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

 

[Mahboi]

Given Nvidia’s pricing - that shouldn’t be a problem. Need lots of marketing dollars (and a good strategy)

AMD challenge level: Hard

, best of breed hardware kits to all influencers

Easy

and really good day zero drivers.

Impossible

Interviews with top Radeon Brass to pump up the excitement would be nice.

More Naffziger and Papermaster, why not. They're nice.

 

[Mahboi]

Usually the people on forums are the loudest minority...

The ecosystem is interesting. Because you have a bunch of people who are loud and try to spread the Gospel of Jensen everywhere, but they all tend to immediately change tune as soon as their television changes tunes.
I'm pretty sure pro shills like Digital Foundry are one thing, but I wonder how many will rely on the annoyingly repetitive Tim from HWUB. It may not be as easy to pay off all the shills as adroc pretends. I suppose AMD can pay some of the most obvious ones though.

 

[Mahboi]

Tons of companies were second or third to a trend and have won big. Facebook vs Myspace, Chrome vs anything, Zoom was created much later than webex or Skype, Android vs. Windows mobile etc.

What the heck?
Those are all abysmal examples, and totally unrelated to my point too.

AMD's problem is in culture, not in the fact that they are N°2.

AMD's Zen development has been rock solid and great. Much closer to what Nvidia is doing in GPUs than what AMD is doing in GPUs. Developing a rock solid foundation for their stack top-to-bottom. Solid roadmap, solid execution, and nice niche things like Threadripper and interesting experiments like 3d cache. They also had great support and longevity with AM4 and hopefully will have longevity with AM5 as well. They single handedly broke Intel's HEDT by giving 8 and 16 strong cores to the masses, after more than a decade of quad cores.

What does this have to do with anything? I'm not AMD, why are you writing me a love letter to them?

You're basically describing any small company winning against a big company.

No.

They gambled on chiplets and TSMC and it worked great, they also created threadripper and 3d-cache. Things that didn't really exist before.

Totally unrelated to anything Radeon.

 

[linkgoron]

What the heck?
Those are all abysmal examples, and totally unrelated to my point too.

You said: "The corpo that takes hard chances in a growing market is winning. The corpo that waits for trends and follows is N°2. That's just how an economy works." (bold mine)

That's just not how "economy" works. Tons of companies learned from the mistakes of the company that worked hard and set a trend and was first, and still won the market.

Totally unrelated to anything Radeon.

I assumed that you were talking about Zen when you said: "They play it safe mostly, take little risks and hardly invest really heavily", as you were talking about Intel a sentence before. However, even if you're talking about GPUs - Investing in chiplets without a real backup plan for GPUs was a huge risk, and it appears to have blown up in their face. This also happened to them before with HBM and Fiji/Fury.

 

[adroc_thurston]

However, even if you're talking about GPUs - Investing in chiplets without a real backup plan for GPUs was a huge risk, and it appears to have blown up in their face

it didn't blow up in anyone's face.
They just focus on raw execution.

 

[Mahboi]

You said: "The corpo that takes hard chances in a growing market is winning. The corpo that waits for trends and follows is N°2. That's just how an economy works." (bold mine)

Yes.

That's just not how "economy" works. Tons of companies learned from the mistakes of the company that worked hard and set a trend and was first, and still won the market.

No.

I assumed that you were talking about Zen when you said: "They play it safe mostly, take little risks and hardly invest really heavily", as you were talking about Intel a sentence before.

I didn't mention Zen at all here. And AMD's CPU business is doing amazing because Intel fell flat on its face and never got back up. Hence the typical AMD problems are very visible in GPU space. The typical AMD idiosyncrasies do exist in both.
AMD just makes MVPs.
You can look at MTL vs Phoenix, it's a great example. Intel brought out the entire chiplet/tile all on a silicon interposer, a complete high end product. The 5 star restaurant treatment with gloves & all. Sure it's fecal matter that they serve, but that's a problem they have to fix.
AMD? They designed a simple, monolithic chip that got a large bunch of (poor) RDNA 3 WGPs shoved on it. No Vcache, no interposer, no BVH walker in the CUs, nothing. Just the smallest viable silicon they could sell.

