Discussion RDNA4 + CDNA3 Architectures Thread

[DisEnchantment]

With the GFX940 patches in full swing since first week of March, it is looking like MI300 is not far in the distant future!
Usually AMD takes around 3Qs to get the support in LLVM and amdgpu. Lately, since RDNA2 the window they push to add support for new devices is much reduced to prevent leaks.
But looking at the flurry of code in LLVM, it is a lot of commits. Maybe because US Govt is starting to prepare the SW environment for El Capitan (Maybe to avoid slow bring up situation like Frontier for example)

See here for the GFX940 specific commits

History for llvm/lib/Target/AMDGPU - llvm/llvm-project

The LLVM Project is a collection of modular and reusable compiler and toolchain technologies. - History for llvm/lib/Target/AMDGPU - llvm/llvm-project

github.com

Or Phoronix

More AMD "GFX940" Enablement Work Landing In LLVM - Phoronix

www.phoronix.com

There is a lot more if you know whom to follow in LLVM review chains (before getting merged to github), but I am not going to link AMD employees.

I am starting to think MI300 will launch around the same time like Hopper probably only a couple of months later!
Although I believe Hopper had problems not having a host CPU capable of doing PCIe 5 in the very near future therefore it might have gotten pushed back a bit until SPR and Genoa arrives later in 2022.
If PVC slips again I believe MI300 could launch before it

This is nuts, MI100/200/300 cadence is impressive.



Previous thread on CDNA2 and RDNA3 here

Question - Speculation: RDNA3 + CDNA2 Architectures Thread

Man I have been dying to make this one for a while now. First rumours for RDNA3 are here so new thread time! Just going to start off with this one for now: kopite7kimi on Twitter: "@VideoCardz Ah, I mean a simple mcm design with 10240 cores is not enough. Because the lift from RDNA2 to RDNA3...

forums.anandtech.com

 

[soresu]

Well he was right about Zen5's IPC boost.

That's not the point.

If I wanted to see his face I'd follow his channel, I don't want to see it when I come here 🤣

 

[SK10H]

Well he was right about Zen5's IPC boost.

It's whoever gave him the slides way back was right, not him.

When a month or so ago, the forum twitter guys set him up the trap which expose his professional standard presenting 1 single source as multiple to boost his credibility on the story, and if those twitter guys manage to be right, you can also claim that particular story he post was right too. I encourage more traps to be planted on these guys so there's more drama later on.

 

[Kaluan]

Could be Blackwell.

IDK man, I doubt Lisa (or whoever it was) would talk about the competition (even if indirectly and on a tangential subject).

 

[BurnItDwn]

I removed the link

 

[tajoh111]

See I think the leakers are wrong about the performance.

But it isn't necessarily about performance if the delay ends up happening. It's more about getting RDNA2/3 supply gone so AMD doesn't have to eat the losses on it. And people are going to complain about the prices no matter what AMD ends up pricing RDNA4 at.

A 10 percent performace boost on a die 240mm2, would represent a 54% performance per mm2 increase on a process that is relatively the same. That's actually really good. Something like a bit better than a 7900xt at 240mm2 would represent a 90% increase in performance per mm2. That's a miracle. That has never happend any time recently and certainly without the aid of a much better node increase.

It would be something akin to the RTX 3090 to an imaginary RTX 4090 ti which had a massively different process node improvement.

I don't know how people are getting such wild performance expectations from a 240mm2 4nm die. Huge race tracing improvement + a 90% improvement in performance per die area? That seems simply impossible. If Navi4 was such a good architecture, AMD should have just gone for a monolithic rdna4. I.E a 450mm2 die with 80-90% more performance on 4nm which would have been rediculously good.

But that's why I have doubts about the performance...Look back at my performance predictions about RDNA3... and you will see I was the only one on this forum that predicted it's performance accurately.

 

[gdansk]

Consider that even at same clocks as 7800 XT with the rumored specs (64 CU at 240mm²), it would be 54% better performance per area.

So if those are right then it's pretty likely that it has a big ppa increase.

 

[maddie]

I removed the link

Not specifically to you.

So we're censoring or self censoring now?

Why does it seem so impossible to admit that an unliked person was right about anything? What the hell is happening?

By the way being right in a complex situation with multiple alternatives is NOT 50-50. The chance of being wrong is a lot more that being right.

 

[Timorous]

Consider that even at same clocks as 7800 XT with the rumored specs (64 CU at 240mm²), it would be 54% better performance per area.

So if those are right then it's pretty likely that it has a big ppa increase.

How are you calculating that?

