Discussion RDNA 5 / UDNA (CDNA Next) speculation

[marees]

What is the expectation on the memory situation?
When will it cool down?
Will lpddr5x & lpddr6 be affected?

When will RDNA 5 launch now
What will be the SKUs

Will it be something like this

1. 50xt 12gb (AT4 24 CU) — $300
2. 60 16gb (AT3 40? CU) — $400
3. 60xt 16gb/24gb/32gb (AT3 48 CU) — $500/$550/$600
4. 70 18gb (AT2 56/60 CU) — $650
5. 70xt 18gb/24gb (AT2 72 CU) — $700/$800
6. 80xt 24gb (AT0 128? CU) — $1200+
7. 90xt 36gb (AT0 144? CU) — $1500+

 

[jpiniero]

Will lpddr5x & lpddr6 be affected?

Def seems like it should.

I have a feeling that the LPDDR products will get cancelled, as well as any AT0 gaming products. Mainly due to the prices.

 

[adroc_thurston]

I have a feeling that the LPDDR products will get cancelled, as well as any AT0 gaming products. Mainly due to the prices.

not how any of that works

 

[jpiniero]

not how any of that works

You have to realize that AMD and nVidia are probally going to raise MSRP by $50 for every 8 GB very soon. Sales are gonna tank. AT0 is going to be too expensive to expect people who would buy AMD GPUs to buy it, regardless of performance. May as well turn any of the duds into AI cards.

Maybe you will get lucky that the AI bubble will burst by the time AMD has to commit to a plan for RDNA 5.

 

[adroc_thurston]

You have to realize that AMD and nVidia are probally going to raise MSRP by $50 for every 8 GB very soon

and

Sales are gonna tank.

they can just cut volumes and leverage the shortage

AT0 is going to be too expensive to expect people who would buy AMD GPUs to buy it, regardless of performance

does not matter.

Maybe you will get lucky that the AI bubble will burst by the time AMD has to commit to a plan for RDNA 5.

They commited to a plan 2 years ago.

 

[maddie]

and

they can just cut volumes and leverage the shortage

does not matter.

They commited to a plan 2 years ago.

You argue from a shareholder POV. I think its safe to say that most here, approach the DRAM crisis from a consumer/PC enthusiast angle.

With that said, it will impact sales significantly with consumers and smaller enterprises will suffering the most.

Regarding plans, and quoting an aphorism. "No plan survives contact with the enemy".

I've seen some claiming that a fall in sales will cause prices to retreat. Standard supply and demand thinking. This is wrong. The AI enterprises want much more RAM that the market can supply. A fall in consumer sales is exactly what they want. Barring a massive increase in production capacity, or a political intervention reacting to traditional profitable PC markets being disrupted, think employment, as an example. Alternately, seeing that a lot of US economic growth is being driven by AI and related CAPEX, the choices are quite stark.

 

[gdansk]

Maybe prices will be worse, maybe some SKUs for gamers will be taken out back. But there is good reason to have chips with flexible memory configuration taped out as planned. There will be one company that has the worst HBM. Yet all three have new facilities in construction now that plan to start production in 2026/2027.

 

[jpiniero]

re: PS6, I suppose they could go with 4 GB chips on a 160 bit bus and do partial clamshell. That would depend of course if RDNA5 supports that or even if there's any kind of cost savings. It'd be 8x4GB instead of 10x3GB.

Still feels like a bad idea.

 

[adroc_thurston]

I think its safe to say that most here, approach the DRAM crisis from a consumer/PC enthusiast angle

GPU product stacks are not built around commodity IC pricing in a DIY market.

I've seen some claiming that a fall in sales will cause prices to retreat. Standard supply and demand thinking. This is wrong. The AI enterprises want much more RAM that the market can supply. A fall in consumer sales is exactly what they want. Barring a massive increase in production capacity, or a political intervention reacting to traditional profitable PC markets being disrupted, think employment, as an example. Alternately, seeing that a lot of US economic growth is being driven by AI and related CAPEX, the choices are quite stark.

Oh you see that's the thing, consumer sales being strong are the problem.
DRAM vendors aren't bulking up on HBM capacity lmao, it's all DDR5 now.

 

[Thunder 57]

You argue from a shareholder POV. I think its safe to say that most here, approach the DRAM crisis from a consumer/PC enthusiast angle.

I don't see it that way. He is arguing from a realistic POV. As consumers we would all want more supply/cheaper prices but that isn't always possible.

