Question Speculation: RDNA2 + CDNA Architectures thread

[uzzi38]

All die sizes are within 5mm^2. The poster here has been right on some things in the past afaik, and to his credit was the first to saying 505mm^2 for Navi21, which other people have backed up. Even still though, take the following with a pich of salt.

Navi21 - 505mm^2

Navi22 - 340mm^2

Navi23 - 240mm^2

Source is the following post: https://www.ptt.cc/bbs/PC_Shopping/M.1588075782.A.C1E.html

 

[RnR_au]

Why is corporate communications so hard for AMD? Consumers don't give a hoot about 'code paths'. They are only concerned about features.

They also kinda mangled their corp language when they provided an update to Toms Hardware when they mentioned game optimisation is still coming 'as required by market needs'. Not sure why they felt the need to add that part. Its always been up to AMD to decide which games gets supported at the driver level... its a weird thing to say.

 

[SPBHM]

But what is the real impact?
It shouldn't be to hard for GPU reviewers to test it.

Grab a 5700xt and install a driver from 3 years ago and the most recent one. Test some games released after that older driver, especially games that had game ready drivers and then show the impact.

it's far from that simple because in general there is also a game side impact, when a product is no longer supported the games also don't care about it, the other software also moves on from being any type of optimized for the old stuff

 

[GaiaHunter]

it's far from that simple because in general there is also a game side impact, when a product is no longer supported the games also don't care about it, the other software also moves on from being any type of optimized for the old stuff

It is more likely that newer GPUs either don't work or have poor performance in old games than newer games not using old hardware to their best capability.

And anyone can see that with Intel ARC GPUs, old games using hardware physX being slower in newer NVIDIA GPUs, GoG work on games to allow it use new hardware and OSes, etc.

I'm waiting for proof of old GPUs being more performant in new games using new drivers as opposed to drivers from say 2-3 years after their release.

 

[GaiaHunter]

Sure looks like they care.

https://www.amd.com/en/blogs/2025/continued-support-for-every-radeon-gamer.html

They are saying "we are still doing what we said we were going to do but look it is in a nice wrapper".

What they should have done, from their perspective, is saynothing. People wouldn't notice it because it has no effect on them.

This is the case of people punishing transparency, which shouldn't be rocket science as it is exactly the same as in other areas of society. People don't want the truth, they want to believe companies are still working hard on their completly mature products and those products still have untapped performance...

 

[Vikv1918]

People don't want the truth, they want to believe companies are still working hard on their completly mature products and those products still have untapped performance...

Not untapped performance, I think thats almost maxed out. But FSR4 would be nice.

 

[ToTTenTranz]

this is all PRese for "we're moving non-WMMA GPUs into legacy"?

Which is completely different from their initial "we're only doing security and stability updates for RDNA2 from now on" message.

What they should have done, from their perspective, is saynothing. People wouldn't notice it because it has no effect on them.

People noticed. Which is why they backtracked.

People don't want the truth, they want to believe companies are still working hard on their completly mature products and those products still have untapped performance...

They only need to say the INT8 version of FSR4 isn't coming to RDNA2 because the performance overhead is so big their users should just stick to FSR3.1 at higher base resolutions. Which is observably true, so it isn't a bullsh excuse like the one Nvidia used to justify not having framegen on RTX30, or >2x framegen on RTX40.

 

[Vikv1918]

They only need to say the INT8 version of FSR4 isn't coming to RDNA2 because the performance overhead is so big their users should just stick to FSR3.1 at higher base resolutions. Which is observably true, so it isn't a bullsh excuse like the one Nvidia used to justify not having framegen on RTX30, or >2x framegen on RTX40.

Its absolutely a BS excuse. AMD doesnt care about the glaring artifacts and ghosting that FSR2/3 still has years after launch, but I'm supposed to believe that FSR4 being 10% slower is a reason to not release it?

