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

 

[marees]

What if AMD is maintaining Radion silence on RDNA 4 because they don't want to osborne the RDNA 3 cards in the $500 range

What do they need to do to clear stocks of 7800xt / 7900gre / 7900xt ??

Osborne effect - Wikipedia

en.wikipedia.org

 

[blckgrffn]

My $.02 is that just isn’t top priority right now.

I don’t know what is, but whenever it happens if it’s only the one SKU all RDNA 3 below 7900xt gets pushed down $50 until full stack launch?

Who knows if they are trying to save it for a Q4 or Q1 bump or what, the motivation might be keyed to compensation for someone as much as anything else.

 

[gaav87]

I think they rushed ZEN5 as they needed more capacity with TSMC for RDNA4 as they are both on N4P.

 

[marees]

I think they rushed ZEN5 as they needed more capacity with TSMC for RDNA4 as they are both on N4P.

How long does it take for TSMC to supply if AMD places order for one batch of n48

 

[SolidQ]

There rumor from mlid, RDNA3 has stopped producing

 

[marees]

There rumor from mlid, RDNA3 has stopped producing

RDNA 2 (7nm) production was stopped 2 years ago !!!!

 

[marees]

My $.02 is that just isn’t top priority right now.

I don’t know what is, but whenever it happens if it’s only the one SKU all RDNA 3 below 7900xt gets pushed down $50 until full stack launch?

Who knows if they are trying to save it for a Q4 or Q1 bump or what, the motivation might be keyed to compensation for someone as much as anything else.

If battlemage is a success then surely some heads are going to roll in AMD

 

[linkgoron]

If battlemage is a success then surely some heads are going to roll in AMD

View attachment 106760

I'm sure everyone at AMD's GPU division got huge raises and promotions after the great successes of RDNA3 and RDNA4.

 

[adroc_thurston]

I think they rushed ZEN5 as they needed more capacity with TSMC for RDNA4 as they are both on N4P.

Wafers aren't the limiting factor at all.

If battlemage is a success then surely some heads are going to roll in AMD

LNL has more $ and more membw, matters somewhat for iGPs.
Not apples to apples stuff.

 

[blckgrffn]

I'm sure everyone at AMD's GPU division got huge raises and promotions after the great successes of RDNA3 and RDNA4.

Like Nvidia, it feels like consumer GPU is a secondary - or tertiary - concern. Money is made on AI acceleration these days. It’s way more important to ensure they are relevant in that market than to do more than maintain status quo in client GPU. - some manager at AMD, probably.

 

[Mahboi]

If battlemage is a success then surely some heads are going to roll in AMD

View attachment 106760

I am withholding every bit of faith until it's out.
Intel claiming to go from barely usable ACM to BMG somehow diffing RDNA 3.5 seems like a fat load of "believe me bruh".
We'll see the actual perf when it's out, and the state of the drivers too. I'm expecting big oofs.

 

[SolidQ]

today announced that Keith Strier has joined the company as senior vice president of global AI markets

AMD Appoints AI Industry Veteran Keith Strier to Expand Global AI Capabilities and Engagements

Browse AMD’s company-wide and financial press releases.

ir.amd.com

gonna be NV style

 

[ToTTenTranz]

RDNA 2 (7nm) production was stopped 2 years ago !!!!

That's pretty much impossible considering there are RDNA2 in automotive products like the Tesla Infotainment and those have supply contracts of like 10 years for repairs.

 

[marees]

That's pretty much impossible considering there are RDNA2 in automotive products like the Tesla Infotainment and those have supply contracts of like 10 years for repairs.

Afaik TSMC wants to move 7nm customers to 6nm

So maybe AMD might introduce a version of w6600 pro on 6nm for Tesla 🤔

https://videocardz.com/newz/tesla-car-computer-features-zen-ryzen-embedded-apu-and-discrete-navi-23-gpu

 

[Aapje]

That's pretty much impossible considering there are RDNA2 in automotive products like the Tesla Infotainment and those have supply contracts of like 10 years for repairs.

It's quite common for companies to just put stock on the shelves for repairs and shut down the production pipelines. That's one of the reasons why you can find NOS-products long after production ends.

 

[blckgrffn]

It's quite common for companies to just put stock on the shelves for repairs and shut down the production pipelines. That's one of the reasons why you can find NOS-products long after production ends.

Yeah, but… IDK in this case. Like embedded Ryzen it seems that AMD is promising availability of the product for that period of time. Even if it’s continued production in small scale there are chips being fabbed on older processes all the time. Isn’t that one of the secondary booming markets since those larger lithographies are very stable?

