Discussion RDNA 5 / UDNA (CDNA Next) speculation

[Win2012R2]

but what you're getting is a normal console.

Consoles don't sell in volume at prices that MSFT appeared to have planned even before RAM spikes, it will be cheaper/saner for them to just cancel it completely, you've heard it on here first.

 

[adroc_thurston]

Consoles don't sell in volume at prices that MSFT appeared to have planned

They planned a normal box for the normal price.
MS 10th gen gamble is being a good CODbox once again.

 

[jpiniero]

They planned a normal box for the normal price.
MS 10th gen gamble is being a good CODbox once again.

They won't do that, since MS is pivoting to Windows.

It has to be expensive. Even the PS6 is probably going to be at least $699, and that's if RAM prices come back down.

 

[adroc_thurston]

They won't do that, since MS is pivoting to Windows.

Nope.
I get it, your kind (PCels) is desperate for subsidized gaming hardware. But what you get is a console that will have games. maybe.

It has to be expensive

Normal console prices (well, adjusted for the BOM bump).

PS6 is probably going to be at least $699

599 probs.
They'll try to not repeat the PS3 launch.

 

[jpiniero]

Console subsidies are more or less over, even now.

 

[adroc_thurston]

Console subsidies are more or less over, even now.

a) no they ain't
b) the last actually subsidized console was like PS3.

 

[Joe NYC]

No volume = no monies.

Fixed costs pretty high, plus big opportunity cost of using expensive devs doing something that won't make big monies.

NVidia is making a fortune on 5090 and its derivatives. The market is there, the only question is if AMD can gain sufficient percentage of it.

And only way to gain a sufficient share of a market is by shipping products into that market / segment. Not supping, with RDNA4, was clearly a mistake.

 

[adroc_thurston]

question is if AMD can gain sufficient percentage of it.

ah no they don't care.

And only way to gain a sufficient share of a market is by shipping products into that market / segment

Not a single exec at AMD wants to compete with NV in client gfx.

 

[jpiniero]

ah no they don't care.

Not a single exec at AMD wants to compete with NV in client gfx.

It is kind of rough when people expect your product to be faster and $100+ cheaper to even consider buying.

 

[Win2012R2]

They planned a normal box for the normal price.
MS 10th gen gamble is being a good CODbox once again.

Xbox stopped selling and COD will run just fine on PS5 and 6, they can't even afford to make it time exclusive to Xbox, there is zero reason to buy Xbox now, the war is completely lost now.

NVidia is making a fortune on 5090 and its derivatives

Are you having a laugh comparing high end market like that (a few millions units max) with console buyers (target for generation: 60-100 mln)? I don't think 5090 specifically sells well, 4090 did sell very well however.

Not a single exec at AMD wants to compete with NV in client gfx.

False, this can't be true and if you know that (and you should) then it makes you a liar.

 

[adroc_thurston]

Xbox stopped selling and COD will run just fine on PS5 and 6, they can't even afford to make it time exclusive to Xbox, there is zero reason to buy Xbox now, the war is completely lost now.

It's okay Arcanum 2 will be a system-seller.

 

[Win2012R2]

It's okay Arcanum 2 will be a system-seller.

Xbox hardware is DEAD, ok?

Before 2025 it was 1 xbox for 2 PS5, now this is 1 for 3.5-4, that's death of the "platform". No considerably higher priced model can fix it, all exclusives now go to PS5 too, that's sealed the fate completely.

Their hope was to release this earlier than PS5 but now with higher memory prices that just won't be happening or they will sell about nothing. It will be easier for publishers to get a few extra bucks from Sony to make their release exclusive there (plus PC release) than bother porting it to Xbox and wasting money and time trying to make it all work on crippled S series: which is really the elephant in the room, because it makes up 2/3 of Xbox sales and devs required to support both, it will be basically cheaper to not bother and 5-7 mln sales of higher powered Xbox2 won't change the equation.

 

[adroc_thurston]

Xbox hardware is DEAD, ok?

no.

Before 2025 it was 1 xbox for 2 PS5

no lol, it was 1:4 ratio usually.