However, even if you're talking about GPUs - Investing in chiplets without a real backup plan for GPUs was a huge risk, and it appears to have blown up in their face. This also happened to them before with HBM and Fiji/Fury.

I disagree, I think they invested into chiplets literally because they had no choice. They couldn't compete with Intel's capacity or Nvidia's years ahead of research. They found a quick and efficient way to scale at the lowest cost, and luckily, they came out with the way right on cue for TSMC to offer everything they could have hoped for on the menu. CoWoS-L, CoWoS-R, CoWoS-AEIOUY.

It's not a bold design decision, it is the most practical and cheapest way to make a small Zen 2 core and run it off giant servers with 32 or 64 cores as well as ship it in 3600x for $180. It's economies of scale taken to the max. Aka penny pinching.

AMD tries its best to save money and tries to win big. Nvidia just goes big and tries to win big. I think I wrote that somewhere earlier. That's the fundamental difference.

 

[linkgoron]

it didn't blow up in anyone's face.
They just focus on raw execution.

Obviously they had to lie about Navi 31's performance, had to release Navi 32 extremely late, cancel Navi 41/42 and will probably not release any chiplet based consumer GPU until at least 2026 because it's such a success.

 

[gdansk]

Obviously they had to lie about Navi 31's performance, had to release Navi 32 extremely late, cancel Navi 41/42 and will probably not release any chiplet based consumer GPU until at least 2026 because it's such a success.

While funny - it didn't blow up in their face because of the chiplets. The dud wouldn't clock right even when monolithic (Phoenix) and on N4.

 

[branch_suggestion]

they came out with the way right on cue for TSMC to offer everything they could have hoped for on the menu. CoWoS-L, CoWoS-R, CoWoS-AEIOUY.

Who drives TSMC's advanced packaging roadmap?
Because they don't develop the stuff in the hope customers might want to use it, the customer requests it and they work together to make it happen.
First CoWoS was Xilinx (now AMD), first HBM was AMD, first SoIC was AMD, CoWoS-L which is what B100 uses was first used by Apple's M1 Ultra.
It wasn't luck, it was intentional. Their old HyperTransport had to be massively overhauled into Infinity Fabric to stitch the whole thing together.
Give AMD some credit, please.

 

[Mahboi]

Give AMD some credit, please.

Are you gonna fanboying down my ears every time I'm not gonna be praising Lord AMD every day? Cause that's gonna get old extremely fast. Stop. Right now.
You sound like MCH, but the Red version of him.

 

[igor_kavinski]

AMD tries its best to save money and tries to win big.

Can't really blame them for doing that. They almost ceased to exist during the Bulldozer era. It was only their penny pinching that probably kept them alive. Intel in the same situation would've been reduced to something much, much worse coz they are used to throwing away money and it would be hard to stop even when things aren't looking so rosy. AMD tries their best to be efficient. I can't really fault them for that strategy. Let them get at least 60% CPU marketshare and maybe then they will be comfortable risking big money.

By the way, I was able to get someone to buy a Ryzen laptop (coz it was super cheap. Ryzen 5825U for the price of a Core i5). Only my third Ryzen sale so far (and I make zero profit on these sales coz I'm the helpful techie). This is after more than 10 years in my company. So you can imagine what kind of hurdle AMD faces.

 

[branch_suggestion]

Are you gonna fanboying down my ears every time I'm not gonna be praising Lord AMD every day? Cause that's gonna get old extremely fast. Stop. Right now.
You sound like MCH, but the Red version of him.