7800XT is 200mm + 4x36mm. Of that 200mm there are 4 GCD MCD links and of that 36mm there is a GCD MCD link, 64 bit GDDR6 bus and 16mb L3 cache.

If you remove the GCD to MCD links then that means you can reduce the total area of a monolithic design by around 30mm. So in total a monolithic N32 would be about 290mm assuming the GDDR6 bus and the 64MB L3 does not shrink at all going from the N6 MCD to the N5 GCD.

 

[gdansk]

Calculated without adjustment of any sort.

 

[marees]

The next generations of video cards from AMD and Intel will not appear until early 2025. Tweakers verified this with several sources during the Computex 2024 fair in Taipei. According to previous rumors, competitor Nvidia will soon release a new GeForce RTX 5000 series.

Officially, both manufacturers are not achieving their self-imposed goals for the new video cards. AMD's next generation of graphics cards, which uses the RDNA4 architecture and is likely to be marketed as the Radeon RX 8000 series, is expected to launch in 2024, according to the roadmap . Intel also said at the end of last year that the new generation of Arc video cards, better known under the code name Battlemage, would be released this year.

It is possible that AMD and Intel will announce their new GPUs just before the end of the year to meet the aforementioned deadlines, but an actual release at the CES fair in early January is more likely. In any case, availability will only start in the new year.

About the new AMD lineup, sources confirm to Tweakers that no real high-end models will appear in the RX 8000 series. Rumors already appeared last year that the Navi 4x GPU with the largest number of cores had been canceled. Shortly afterwards, the CEO of AMD's video card division left . The company would no longer aspire to compete with Nvidia's top models, but would mainly like to focus on the mainstream segment with cards such as the RX 7700 XT and RX 7800 XT.

'Volgende generaties AMD Radeon- en Intel Arc-gpu's verschijnen pas begin 2025'

De volgende generaties videokaarten van AMD en Intel verschijnen pas begin 2025. Dat heeft Tweakers geverifieerd bij meerdere bronnen tijdens de Computex 2024-beurs in Taipei. Volgens eerdere geruchten komt concurrent Nvidia al eerder met een nieuwe GeForce RTX 5000-reeks.

tweakers-net.translate.goog

Above From below:

Powercolor refers to the fact that the RX 8000 series would not be released until early 2025, something Tweakers was previously able to report on Computex based on several sources .

Powercolor soldeert npu op RX 7900 XTX voor tot 22 procent extra efficiëntie

Powercolor toont op Computex 2024 een prototype van een efficiëntiemodule voor de AMD RX 7900 XTX. In principe zorgt een op de videokaart gesoldeerde npu voor een preciezere stroomtoevoer, wat voor een tot 22 procent lager stroomverbruik van de gpu moet zorgen.

tweakers-net.translate.goog

Via wccftech

PowerColor's "Next-Gen" AMD Radeon GPUs To Work With NPUs To Reduce Power Consumption While Gaming

PowerColor teased a demo of its upcoming AMD Radeon RX GPUs that would utilize the power of NPUs to lower power consumption while gaming.

wccftech.com

 

[marees]

The next generations of video cards from AMD and Intel will not appear until early 2025. Tweakers verified this with several sources during the Computex 2024 fair in Taipei. According to previous rumors, competitor Nvidia will soon release a new GeForce RTX 5000 series.

Officially, both manufacturers are not achieving their self-imposed goals for the new video cards. AMD's next generation of graphics cards, which uses the RDNA4 architecture and is likely to be marketed as the Radeon RX 8000 series, is expected to launch in 2024, according to the roadmap . Intel also said at the end of last year that the new generation of Arc video cards, better known under the code name Battlemage, would be released this year.

It is possible that AMD and Intel will announce their new GPUs just before the end of the year to meet the aforementioned deadlines, but an actual release at the CES fair in early January is more likely. In any case, availability will only start in the new year.

About the new AMD lineup, sources confirm to Tweakers that no real high-end models will appear in the RX 8000 series. Rumors already appeared last year that the Navi 4x GPU with the largest number of cores had been canceled. Shortly afterwards, the CEO of AMD's video card division left . The company would no longer aspire to compete with Nvidia's top models, but would mainly like to focus on the mainstream segment with cards such as the RX 7700 XT and RX 7800 XT.