It hasnt been released because they want people to upgrade from RDNA1/2/3 to 4. A significant chunk of Radeon buyers upgrade from Radeon to Radeon, so they want to upsell those people.

 

[jpiniero]

Which is completely different from their initial "we're only doing security and stability updates for RDNA2 from now on" message.

They never said that. They said that games would continue to work and not much more. And (IMO) that hasn't changed.

 

[adroc_thurston]

Which is completely different from their initial "we're only doing security and stability updates for RDNA2 from now on" message.

It's the same thing.
Your corporatese is lacking.

 

[GaiaHunter]

People noticed. Which is why they backtracked.

People didn't notice anything about their drivers - they were told. And it is not backtracked, just reworded.

And people were getting ready to be outraged about the usb c port on the 7900 xtx losing features until someone actually tested that was still working and it seems(?) a previous driver had already had those notes.

 

[poke01]

people were getting ready to be outraged about the usb c port on the 7900 xtx losing features until someone actually tested that was still working and it seems(?) a previous driver had already had those notes.

Funny cause some told that usb c feature was disabled much earlier and no one noticed .

 

[coercitiv]

Funny cause some told that usb c feature was disabled much earlier and no one noticed .

Funny because AMD stated the USB C feature was not disabled in the first place.

We’d like to inform you that the release notes for AMD Software Adrenalin Edition 25.10.2 posted today included misinformation that has since been corrected. There is no change to USB-C functionality on the RX 7900 series GPUs in the 25.10.2 driver. There was an incorrect line in the originally posted release notes that has been removed, and the release notes have been updated.

 

[poke01]

Funny because AMD stated the USB C feature was not disabled in the first place.

Then looks like those people were lying, ahh reddit

 

[gdansk]

I have a 7900 XTX with USB C but I never use that. I cannot tell you if it works or doesn't.
But I paid for it. It should work. What's weirder is that AMD didn't know if it worked. Apparently.

 

[GaiaHunter]

I have a 7900 XTX with USB C but I never use that. I cannot tell you if it works or doesn't.
But I paid for it. It should work. What's weirder is that AMD didn't know if it worked. Apparently.

I wonder if at some point there was some bug and they briefly disabled it or planned to and this is copy/pasta.

 

[marees]

"Easier to build new hardware than fix issues with existing hardware"

https://twitter.com/x/status/1985545276917010483

 

[marees]

Comment by Ryan Smith

(Move must have been planned for a long time.
Communication should have been better. Should have remained ahead / pro-active)

https://twitter.com/x/status/1985549643644878963

 

[Vikv1918]

Tom's hardware edited their article yesterday with this addition :

Edit 11/3/2025: To be clear, AMD gave Tom's Hardware both an on-the-record verbal statement clarifying that day-zero support would be enabled, and also provided the following written statement.]

"New features, bug fixes and game optimizations will continue to be delivered as required by market needs in the maintenance mode branch," an AMD spokesperson told Tom's Hardware.

So Toms clearly got verbal communication from AMD that day zero support would be available, but they chose not to print that in the original article, only printing the written statement which is much more unclear.

So this is bad communication from AMD, and bad journalism from the tech press. I think AMD should openly make this point clear and publicly refute the tech press who now claim "backtracking" happened as a result of their "journalism".
And while they're at it, they should fire whoever in marketing or driver communications is responsible for this mess.

EDIT: Take a look at it this tweet and the replies

https://twitter.com/x/status/1985459928879022510

HWU : "we totally made the big evil company backtrack"

Repliers who are totally not shills : "Wow lost all trust in AMD, will never buy them will only buy nvidia"

 

[coercitiv]

So this is bad communication from AMD, and bad journalism from the tech press. I think AMD should openly make this point clear and publicly refute the tech press who now claim "backtracking" happened as a result of their "journalism".

But AMD totally backtracked. The only thing that Toms failed to convey properly was how much they backtracked.