 

[Joe NYC]

ah no they don't care.

Not a single exec at AMD wants to compete with NV in client gfx.

But, at the same time, not a single AMD exec wants the repeat of NVidia running away in a market where AMD meaningfully participated in, leaving AMD half a decade behind, with half a decade of playing catch up (meaning burning OpEx, barely covered by revenue).

I don't think AMD wants to repeat that blunder.

 

[adroc_thurston]

But, at the same time, not a single AMD exec wants the repeat of NVidia running away in a market where AMD meaningfully participated in, leaving AMD half a decade behind, with half a decade of playing catch up (meaning burning OpEx, barely covered by revenue).

?

I don't think AMD wants to repeat that blunder.

Gaming is a really bad market to be in so they just don't.

 

[jpiniero]

Gaming is a really bad market to be in so they just don't.

nVidia still makes a ton of money off of it.

 

[Joe NYC]

Are you having a laugh comparing high end market like that (a few millions units max) with console buyers (target for generation: 60-100 mln)? I don't think 5090 specifically sells well, 4090 did sell very well however.

NVidia "gaming" revenue in the last reported quarter was $4.3 billion.

Imagine all of AMD EPYC CPUs + all of AMD datacenter GPUs + networking + other datacenter peripherals that AMD sells. The crown achievement of AMD. Revenue? Exactly the same, $4.3 billion per quarter

Anyone who poo-poos NVidia "gaming" revenue, which includes desktop dGPU, mobile dGPU, workstation dGPU and industrial uses of GPUs derived from the same die - anyone who think this market is not worth pursuing is not living in reality.

AMD stumbles with RDNA3 and incomplete product line up for RDNA4 - those are blunders. That's AMD underestimating the market, not taking it seriously and then paying the price in lost revenue.

 

[Joe NYC]

Gaming is a really bad market to be in so they just don't.

$4.3 billion "gaming" market is a bad market to be in?

You are just being silly. Sometimes AMD just plainly screws up, comes short or misses huge opportunities.

It's not some sort of virtue or superior vision of AMD management to be missing out on $4.3 billion revenue even while spending most of the OpEx to participate in this segment.

Just compare how much R&D AMD has to spend on its entire datacenter line up CPU / GPU, including cost of acquisitions like Pensando. That's a multiple by several X AMD is spending vs. what NVidia spends on "gaming" to derive the same revenue.

 

[adroc_thurston]

nVidia still makes a ton of money off of it.

They got lucky and did not commit brand suicide.
Blame RV770.

$4.3 billion "gaming" market is a bad market to be in?

Yeah, it's a market full of petulent children that hate Radeon.

Just compare how much R&D AMD has to spend on its entire datacenter line up CPU / GPU, including cost of acquisitions like Pensando. That's a multiple by several X AMD is spending vs. what NVidia spends on "gaming" to derive the same revenue.

Those are strategic markets where AMD will win major share. They can't win gaming, that one's been lost due to RV770.

 

[Joe NYC]

Yeah, it's a market full of petulent children that hate Radeon.

AMD won over majority of formerly loyal Intel fans for its CPU business. Just a few remaining Loony Tunes left, "tuning" their degrading Raptor Lakes.

The same can be done in GPU.

Those are strategic markets where AMD will win major share.

What do you mean, like the dollars from those markets are coated with Pixie Dust or cocaine powder?

Before the recent upturn in server CPU, it looked like a $5 billion market for x86 CPU, and AMD already has 40% of it. Which is 1/2 of $10 billion client x86 CPU market.

The "gaming" market, between AMD and NVidia is currently same or bigger than the "highly strategic" server CPU market.

They can't win gaming, that one's been lost due to RV770.

Most gamers don't even know what RV770 means.

Part of gaining credibility for premium / Halo notebook (and MiniPC) APUs can be gained by gaining reputation in client dGPU.

It seems like AMD is aiming to do exactly that with RDNA5, in all its existing segments.

Additionally, hoping that some of that client Aye Eeee drops in AMD's lap.

 

[adroc_thurston]

AMD won over majority of formerly loyal Intel fans for its CPU business.

CPU market never had any loyalties.

