a fixed frequency/voltage curve and depending on the limits the core falls somewhere along that curve, that's also why there is no frequency limit with those cards because the clocks are always bounded by the curve.

As far as I can tell most programs including AMD's own software display the effective shader clock which usually reports the lowest clocks of all and I think that's also what the "max frequency limit" keeps track of. I think I posted on this thread a screenshot of my 7900XT hitting 3.2 Ghz on the shader clock with an unstable undervolt, the only limit in practice these cards have is the power limit.

This!

In short, Nvidia operates at the top of the voltage-frequency curve, unless the card hits a power or thermal limit. AMD, on the other hand, operates at a power limit, unless the card hits a thermal wall or it maxes out the allowed voltage-frequency limits/offsets.

That's why Nvidia cards operate at a relatively constant clock speed at various power consumption levels within the power and thermal headroom. AMD cards operate at maximum power consumption with more fluctuating clock speeds depending on the load.