Here's what started all of this, a line in their drier release notes.

New Game Support and Expanded Vulkan Extensions Support is available to Radeon RX 7000 and 9000 series graphics products.

This is how they let us know they were altering support for older RDNA versions, by announcing "New Game Support" is available to Radeon RX 7000 and 9000 series. This statement was followed by clarifications offered to PCGamesHardware.de which only guaranteed security and bug fixes for RDNA 1&2. AFAIK "Maintenance mode" is a term is yet to be defined clearly, as it can mean many things and can also change meaning over time (definition is essentially up to AMD).

RDNA 1 and RDNA 2 graphics cards will continue to receive driver updates for critical security and bug fixes. To focus on optimizing and delivering new and improved technologies for the latest GPUs, AMD Software Adrenalin Edition 25.10.2 is placing Radeon RX 5000 and RX 6000 series graphics cards (RDNA 1 and RDNA 2) into maintenance mode. Future driver updates with targeted game optimizations will focus on RDNA 3 and RDNA 4 GPUs.

Spoiler: Original in German

RDNA-1- und RDNA-2-Grafikkarten werden weiterhin Treiber-Updates für kritische Sicherheits- und Fehlerbehebungen erhalten. Um sich auf die Optimierung und Bereitstellung neuer und verbesserter Technologien für die neuesten GPUs zu konzentrieren, versetzt AMD Software Adrenalin Edition 25.10.2 Grafikkarten der Serien Radeon RX 5000 und RX 6000 (RDNA 1 und RDNA 2) in den Wartungsmodus. Zukünftige Treiber-Updates mit gezielten Spieloptimierungen werden sich auf RDNA-3- und RDNA-4-GPUs konzentrieren.

Since the clarification clarified very little, AMD followed with another longer clarification that contained the following quote:

Your Radeon RX 5000 and RX 6000 series GPUs will continue to receive:

• Game support for new releases
• Stability and game optimizations
• Security and bug fixes

followed by another segment where they take a very winding path to explain they will stop adding new features for RDNA 1&2 (unless they decide otherwise):

The difference is that these products now benefit from a dedicated, stable driver branch, one built on years of tuning and optimization. This approach helps deliver a smoother, more consistent experience for your games while insulating previous generation GPUs from rapid changes designed for newer architectures.

Our goal is simple: to give every Radeon gamer the best experience possible. By separating the code paths, our engineers can move faster with new features for RDNA 3 and RDNA 4, while keeping RDNA 1 and RDNA 2 stable and optimized for current and future games.

So let's compare some statements:

• Driver release notes - New Game Support and Expanded Vulkan Extensions Support is available to Radeon RX 7000 and 9000 series graphics products.
• Second official statement - Your Radeon RX 5000 and RX 6000 series GPUs will continue to receive New Game Support. (paraphrased, see above exact quote)

Backtracking: 100%

• First official statement: Future driver updates with targeted game optimizations will focus on RDNA 3 and RDNA 4 GPUs.
• Second official statement: Your Radeon RX 5000 and RX 6000 series GPUs will continue to receive stability and game optimizations.

Backtracking: 75%


What likely happened with Tom's Hardware is they got information from two sources or via two channels:

• Source 1: New features, bug fixes and game optimizations will continue to be delivered as required by market needs in the maintenance mode branch
• Source 2: Day-zero support will be enabled for RDNA 1&2.

Even if Tom's failed to report the second quote, the fact that the two statements from AMD are not fully compatible with each-other shows how non-committal AMD was during these days. We can blame journalists for many things nowadays, but in this case the damage is 100% self-inflicted by AMD who didn't even know how to backtrack properly.

 

[ToTTenTranz]

Its absolutely a BS excuse. AMD doesnt care about the glaring artifacts and ghosting that FSR2/3 still has years after launch, but I'm supposed to believe that FSR4 being 10% slower is a reason to not release it?