You're dismissing strategic decisions as luck, this isn't a woe is me situation this is just setting the record straight.
You cannot build the 747 before you build the 707, no company builds something big and crazy without a solid foundation first, or they inevitably crash and burn.
Even with leadership and a strong base, taking too many risks at once is very dangerous, something all 3 companies have been through to some degree.
Establish a sound roadmap, execute, build your base and then get more ambitious over time in tandem with what the market expects.
The DC accel market looks to remain massive for the meanwhile, so due to the massive TAM opportunity you can bet everyone is pushing their roadmap as hard as they dare.

 

[KompuKare]

Who drives TSMC's advanced packaging roadmap?
Because they don't develop the stuff in the hope customers might want to use it, the customer requests it and they work together to make it happen.
First CoWoS was Xilinx (now AMD), first HBM was AMD, first SoIC was AMD, CoWoS-L which is what B100 uses was first used by Apple's M1 Ultra.
It wasn't luck, it was intentional. Their old HyperTransport had to be massively overhauled into Infinity Fabric to stitch the whole thing together.
Give AMD some credit, please.

HBM is an strange one.
An almost bankrupt AMD along with Hynix were the instigators, but once done AMD had nothing to really use it with.

https://en.m.wikipedia.org/wiki/High_Bandwidth_Memory#Development

Fuji almost looked a pride thing.
And that is still mostly the case.
In the meantime, Nvidia and even Intel have shipped far more HBM products than AMD.

It all just seemed like such a strange thing for an almost bankrupt company to do especially since their competitors have benefitted more from it than AMD have.

 

[Mahboi]

Can't really blame them for doing that. They almost ceased to exist during the Bulldozer era. It was only their penny pinching that probably kept them alive. Intel in the same situation would've been reduced to something much, much worse coz they are used to throwing away money and it would be hard to stop even when things aren't looking so rosy. AMD tries their best to be efficient. I can't really fault them for that strategy.

Fully agreed.
I don't think AMD is doomed to behave like a low funds corpo forever either.
But I'm getting a little tired of all the fanboys that don't understand a thing about running a company, whining about technical solutions to market problems.

AMD has been reborn through the dark years thanks to lean, cheap and efficient designs. Nobody will deny that.
However those years are long past. A lot of people still act like this is the pre Zen 2 company, when we're entering Zen 5.

- Pre Zen 2 was price competitive, not performance competitive
- Zen 2 required cheap solutions like IOD connected through traces to the CCDs
- Zen 2 was competitive while being a much smaller core
- Zen 3 took the performance crown across the stack (I even run Zen 3, happily)
- Zen 4 translated the success seamlessly into a new platform, new RAM, etc

Now Zen 5 is holding before the gate into the arena, and promises to be the first true new thing since Zen 2. We are 5 minutes before the bell rings, and the crowd is excited, yet everyone acts like this is going to be the Zen 2 fight all over again. This is typical, people admire what worked, instead of thinking about what doesn't work anymore.
I think AMD and its fans have not been able to shed that penny pinching mentality since it's what brought them victory before. But that's not how a market works. You can't be meek and try to save money when you are about to break the market. On the contrary you should behave like Intel or Nvidia, invest massively, risk a lot, and take everything you can. Playing second fiddle is fine when you're second, top dogs must play hardball. I feel like way too many AMD fans are still beholden to that "play smart not hard" mentality that makes absolutely no sense anymore.

You have the money, you have the IP, you have the weapons. You HAVE to hit hard, AMD. And not just with good technicals, but with expensive, ambitious stuff. That's what I'm seeing with Zen, steadily, with stuff like Zen 6 or Strix Halo, and I'm not seeing with Radeon, at all. Just the fact that everyone expected SEO and a BVH walker for RDNA 3 and that it only comes 2 years later, most likely because Sony requested it, tells me so much about the mentality at AMD when it comes to getting ahead. Same with RDNA 4 dropping anything past a N23 sized die. Admittedly their biggest problem is software anyway.

Let them get at least 60% CPU marketshare and maybe then they will be comfortable risking big money.