'Volgende generaties AMD Radeon- en Intel Arc-gpu's verschijnen pas begin 2025'

De volgende generaties videokaarten van AMD en Intel verschijnen pas begin 2025. Dat heeft Tweakers geverifieerd bij meerdere bronnen tijdens de Computex 2024-beurs in Taipei. Volgens eerdere geruchten komt concurrent Nvidia al eerder met een nieuwe GeForce RTX 5000-reeks.

tweakers-net.translate.goog

Above From below:

Powercolor refers to the fact that the RX 8000 series would not be released until early 2025, something Tweakers was previously able to report on Computex based on several sources .

Powercolor soldeert npu op RX 7900 XTX voor tot 22 procent extra efficiëntie

Powercolor toont op Computex 2024 een prototype van een efficiëntiemodule voor de AMD RX 7900 XTX. In principe zorgt een op de videokaart gesoldeerde npu voor een preciezere stroomtoevoer, wat voor een tot 22 procent lager stroomverbruik van de gpu moet zorgen.

tweakers-net.translate.goog

Via wccftech

PowerColor's "Next-Gen" AMD Radeon GPUs To Work With NPUs To Reduce Power Consumption While Gaming

PowerColor teased a demo of its upcoming AMD Radeon RX GPUs that would utilize the power of NPUs to lower power consumption while gaming.

wccftech.com

Btw the first article by tweakers above was tweeted 2 minutes before hassan mujtaba's tweet that RDNA 4 will not release this year

https://twitter.com/x/status/1798062177737409010

 

[marees]

Not specifically to you.

So we're censoring or self censoring now?

Why does it seem so impossible to admit that an unliked person was right about anything? What the hell is happening?

By the way being right in a complex situation with multiple alternatives is NOT 50-50. The chance of being wrong is a lot more that being right.

In general I avoid videos & prefer links because I can scan thru the information when reading. In addition if the voice or face is not tolerable then the value quotient of the video has to be extremely high for me to watch that. This is just me, ofc.

 

[marees]

yesterday Moore's Law is Dead talked a bit about RDNA 4. Likely Dec 2024, but Nov possible.

Others assert that you won't see stock on shelves until 2025

Powercolor has taken it as a fait accompli that there won't be RDNA 4 cards released this year

 

[DaaQ]

Elden Ring is a 2022 game, and Baldur's Gate 3 isn't even a full year old. They're more than valid tests.

Yea I've been playing D4, which is a 2023 game, but I can run BG3 and see what i get, I haven't spent alot of time on it since it is not my personal fav atm, but what should I use, the adrenaline, or AB plus HWinfo? Because I don't think anything I post will convice him.
I will be upgrading, but I had IRL issues regarding my grand daughters well being that took priority. Then I spent the top dollar on the biggest upgrade from a GTX 780 for the time being. I will go Z5 and 870XE w ddr5 in my own time line.

But what I have shows Vishera is not a slouch. Especially at 4.8 and 2400 mem. Ahead of it's time I think.

 

[SolidQ]

Black Myth Wukong system requirements​

No Mention AMD in RT settings. That going also more sales for RTX.
AMD need release RDNA4 earlier as possible

Spoiler

 

[branch_suggestion]

Black Myth Wukong system requirements​

No Mention AMD in RT settings. That going also more sales for RTX.
AMD need release RDNA4 earlier as possible

Spoiler

NV has effectively infinite MDF to bribe all sorts of inane nonsense.
And flood the channel as well

 

[Mahboi]

The next generations of video cards from AMD and Intel will not appear until early 2025. Tweakers verified this with several sources during the Computex 2024 fair in Taipei. According to previous rumors, competitor Nvidia will soon release a new GeForce RTX 5000 series.

Officially, both manufacturers are not achieving their self-imposed goals for the new video cards. AMD's next generation of graphics cards, which uses the RDNA4 architecture and is likely to be marketed as the Radeon RX 8000 series, is expected to launch in 2024, according to the roadmap . Intel also said at the end of last year that the new generation of Arc video cards, better known under the code name Battlemage, would be released this year.

It is possible that AMD and Intel will announce their new GPUs just before the end of the year to meet the aforementioned deadlines, but an actual release at the CES fair in early January is more likely. In any case, availability will only start in the new year.

Well now I'm getting mad.

 

[SolidQ]

to bribe all sorts of inane nonsense.

AMD also need same, and choose good games. For FSR3 presentation they chose worsts games.

Via wccftech

Interesting. Maybe that why RDNA4 is delay?

 

[eek2121]

N31 is extremely PL limited so you get high clocks in not that demanding games (including mostly CPU-limited MMOs) and in the other games and stress tests it drops to rock bottom or starts to eat upwards of 800W (check evc2se TS/TSE OC results at OCN). Either AMD aimed at 600W tdp initially or something is really broken either in some raster-related shit inside the CUs or the fixed part of the pipeline (whatever it is called now)

It should have clocked in around 3.5-3.8ghz. Instead, we got cards that flirted with 3ghz.