These aren’t fenders on discontinued models and they could see a big spike in demand in the future. Stopping making them altogether when they could maintain a healthy supply chain seems unlikely to me.

 

[jpiniero]

You can fab an N7 design on N6 with minimal changes, although you don't get the shrink benefits.

 

[marees]

Looking at NewEgg there is $150 gap between the 7900gre & 7900xt

That is a significant hole in the segment of the market which is selling big now

If only AMD had a $600 card with RT of 4070 ti or 4070 ti super that they could launch 🤔

Visual illustration of the point, that AMD needs a new high ASP card (if it has one ready, that is !!!!!)

https://twitter.com/x/status/1832743185417744634

A few extracts:

drop in sales has hit AMD much harder than nVidia: AMD lost 9290 units in sales compared to the previous quarter, while nVidia only lost 3540 units. This effectively reversed the market share in terms of units within just one quarter: 53.9% to 45.0% in the first quarter became 46.9% to 52.0% in the second quarter.

AMD is now losing more significantly in terms of ASPs, while nVidia is actually gaining again.

AMD fell by -43% in just one quarter, while nVidia managed to limit its losses to -13%.

AMD is still dependent on RDNA2 GPUs to a significant extent in some market segments - and as soon as these(old RDNA 2 stocks) are missing one day, AMD sales will probably decline even further.

AMD's RDNA3 generation lost even more noticeably than AMD's RDNA2 generation - which is actually unhealthy, because the latter is likely to leave the market soon

Grafikkarten-Verkaufsstatistik Mindfactory Q2/2024 | 3DCenter.org

Sonntag, 8. September 2024 / von Leonidas Mit dem abgelaufenen zweiten Jahresquartal läßt sich eine neue Grafikkarten-Verkaufsstatistik der diesbezüglichen Absätze & Umsätze des deutschen Einzelhändlers 'Mindfactory'

m-3dcenter-org.translate.goog

In summary:
The graphical analysis of these figures makes it clearer how much AMD is currently (still) dependent on RDNA2 sales to keep up with nVidia. But this has by no means always been the case, because just two quarters back in the final quarter of 2023, RDNA3 and Ada Lovelace were effectively on par in terms of Mindfactory sales figures

While RDNA2 sales were still a useful extra income for AMD six months ago, they are now downright dependent on them in order to achieve somewhat similar figures to nVidia in the overall calculation.

this will cause problems for AMD as soon as RDNA2 really goes off the market: Then it will become clear that the RDNA3 generation cannot keep up with previous AMD generations in terms of business either. The currently still notable RDNA2 sales have simply glossed over this point for quite some time.

 

[marees]

Visual illustration of the point, that AMD needs a new high ASP card (if it has one ready, that is !!!!!)

https://twitter.com/x/status/1832743185417744634

A few extracts:

drop in sales has hit AMD much harder than nVidia: AMD lost 9290 units in sales compared to the previous quarter, while nVidia only lost 3540 units. This effectively reversed the market share in terms of units within just one quarter: 53.9% to 45.0% in the first quarter became 46.9% to 52.0% in the second quarter.

AMD is now losing more significantly in terms of ASPs, while nVidia is actually gaining again.

AMD fell by -43% in just one quarter, while nVidia managed to limit its losses to -13%.

AMD is still dependent on RDNA2 GPUs to a significant extent in some market segments - and as soon as these(old RDNA 2 stocks) are missing one day, AMD sales will probably decline even further.

AMD's RDNA3 generation lost even more noticeably than AMD's RDNA2 generation - which is actually unhealthy, because the latter is likely to leave the market soon

Grafikkarten-Verkaufsstatistik Mindfactory Q2/2024 | 3DCenter.org

Sonntag, 8. September 2024 / von Leonidas Mit dem abgelaufenen zweiten Jahresquartal läßt sich eine neue Grafikkarten-Verkaufsstatistik der diesbezüglichen Absätze & Umsätze des deutschen Einzelhändlers 'Mindfactory'

m-3dcenter-org.translate.goog

In summary:
The graphical analysis of these figures makes it clearer how much AMD is currently (still) dependent on RDNA2 sales to keep up with nVidia. But this has by no means always been the case, because just two quarters back in the final quarter of 2023, RDNA3 and Ada Lovelace were effectively on par in terms of Mindfactory sales figures

While RDNA2 sales were still a useful extra income for AMD six months ago, they are now downright dependent on them in order to achieve somewhat similar figures to nVidia in the overall calculation.

this will cause problems for AMD as soon as RDNA2 really goes off the market: Then it will become clear that the RDNA3 generation cannot keep up with previous AMD generations in terms of business either. The currently still notable RDNA2 sales have simply glossed over this point for quite some time.