Sorry but that's never going to work. As one guy put it, you can play to not lose, or you can play to win. To not lose means to avoid losses and play safe. To win means to take the risks of attacking. In this case, to try possibly mad and ambitious things. I'm just not seeing that happening especially when all the fans are praising the "play smart not hard" strategy that belongs to the 2017-2022 era.
And you never ever get the dominant position by playing to not lose. You either take the risks and attack the top dog, or you don't and you stay an eternal second. Those 60% marketshare will take 15 years to happen with those couch strategies.

By the way, I was able to get someone to buy a Ryzen laptop (coz it was super cheap. Ryzen 5825U for the price of a Core i5). Only my third Ryzen sale so far (and I make zero profit on these sales coz I'm the helpful techie). This is after more than 10 years in my company. So you can imagine what kind of hurdle AMD faces.

Yeah I also got a friend who was like "wait what do you mean AMD...are you trying to make me buy a bum CPU?". But frankly I blame the marketing on this. Sure, fighting Intel's rep is a gargantuan task. And they don't have much of an argument against Nvidia yet. But their public presence is just poor. I don't know, it's 2024's marketing, make tiktoks or youtube shorts of Ryzens eating through workloads much faster than 13900ks and show the power consumption all along. 10 second video of a blender render comparison with some background music so you can send that to people. Tell everyone that the PS5 is AMD hardware through and through and is awesome.

 

[branch_suggestion]

HBM is an strange one.
An almost bankrupt AMD along with Hynix were the instigators, but once done AMD had nothing to really use it with.

https://en.m.wikipedia.org/wiki/High_Bandwidth_Memory#Development

Fuji almost looked a pride thing.
And that is still mostly the case.
In the meantime, Nvidia and even Intel have shipped far more HBM products than AMD.

It all just seemed like such a strange thing for an almost bankrupt company to do especially since their competitors have benefitted more from it than AMD have.

AMD finally has a nice enough product to put HBM with, it was worthwhile, just not as soon as hoped, pretty common thread.

I think AMD and its fans have not been able to shed that penny pinching mentality since it's what brought them victory before. But that's not how a market works. You can't be meek and try to save money when you are about to break the market. On the contrary you should behave like Intel or Nvidia, invest massively, risk a lot, and take everything you can. Playing second fiddle is fine when you're second, top dogs must play hardball. I feel like way too many AMD fans are still beholden to that "play smart not hard" mentality that makes absolutely no sense anymore.

AMD's R&D/MDF spend as % of revenue has been highly competitive over the last 5 years.

You have the money, you have the IP, you have the weapons. You HAVE to hit hard, AMD. And not just with good technicals, but with expensive, ambitious stuff. That's what I'm seeing with Zen, steadily, with stuff like Zen 6 or Strix Halo, and I'm not seeing with Radeon, at all.

It is the plan, now just 2026 instead of 2025.

Just the fact that everyone expected SEO and a BVH walker for RDNA 3 and that it only comes 2 years later, most likely because Sony requested it, tells me so much about the mentality at AMD when it comes to getting ahead.

Market adoption of that stuff is still meh, it isn't part of API's so it is a meme, end of story.

Same with RDNA 4 dropping anything past a N23 sized die. Admittedly their biggest problem is software anyway.

You're getting a futureproof $400 1440p card and you are going to like it.

Sorry but that's never going to work. As one guy put it, you can play to not lose, or you can play to win. To not lose means to avoid losses and play safe. To win means to take the risks of attacking. In this case, to try possibly mad and ambitious things. I'm just not seeing that happening especially when all the fans are praising the "play smart not hard" strategy that belongs to the 2017-2022 era.
And you never ever get the dominant position by playing to not lose. You either take the risks and attack the top dog, or you don't and you stay an eternal second. Those 60% marketshare will take 15 years to happen with those couch strategies.

AMD played those games in the Athlon era, it led them to Bulldozer. Need I say more?