It's not a BS excuse, it's the truth. 99.9% of the people are better off using FSR3.1 Quality at higher framerates than FSR4 Ultra Performance at lower framerates. And it's not a 10% difference either.
RDNA2 is just not adequate for FSR4 because the shader units can't do WMMA, period.
Ghosting problems are usually a bad implementation of motion vectors, not a limitation of the upscaler nor the hardware. Artifacts can also be mitigated, to the point where most people won't even notice they're there, especially if it's in motion.

I played on a 6900XT for almost 5 years until August which is when I upgraded to a RTX 5090 so I went from FSR3 to DLSS4 overnight. And I still turn FSR 3 on when I play on my Legion Go.
DLSS3/4 isn't that much better than a decent FSR 3.1 implementation. All those publications that keep dissing FSR3 are IMO doing the world a disservice by putting a spotlight on differences that can only be noticed when they freeze a frame and do a 5x zoom.

It hasnt been released because they want people to upgrade from RDNA1/2/3 to 4. A significant chunk of Radeon buyers upgrade from Radeon to Radeon, so they want to upsell those people.

A significant chunk of Radeon buyers are buying RDNA2 GPUs today, specifically iGPUs. AMD would've loved to tell all their brand new Z2 Go and Z2A clients they can use FSR4. So no, upselling is not the main reason for not getting FSR4 on RDNA2.

People didn't notice anything about their drivers - they were told.

Hence they noticed. That's why outlets like Gamersnexus and Hardware Unboxed have an increasingly large audience. Because for better or worse, they trust what the outlet says.

Then looks like those people were lying, ahh reddit

No one lied. AMD's release notes for the 25.10.2 driver had a line under "Important Information" that said:
USB-C power charging has been disabled for Radeon™ RX 7900 series graphics products. Users requiring this feature are recommended to use AMD Software: Adrenalin Edition version 25.3.1.
AMD has since said the line was there by mistake and erased it from their webiste.

It's the same thing.

No, it's not. This is a strange hill for you to die on, as the evidence that they backtracked is unsurmountable.

AMD's statement to PCGH clearly said only RDNA3 and 4 would be getting game improvements and RDNA1 and 2 would only be getting "critical security and bug fixes":

RDNA-1 and RDNA-2 graphics cards will continue to receive driver updates for critical security and bug fixes. To focus on optimizing and delivering new and improved technologies for the latest GPUs, AMD Software Adrenalin Edition 25.10.2 puts Radeon RX 5000 and RX 6000 and RX 6000 and RDNA 2 graphics cards into maintenance mode. Future driver updates with targeted game optimizations will focus on RDNA-3 and RDNA-4 GPUs

Yesterday, AMD reached out to HUB and told them RDNA1 and 2 would be getting game optimizations at the same time as the others:

𝙔𝙚𝙨, 𝙜𝙖𝙢𝙚 𝙤𝙥𝙩𝙞𝙢𝙞𝙯𝙖𝙩𝙞𝙤𝙣𝙨 𝙖𝙣𝙙 𝙨𝙪𝙥𝙥𝙤𝙧𝙩 𝙛𝙤𝙧 𝙖𝙡𝙡 𝙍𝘿𝙉𝘼 𝙎𝙚𝙧𝙞𝙚𝙨 1 𝙩𝙝𝙧𝙤𝙪𝙜𝙝 4 𝙬𝙞𝙡𝙡 𝙧𝙤𝙡𝙡 𝙤𝙪𝙩 𝙖𝙩 𝙩𝙝𝙚 𝙨𝙖𝙢𝙚 𝙩𝙞𝙢𝙚 𝙞𝙣 𝙗𝙤𝙩𝙝 𝙙𝙧𝙞𝙫𝙚𝙧 𝙥𝙖𝙘𝙠𝙖𝙜𝙚𝙨 – 𝙞𝙣𝙘𝙡𝙪𝙙𝙞𝙣𝙜 𝙗𝙪𝙩 𝙣𝙤𝙩 𝙡𝙞𝙢𝙞𝙩𝙚𝙙 𝙩𝙤 𝘾𝙤𝘿, 𝘾𝙧𝙞𝙢𝙨𝙤𝙣 𝘿𝙚𝙨𝙚𝙧𝙩, 𝙖𝙣𝙙 𝙍𝙚𝙨𝙞𝙙𝙚𝙣𝙩 𝙀𝙫𝙞𝙡.