Quarter sales summary image, for those who can't access twitter

 

[marees]

Visual illustration of the point, that AMD needs a new high ASP card (if it has one ready, that is !!!!!)

https://twitter.com/x/status/1832743185417744634

A few extracts:

drop in sales has hit AMD much harder than nVidia: AMD lost 9290 units in sales compared to the previous quarter, while nVidia only lost 3540 units. This effectively reversed the market share in terms of units within just one quarter: 53.9% to 45.0% in the first quarter became 46.9% to 52.0% in the second quarter.

AMD is now losing more significantly in terms of ASPs, while nVidia is actually gaining again.

AMD fell by -43% in just one quarter, while nVidia managed to limit its losses to -13%.

AMD is still dependent on RDNA2 GPUs to a significant extent in some market segments - and as soon as these(old RDNA 2 stocks) are missing one day, AMD sales will probably decline even further.

AMD's RDNA3 generation lost even more noticeably than AMD's RDNA2 generation - which is actually unhealthy, because the latter is likely to leave the market soon

Grafikkarten-Verkaufsstatistik Mindfactory Q2/2024 | 3DCenter.org

Sonntag, 8. September 2024 / von Leonidas Mit dem abgelaufenen zweiten Jahresquartal läßt sich eine neue Grafikkarten-Verkaufsstatistik der diesbezüglichen Absätze & Umsätze des deutschen Einzelhändlers 'Mindfactory'

m-3dcenter-org.translate.goog

In summary:
The graphical analysis of these figures makes it clearer how much AMD is currently (still) dependent on RDNA2 sales to keep up with nVidia. But this has by no means always been the case, because just two quarters back in the final quarter of 2023, RDNA3 and Ada Lovelace were effectively on par in terms of Mindfactory sales figures

While RDNA2 sales were still a useful extra income for AMD six months ago, they are now downright dependent on them in order to achieve somewhat similar figures to nVidia in the overall calculation.

this will cause problems for AMD as soon as RDNA2 really goes off the market: Then it will become clear that the RDNA3 generation cannot keep up with previous AMD generations in terms of business either. The currently still notable RDNA2 sales have simply glossed over this point for quite some time.

Coincidentally the RX 6800 stocks seem to be over in NewEgg

Next step from AMD should be heavy discount of 7700xt, 7800xt & 7900 GRE cards 🤔

 

[SolidQ]

So many intel IGPU's

 

[marees]

So many intel IGPU's

I believe only lunar lake / battlemage is useful.
They should split Intel by graphics gen

 

[DokiDoki]

AMD deprioritizing flagship gaming GPUs: Jack Huynh talks new strategy against Nvidia in gaming market

The battle seems to be over before it starts.

www.tomshardware.com

A couple of years later you will get a repeat of this same headline.

Only then the word 'flagship' will be missing and 'Intel' will get added to it.

 

[marees]

AMD deprioritizing flagship gaming GPUs: Jack Huynh talks new strategy against Nvidia in gaming market

The battle seems to be over before it starts.

www.tomshardware.com

A couple of years later you will get a repeat of this same headline.

Only then the word 'flagship' will be missing and 'Intel' will get added to it.

To re-use mahboiiiiii's meme

 

[branch_suggestion]

AMD deprioritizing flagship gaming GPUs: Jack Huynh talks new strategy against Nvidia in gaming market

The battle seems to be over before it starts.

www.tomshardware.com

A couple of years later you will get a repeat of this same headline.

Only then the word 'flagship' will be missing and 'Intel' will get added to it.

It is truly amusing that you think AMD is in dire straits. These interviews are realpolitik, nothing more.
NV could price AMD out of the market completely right now, but they know leaving openings for scraps will keep the FTC off their case.
Intel did the exact same thing during Netburst, except they had the outright inferior product.
What happened is simple, RDNA3 was ~20% weaker than anticipated, AMD lost the momentum they had built with RDNA2, and now they have to start the ball rolling all over again.
AMD has a roadmap and a strategy and has proven performance for said strategy established. So the most baffling thing is where you think Intel is suddenly going to gain momentum from, considering client GPU is basically all on TSMC, and FCS is a single product for now with too much mystery still surrounding it, likely coming out when the market corrects itself and hyperscalers have to slow down CapEx.
If you cannot see the forest for the trees, so be it, but please worry about your camp not imploding into little pieces before telling others they are down bad.