Yeah I also got a friend who was like "wait what do you mean AMD...are you trying to make me buy a bum CPU?". But frankly I blame the marketing on this. Sure, fighting Intel's rep is a gargantuan task. And they don't have much of an argument against Nvidia yet. But their public presence is just poor. I don't know, it's 2024's marketing, make tiktoks or youtube shorts of Ryzens eating through workloads much faster than 13900ks and show the power consumption all along. 10 second video of a blender render comparison with some background music so you can send that to people. Tell everyone that the PS5 is AMD hardware through and through and is awesome.

We all know their marketing to normies remains bad, but they are becoming a bit scummier with marketing in recent times, which annoys us but it does work against the proles.

 

[Mahboi]

You're dismissing strategic decisions as luck, this isn't a woe is me situation this is just setting the record straight.

I'm not dismissing anything, I'm calling the reasoning behind it. They wanted the cheapest way to scale. It was chiplets. Good choice.
Fair enough on AMD pushing TSMC forward though.

You cannot build the 747 before you build the 707, no company builds something big and crazy without a solid foundation first, or they inevitably crash and burn.
Even with leadership and a strong base, taking too many risks at once is very dangerous, something all 3 companies have been through to some degree.
Establish a sound roadmap, execute, build your base and then get more ambitious over time in tandem with what the market expects.

Yaddi yadda, how many times must I read that. I speak of ambitions, you speak of roadmaps. I speak of goals, you speak of techniques.
It's the one thing that can't get through to technical people: you speak of will, they think of way. I'm not talking about the way, I'm talking about the will.

AMD has an understandable technical obsession with finding the cheapest, most cost effective way to do something.
Nvidia doesn't mind throwing extra Si for Tensor Cores, or mislead people into believing that they're good for raytracing or whatnot. They don't mind fully committing to AI. They don't mind buying obscene volume and then be left holding the bag when the crypto madness ends. That's WILL. Nvidia doesn't even have far more competent people than AMD. They just do the things they want to do, and manage to force the market into it. That's why they win.
Same was true of Intel but their internal problems and conceit has completely murdered their products.

When the corporation that thinks of cost-effectiveness is N°2, they thrive. When they're N°1, they sour. Because being on top is about leading the space and raking in the money that funds your further leading. AMD's getting there with Zen, and people still talk cost-effectiveness all the time. Think big.
AMD's not there yet with Radeon, but they absolutely aren't showing any ambition. It's all Si saving, small dies, meek stuff. MI300 is the glorious exception but it comes off to me as a N°2's response to the N°1 having created a monster market. I think MI300 wouldn't have been nearly as ambitious if it weren't for H100's obscene ROI. And there's a lot to say about its generic-ness as well. MI300 is basically throwing a battlecruiser in the bathtub. Everyone's impressed, but when you look at Blackwell forsaking all high precision for the sake of doing more AI, you can see where Nvidia is going. They want to be the AI God. The Alpha and Omega of FP4. MI300 is a huge beast with huge capacity. It's a wonderful piece of technology, but what does it say about AMD's ambitions? It says: "we can do big chip for you". Not "we WANT to do big AI chip" or "we WANT to do big supercomputer compute". Just...we don't know what the market wants, and we don't want anything, so we'll just provide something that sort of satisfies every need.
A way, not a will.

The DC accel market looks to remain massive for the meanwhile, so due to the massive TAM opportunity you can bet everyone is pushing their roadmap as hard as they dare.

That's fair. I'm just not seeing much ambition yet. Just holdovers and promises and "spam transistors". Just RDNA 4 giving up on anything past a 6600 xt's die area. Just more of Sony asking for things, or DC asking for things.
Compared to Nvidia that has basically jumped so hard into AI they've redesigned the entire company around it, it's quite a tame show, that's all. And I don't see anyone pushing in the right direction, just "someday, they'll have Beeg GPU and Nvidia will quake in its bootsies". I don't think the company that has a Dictator for Life with crazy ambitions is going to let himself be beaten by the company of meek nerds. It just isn't what I'm seeing here.

 

[Mahboi]

more of the same
more of the same
more of the same
more of the same
more of the same

AMD played those games in the Athlon era, it led them to Bulldozer. Need I say more?