What do you mean, like the dollars from those markets are coated with Pixie Dust or cocaine powder?

yeah

The "gaming" market, between AMD and NVidia is currently same or bigger than the "highly strategic" server CPU market.

90% of server rev share would be like 9B for AMD.

Most gamers don't even know what RV770 means.

They know that Radeon Sucks and Radeon Sucks (and is for poor people) begins with RV770.

It seems like AMD is aiming to do exactly that with RDNA5, in all its existing segments.

Nope, none of the gfx13 parts compete with Nvidia at all.
None of them are for DIY AICs to begin with!

 

[Joe NYC]

CPU market never had any loyalties.

You only find out the answer by testing these loyalties.

90% of server rev share would be like 9B for AMD.

My guess is that AMD server CPU revenue is ~ $2.5 billion, and if that is 40%, it would make the market $6.25 billion.

The $6.25 billion may be up quite sharply. It was probably closer to $5.5 billion a year ago.

So, we have roughly:
$10.0 billion client CPU
$6.50 billion "gaming"
$6.25 billion server CPU

If AMD can displace Gravitons of the world, then x86 server TAM can grow bigger, but we have not seen any evidence of Arm retreat and x86 gain in server CPU.

The only thing we actually are seeing is x86 server CPU TAM growing due to AI inference growth. This could be the first time a nice windfall will fall into AMD 's laps as a benefit of being prepared.

Nope, none of the gfx13 parts compete with Nvidia at all.
None of them are for DIY AICs to begin with!

Most of the client side AI and AI development is done on cards like 4090 and 5090. These are identical to DIY cards.

Same as with Crypto, Jensen was prepared, at the right place at the right time. AMD was just unprepared, in both cases.

I am pretty sure that inside AMD, they consider missing out on both of these important trends to be blunders, while you think it is some secret wisdom.

 

[adroc_thurston]

You only find out the answer by testing these loyalties.

It never had any.
Neither vendor managed to maintain MSS with horrible product in CPU space.

Most of the client side AI and AI development is done on cards like 4090 and 5090. These are identical to DIY cards.

Ah no that's not a real market.
AT0 is a cloud gaming part anyway.

I am pretty sure that inside AMD, they consider missing out on both of these important trends to be blunders, while you think it is some secret wisdom.

They genuinely do not care about gaming.
It doesn't matter that it makes you upset. No exec at AMD is willing to dump money into gaming. Gaming is dead at AMD.

 

[Joe NYC]

Ah no that's not a real market.

I gave you the size of the market, ~$6.5 billion of client dGPU cards (excluding APU).

It seems like you are in denial, and that NVidia's $4.5 trillion market cap started from "gaming".

AT0 is a cloud gaming part anyway.

How do you know this?

Because equivalent NVidia die is for:
- client gaming
- cloud gaming
- "professional" workstation graphics
- visualization, rendering
- video editing, post production
- crypto
- data analysis
- AI programming development platform
- client AI
- machine learning

$4.5 trillion market cap for NVidia started from these same DIY cards and with investment in software tools, multiplied the uses of these and similar dies.

They genuinely do not care about gaming.
It doesn't matter that it makes you upset. No exec at AMD is willing to dump money into gaming. Gaming is dead at AMD.

It's almost like you are trolling...

 

[adroc_thurston]

I gave you the size of the market, ~$6.5 billion of client dGPU cards (excluding APU)

Yeah those are gaming.

It seems like you are in denial, and that NVidia's $4.5 trillion market cap started from "gaming".

It started with ORNL Titan in 2012.

How do you know this?

says that on gfx13 POR.
gfx13 doesn't have dedicated DIY gaming parts. just hand-me-downs from other markets.

Because equivalent NVidia die is for:
- client gaming
- cloud gaming
- "professional" workstation graphics
- visualization, rendering
- video editing, post production
- crypto
- data analysis
- AI programming development platform
- client AI
- machine learning

It's for gaming.
Rubin one also serves as a prefill accel.

It's almost like you are trolling...

No one at AMD is interested in gaming.
Especially not Lisa, she outright hates gambling with R&D cash.