 

[maddie]

I don't see it that way. He is arguing from a realistic POV. As consumers we would all want more supply/cheaper prices but that isn't always possible.

"they can just cut volumes and leverage the shortage"

I understood this to mean ASP would rise to compensate for reduced sales = same revenue. I never meant it was false, just bad for consumers.

 

[Joe NYC]

"they can just cut volumes and leverage the shortage"

I understood this to mean ASP would rise to compensate for reduced sales = same revenue. I never meant it was false, just bad for consumers.

I don't see it that way.

AMD will just charge its own normal prices, DRAM will increase the price of final cards, and that price increase of final cards may reduce volumes somewhat.

 

[maddie]

I don't see it that way.

AMD will just charge its own normal prices, DRAM will increase the price of final cards, and that price increase of final cards may reduce volumes somewhat.

How does that leverage the shortage?

 

[Joe NYC]

How does that leverage the shortage?

I disagree with @adroc_thurston 's take on it.

 

[adroc_thurston]

I disagree with @adroc_thurston 's take on it.

AMD is in the business of making money and not giving gamers(tm) a freebie.
ASP hikes go in.

 

[Joe NYC]

AMD is in the business of making money and not giving gamers(tm) a freebie.
ASP hikes go in.

AMD is doing quite well with the current strategy with RDNA4 cards. Why change it?

Just adjust prices based on BOM changes.

 

[adroc_thurston]

AMD is doing quite well with the current strategy with RDNA4 cards

They're doing awful, cancelling Navi4C was a bad decision on the level of commiting to Bulldozer.

 

[inquiss]

They're doing awful, cancelling Navi4C was a bad decision on the level of commiting to Bulldozer.

If looking only at the client business I agree, but you'd have thought some of the physical design resources would have been shifted onto speeding their cadence on instinct, which is dollars

 

[Win2012R2]

When will RDNA 5 launch now

2027

 

[Joe NYC]

They're doing awful, cancelling Navi4C was a bad decision on the level of commiting to Bulldozer.

You are comparing vs. hypothetical, of not cancelling Navi4C, which would have strong sales in this environment.

I am comparing it to actual performance, vs. 2024 (with RDNA3).
Q3 2024 graphics revenue: 0.5 billion
Q3 2025 graphics revenue: 1.3 billion.

That's 160% improvement in graphics revenue YoY. Even if majority of the increase was console, it would still be at least +100% revenue.

Which is why I said AMD is doing quite well with the current RDNA4 strategy. AMD just needs to plug the holes in the product line, have a more complete line up.

RDNA5 will just need to have a slightly better gen on gen improvement than NVidia and the market share gain, revenue gain momentum will continue.

Local AI on client is something that can still happen, AMD just needs to be prepared for it (and not miss the boat like with Navi4c cancellation).

 

[Joe NYC]

If looking only at the client business I agree, but you'd have thought some of the physical design resources would have been shifted onto speeding their cadence on instinct, which is dollars

Yup, I think that's what happened.

 

[adroc_thurston]

You are comparing vs. hypothetical, of not cancelling Navi4C, which would have strong sales in this environment.

it's not hypothetical.

Which is why I said AMD is doing quite well with the current RDNA4 strategy. AMD just needs to plug the holes in the product line, have a more complete line up.

RDNA5 will just need to have a slightly better gen on gen improvement than NVidia and the market share gain, revenue gain momentum will continue.

it's a winner-takes-all market and they refuse to ship a bigger config.
They're doomed (but that has been obvious for a long-long while).

 

[Joe NYC]

it's a winner-takes-all market and they refuse to ship a bigger config.
They're doomed (but that has been obvious for a long-long while).

I would have said the same, a year ago, but RDNA4 is defying the odds.

Defying the odds with limited range of products (no high end), desktop only.

 

[adroc_thurston]

but RDNA4 is defying the odds.

It ain't.
At least ASPs are healthy there.

 

[dangerman1337]

Exact RDNA 5 SKUs, pricing and dates can change. Too many factors to decide, especially if the OpenAI Circle Jerk Bubble bursts.

it's not hypothetical.

it's a winner-takes-all market and they refuse to ship a bigger config.
They're doomed (but that has been obvious for a long-long while).

I mean AMD could release a 48GB AT0 (or 64 if they feel like doing 4GB modules, or leave those for pro cards) full fat-SKU against a 6090/6090 Ti if they feel it. Again that leaked MLID chart can change even if Dies get fixed in spec and design wise.