So yes. They backtracked. No, it's not that people got confused with "PRese" or "corporatese". When a corporation says "RDNA1 and 2 will receive updates for critical security and bugs", everyone will assume they'll receive updates for critical security and bugs". This isn't a lack of understanding on people's side, it's the corporation doing a 180º on attempting to save costs in driver development and then trying to gaslight the people who called them out on it. Calling it a "confusion" was honestly kind of insulting.

 

[jpiniero]

So yes. They backtracked.

They did not at all.

See the line about targeted game optimizations.

 

[ToTTenTranz]

They did not at all.

They did.

See the line about targeted game optimizations.

October 30 morning in Germany (i.e. during the 29-30 night in USA): AMD sends an e-mail to PCGH saying RDNA1-2 will only receive driver updates for critical security and bug fixes (in contrast to saying RDNA3-4 will get performance and feature updates).
October 30 in the afternoon and 31 in the morning: Hardware Unboxed, Daniel Owen, a bunch of other youtubers and sites report on AMD's decision to stop making performance improvements on RDNA2 GPUs, referencing AMD's e-mail to PCGH.
October 31 in the afternoon: AMD sends an e-mail to Tomshardware saying RDNA2 are getting "game optimizations by market demand".
November 2: AMD makes a post on their website saying RDNA1-2 will get:

• Game support for new releases
• Stability and game optimizations
• Security and bug fixes

For the PCGH e-mail they clearly only included the last point, and clearly said the first two points would only be done for RDNA3-4.

So yes. They backtracked. Very obviously so.

 

[jpiniero]

See the qualifier about targeted. They've always said they will ensure that games work, and that's all they will do.

 

[ToTTenTranz]

See the qualifier about targeted. They've always said they will ensure that games work, and that's all they will do.

The "qualifier about targeted" is absent in AMD's original response to PCGH in October 30. Whatever AMD said afterwards was made in response to the reactions from several outlets reporting on it.


This is the verbatim statement from AMD towards PCGH:

"RDNA-1- und RDNA-2-Grafikkarten werden weiterhin Treiber-Updates für kritische Sicherheits- und Fehlerbehebungen erhalten. Um sich auf die Optimierung und Bereitstellung neuer und verbesserter Technologien für die neuesten GPUs zu konzentrieren, versetzt AMD Software Adrenalin Edition 25.10.2 Grafikkarten der Serien Radeon RX 5000 und RX 6000 (RDNA 1 und RDNA 2) in den Wartungsmodus. Zukünftige Treiber-Updates mit gezielten Spieloptimierungen werden sich auf RDNA-3- und RDNA-4-GPUs konzentrieren."

Feel free to translate yourself or use whatever translation tool you like.

 

[Ranulf]

Comment by Ryan Smith

(Move must have been planned for a long time.
Communication should have been better. Should have remained ahead / pro-active)

https://twitter.com/x/status/1985549643644878963

A six year cadence means a potential customer should be wary of even rdna4 cards. Now I am glad I didn't splurge and go for a 6800xt 2.5 years ago. I guess we are back to a 2-3 year upgrade cadence for any of the brands despite the price increases. This means I care even less about various technologies these idiot companies tout as the next greatest flashy pretty goofball rays. Why spend any serious money for unoptimized slop games and ever increasing prices for overpriced hardware than can barely get it up to 60fps without rubbing vaseline on the monitor.