No thanks, I wish you'd stop. I've made the same point 3 times already.

We all know their marketing to normies remains bad, but they are becoming a bit scummier with marketing in recent times, which annoys us but it does work against the proles.

The proles are funding the corpo, so instead of getting annoyed at your marketing being more efficient, thank it.

 

[branch_suggestion]

Yaddi yadda, how many times must I read that. I speak of ambitions, you speak of roadmaps. I speak of goals, you speak of techniques.
It's the one thing that can't get through to technical people: you speak of will, they think of way. I'm not talking about the way, I'm talking about the will.

The will is to .ppt flex on the comp with superior products.

AMD has an understandable technical obsession with finding the cheapest, most cost effective way to do something.
Nvidia doesn't mind throwing extra Si for Tensor Cores, or mislead people into believing that they're good for raytracing or whatnot. They don't mind fully committing to AI. They don't mind buying obscene volume and then be left holding the bag when the crypto madness ends. That's WILL. Nvidia doesn't even have far more competent people than AMD. They just do the things they want to do, and manage to force the market into it. That's why they win.
Same was true of Intel but their internal problems and conceit has completely murdered their products.

High concept era of computing died with Moore's Law. Sorry but winning is rather boring at the end of the day for those who like fantasies.

When the corporation that thinks of cost-effectiveness is N°2, they thrive. When they're N°1, they sour. Because being on top is about leading the space and raking in the money that funds your further leading. AMD's getting there with Zen, and people still talk cost-effectiveness all the time. Think big.
AMD's not there yet with Radeon, but they absolutely aren't showing any ambition. It's all Si saving, small dies, meek stuff. MI300 is the glorious exception but it comes off to me as a N°2's response to the N°1 having created a monster market. I think MI300 wouldn't have been nearly as ambitious if it weren't for H100's obscene ROI. And there's a lot to say about its generic-ness as well. MI300 is basically throwing a battlecruiser in the bathtub. Everyone's impressed, but when you look at Blackwell forsaking all high precision for the sake of doing more AI, you can see where Nvidia is going. They want to be the AI God. The Alpha and Omega of FP4. MI300 is a huge beast with huge capacity. It's a wonderful piece of technology, but what does it say about AMD's ambitions? It says: "we can do big chip for you". Not "we WANT to do big AI chip" or "we WANT to do big supercomputer compute". Just...we don't know what the market wants, and we don't want anything, so we'll just provide something that sort of satisfies every need.
A way, not a will.

You can only pull levers once, then you have to get creative since there are only so many levers to pull.
MI300 was building the best part possible for the DoE, so some levers for chasing the AI meme are yet to be pulled.

That's fair. I'm just not seeing much ambition yet. Just holdovers and promises and "spam transistors". Just RDNA 4 giving up on anything past a 6600 xt's die area. Just more of Sony asking for things, or DC asking for things.
Compared to Nvidia that has basically jumped so hard into AI they've redesigned the entire company around it, it's quite a tame show, that's all. And I don't see anyone pushing in the right direction, just "someday, they'll have Beeg GPU and Nvidia will quake in its bootsies". I don't think the company that has a Dictator for Life with crazy ambitions is going to let himself be beaten by the company of meek nerds. It just isn't what I'm seeing here.

Majority of historical analogues end up with the quiet achiever beating the braggart.
"Speak softly and carry a big stick; you will go far."

No thanks, I wish you'd stop. I've made the same point 3 times already.

Nobody gets fired for buying NVIDIA!

The proles are funding the corpo, so instead of getting annoyed at your marketing being more efficient, thank it.

I would if I was an investor, I just like my first party material to be accurate and not lies stretched to hell and back with misleading/irrelevant comparisons.

 

[Mahboi]

So I guess he's not stopping, well I am.