The 7900XTX at 3.5-3.8 ghz would have been an interesting part indeed. It likely would have edged out the 4090 in terms of raster performance.

We will see where the RDNA4 parts land. 60-64CUs at 4Ghz? That would be crazy, right? I haven’t seen anything about final clocks, but there will be an IPC uplift as well. They are also rolling out work graph support (which will help current gen as well) and optimizing dual issue (both hardware and drivers) so I don’t think this will be a boring release. Will they beat a 5090? No, but I think they will outperform expectations.

 

[Tuna-Fish]

We will see where the RDNA4 parts land. 60-64CUs at 4Ghz? That would be crazy, right? I haven’t seen anything about final clocks, but there will be an IPC uplift as well. They are also rolling out work graph support (which will help current gen as well) and optimizing dual issue (both hardware and drivers) so I don’t think this will be a boring release. Will they beat a 5090? No, but I think they will outperform expectations.

... Lets not let the hype train up to that kind of speed again. I doubt it will be anywhere near 4GHz, but even if it somehow is, there is simply not room for it to surprise upwards that much with a 256b GDDR6 memory controller.

 

[RnR_au]

... Lets not let the hype train up to that kind of speed again.

Do you even AMD...

 

[SolidQ]

Microsoft announce they're working on Next gen Xbox. So RDNA5 or 5.5?

 

[dr1337]

and if those twitter guys manage to be right, you can also claim that particular story he post was right too

Not trying to defend Most Likely In Denial, but so far the silicon twitter gang have lost pretty much all of their credibly at this point.

It would be by far the most insane thing for AMD to be sandbagging right now, so given the computex presentation I think it's very safe to say they were all wrong.

And a lot of those same people were extremely wrong about RDNA2 AND RDNA3 as well, those same people are now making interesting statements about RDNA4 and I think we can all safely ignore them.

The problem with leaker type personalities is that its kinda like gambling. There really are only so many decisions AMD is going to make, and its easy enough for people to make bets. I really wouldn't put it past MLID to have gotten lucky with his Zen 5 slide, instead of him actually having a real source. And in the same vein I wouldn't count out AMD doing another polaris type generation for RDNA4.

And really leakers pranking each other with fake leaks is annoying, its not funny as much as it is stupid. Its an odd overlap of people who read AnandTech, a site with highly professional and technical articles, care more about drama than reasonable discussions.

 

[SK10H]

Not trying to defend Most Likely In Denial, but so far the silicon twitter gang have lost pretty much all of their credibly at this point.

And a lot of those same people were extremely wrong about RDNA2 AND RDNA3 as well, those same people are now making interesting statements about RDNA4 and I think we can all safely ignore them.

And really leakers pranking each other with fake leaks is annoying, its not funny as much as it is stupid. Its an odd overlap of people who read AnandTech, a site with highly professional and technical articles, care more about drama than reasonable discussions.

That's the most likely scenario. One thing about those gang is at least they admit they are wrong instead of disappearing like smt4...😌 Not sure since I don't watch but I don't think the typical video leakers ever admit they are wrong about anything since all their bases are covered.

With no direct detail but was just basing on some presentation from oem, those guys sure are way too arrogant, and deserved to get slapped now when every poster got harassed disagreeing against them for months. Here's some tomatoes, help yourself. 🤣

Since people can't help but to keep linking the lying "journalists", might also have some fun since it's nothing professional from these guys anyway unlike Ian Johan days. It's not hard to say this story comes from 1 unreliable source and tell readers to be aware of the story, then you're released from liability as the messager. 😉

 

[tajoh111]

I think the specs of navi 48 are going to similar to the specs of a 7700xt. 240mm2 of die space does not allow for 60 CU when you add back the memory controller and the infinity cache. On navi22, the infinity cache along with the memory controllers and infinity fabric take about a third of die.... and make note this is only 192bit bus. If navi 48 is using a 256bit bus which seems to be the case with the 16GB memory leaks, something has to take a hit. I think Navi 48 is going to have 48CU. Cache and memory controllers does not shrink well and if AMD is indeed changing the cores for RDNA4 and increasing the the ray tracing performance, a toll has to paid somewhere.

You cannot add back all the components of the MCD back into the die for only 40mm2 along with make improvements to the rest of RDNA4 architecture when N4P only provides a 6% improvement to transistor density. Only a positive note, I would not be surprised to see RDNA4 clock at 3.3 or 3.4ghz.