 

[igor_kavinski]

But their public presence is just poor. I don't know, it's 2024's marketing, make tiktoks or youtube shorts of Ryzens eating through workloads much faster than 13900ks and show the power consumption all along. 10 second video of a blender render comparison with some background music so you can send that to people. Tell everyone that the PS5 is AMD hardware through and through and is awesome.

Reasons they may want to avoid a marketing campaign like that is

1) Cost. Instead of increasing the marketing budget two or three fold to take on Intel, the money could be spent on R&D or on bonuses for deserving employees.

2) Inadequate volume. They probably don't have enough volume to satisfy market demand if people indeed start ditching Intel thanks to an aggressive AMD marketing push. It's not like they can go to TSMC and say, we need this many wafers in the next few quarters and get them at their currently contracted prices. TSMC will be like, cha-ching! and charge them through the nose for urgent wafer allotment.

 

[Mahboi]

Reasons they may want to avoid a marketing campaign like that is

1) Cost. Instead of increasing the marketing budget two or three fold to take on Intel, the money could be spent on R&D or on bonuses for deserving employees.

I doubt that internet marketing costs that much. I'm not going to delve into marketing costs ITT but it's broadly speaking doable on a small budget if you're restraining yourself to certain services. Not that they can do that for enterprise clients ofc, but those are a very different business.

2) Inadequate volume. They probably don't have enough volume to satisfy market demand if people indeed start ditching Intel thanks to an aggressive AMD marketing push. It's not like they can go to TSMC and say, we need this many wafers in the next few quarters and get them at their currently contracted prices. TSMC will be like, cha-ching! and charge them through the nose for urgent wafer allotment.

Agreed on volume though, but it is sort of a chicken and egg situation isn't it?
Do they not have enough volume because they don't think they can sell that much, thus they don't market too hard, thus they don't get new clients, thus we're back to not selling that much.
Or do they not have enough volume because they're hard limited by how much they can realistically buy from TSMC and when bidding on more wafers starts being not worth the extra money?

I have no idea about TSMC's actual production limits, so I'll pass on making a claim, but it is a real question right now. The fact that AMD is willing to buy SS4P for Sonoma Valley is telling that they want more, but it's just a super cheap Mendocino tier node whose actual volume is unclear to me. I honestly expected them to go SS for a lot of mobile and lower cost stuff, not just one node. May be the beginning of bigger volumes though!

 

[igor_kavinski]

May be the beginning of bigger volumes though!

Yeah. Could be a trial run. Samsung better not mess this up.

 

[Mahboi]

Yeah. Could be a trial run. Samsung better not mess this up again.

smol fix

 

[Tuna-Fish]

It all just seemed like such a strange thing for an almost bankrupt company to do especially since their competitors have benefitted more from it than AMD have.

AMD/ATi was the driving force behind a lot of the earlier GDDR standards, they co-operated with a memory manufacturer to develop both GDDR3 and GDDR5. They have benefited a lot from this in the past, RV770 wouldn't have been such a success if they had not been able to ship a card with a 256b bus width that nearly matched their earlier 512b design in bandwidth. HBM was just continuing the trend.

Ultimately, they wanted to drive HBM prices down so that they'd ship full stack of GPU products with it, but that never happened.

 

[Mopetar]

My only gripe is that Navi 40 now got cut at the head, shoulders, and chest. All that's left is below the crotch. N48 is a 6600 xt's size. N44 is a 130mm² tiny little thing meant to go into laptops.

AMD goes small then tries to win big anyway. RDNA 4 is just another one in the long list of "meh" moments.

I don't mind this and think it could be good for AMD in the long run. Navi 32 proved popular for AMD because they had a mass market product at a compelling price. Navi 48 looks like it could do more of that and help get some consumers to buy their first AMD card.

Chasing after NVidia on the high end hasn't worked out well for them and when they are constrained on wafers and can make more money producing more Zen dies for the server market, they don't have a lot of incentive to increase their GPU production.

With smaller dies AMD can produce more product and actually try to grow market (and mind) share for a change, which will help them if and when they can better execute in the future.