RX 7900XT, 15,360 cores, MCM, Tapeout Q4

[Grim5]

There is of coure another way to approach pricing:

1. How much would it cost to make
2. And, if wafers are still in short supply (which they will be...) how much profit could they make producing CPUs instead (this is mainly revelant for AMD)?

Let's say AMD go all out and produce a 600mm² monster for the flagship.
We can plug that into a wafer yield calculator (taking the 0.09 defect per cm which is something TSMC mentioned for 7nm - although that was a few years ago now):
https://caly-technologies.com/die-yield-calculator/
Now 600mm² is 24.5mm x 24.5mm, which gets us this:

For the die (ignoring testing, packaging, VRAM, board etc.), the only other consideration is how much wafers cost.
TSMC 7nm was meant to be about $10,000 per wafer but prices are rumourd to actually have gone up a bit.
6nm might be very similar rather than costing more as EVU might be super expensive but throughput and amount of masks is reduced.
5nm will probably be more, maybe $15,000 per wafer.
Anyway, $10k get us to $190 per die, $15k to around $280.
(This ignores that with redudancy not all dies with defects will be wasted, although against that not every good die is able to hit clocks/power targets.)
Those numbers aren't that bad actually: in theory that means a dual chiplet card might be doable for around $1,200 or so even with the $15k figure.

However, AMD at least have more profitable things they can do with wafers and that would be Zen3 or Zen4 CCDs or even APUs.
The current Zen CCD is 83.7mm² which means the wafer costs between $15 and $23 ($10k / $15k). Zen 4 might be a bit bigger, and of course the IO dies come on top of that.
In other words, CPUs have far higher margins.
We see that with Zen3 vs Navi21. A 5600X @ $300 versus a RX 6800 @ $579 using the $10k figure gives margins of 95% and 72% (with $15k per wafer it is 92% and 58%).

Hard to see how things will play out.
AMD seem to madly focused on margins ATM even at the cost of marketshare and mindshare. The design and mask costs are fixed, so once wafer are available it makes no sense to price too high, but what do we know?

Just an FYI

As of 12 months ago, TSMC was charging $16,986 per 5nm wafer. Since then they've increased their prices twice but by how much I don't know.

Various tech reporters have tried to dig out what the price changes were - the average from what I'm reading is an estimated 20% increase from pre to post covid.

So 16,986 + 20% after TSMC's increases = $20,383

So Redo your calculations using $20k or $21k per 5nm wafer as the cost and then you'll have some idea of RDNA3 prices and margins

 

[CAT-THE-FIFTH]

AMD just made record margins and profits. Its quite clear any production/shipping price increases have been swamped by the price increases in their own products. No different than Intel or Nvidia it appears!

 

[KompuKare]

Just an FYI

As of 12 months ago, TSMC was charging $16,986 per 5nm wafer. Since then they've increased their prices twice but by how much I don't know.

Various tech reporters have tried to dig out what the price changes were - the average from what I'm reading is an estimated 20% increase from pre to post covid.

So 16,986 + 20% after TSMC's increases = $20,383

So Redo your calculations using $20k or $21k per 5nm wafer as the cost and then you'll have some idea of RDNA3 prices and margins

$20k? That's a lot - those EUV machines are going to be paid for soon then.
This is from the spreadsheet I used now with $20k wafers too:


Obviously, Zen2 CCDS is 7nm so not $20k per wafer.
The same for the consoles. We don't know now much the charge Sony/MS but they aren't going to be loosing money on those consoles. (Although maybe opportunity cost as the consoles are eating into scarce wafers.)
Ran various dies through the die yield calculator from 600 down to 100 in 100 increments just to see.

 

[KompuKare]

Well, can always make GPUs with a chiplet of 83.7mm² as well?
Sure, you have the vRAM and other "stuff", but if doable for smaller sizes, then normally AMD should be able to play around with them in low / mddle end as well. Theoretically, perhaps, at lower prices than current gen.

That is the holy grail for GPUs with increased wafer and mask prices. Have one small chiplet size die which can support low/med/high/ultra with various combinations.
I would imagine if they are ready for that, the chiplet would be between 100mm² and 150mm², but we will have to wait.

 

[muon]

The lovelace and rdna3 rumours are pretty crazy.

 

[humbug]

AMD just made record margins and profits. Its quite clear any production/shipping price increases have been swamped by the price increases in their own products. No different than Intel or Nvidia it appears!

From 46% to 48% margins, off the back of $3.85 Billion with just short of $900 Million free Cash flow for this quarter, that's up 25% over last quarter and up 99% over the same quarter last year.

So yes, however they are selling more to OEM's and Consoles.

Their R&D cost has also gone up 40%, no doubt preparing RDNA3, Chips for Tesla Cars, Samsung Mobile and Steam Deck for Q3, 4 and Q1 2022 at which point they are forecast to break $20 Billion annual revenue, its $16 to $17 Billion for 2021.

AMD are still growing, rapidly, but given its at 2 percentage points margin growth and with all the new products AMD are expanding into i would argue while yes prices have gone up but its more about that expansion, Intel's margins are still falling, they admit its due to tougher competition, from who is that?

Competition is good but its tainted by a lot of craziness right now, as we all know, and that competition is NOT yet secured because while AMD are growing to a $20 Billion revenue company Intel are still racking in $70 Billion and until AMD become a whale of similar size they are still prey.

 

[CAT-THE-FIFTH]

From 46% to 48% margins, off the back of $3.85 Billion with just short of $900 Million free Cash flow for this quarter, that's up 25% over last quarter and up 99% over the same quarter last year.

So yes, however they are selling more to OEM's and Consoles.

Their R&D cost has also gone up 40%, no doubt preparing RDNA3, Chips for Tesla Cars, Samsung Mobile and Steam Deck for Q3, 4 and Q1 2022 at which point they are forecast to break $20 Billion annual revenue, its $16 to $17 Billion for 2021.

AMD are still growing, rapidly, but given its at 2 percentage points margin growth and with all the new products AMD are expanding into i would argue while yes prices have gone up but its more about that expansion, Intel's margins are still falling, they admit its due to tougher competition, from who is that?

Competition is good but its tainted by a lot of craziness right now, as we all know, and that competition is NOT yet secured because while AMD are growing to a $20 Billion revenue company Intel are still racking in $70 Billion and until AMD become a whale of similar size they are still prey.

OEM/console sales are lower margin,so it kind of explains the relatively large price increases for DIY parts,as higher margin aspects of the business help prop the other aspects up. I fear we will be seeing much higher prices even next year and the year after. We also all know Nvidia and Intel are quite complicit in all of this,so its a bit concerning longer term WRT to this hobby.

 

[Dave2150]

From 46% to 48% margins, off the back of $3.85 Billion with just short of $900 Million free Cash flow for this quarter, that's up 25% over last quarter and up 99% over the same quarter last year.

So yes, however they are selling more to OEM's and Consoles.

Their R&D cost has also gone up 40%, no doubt preparing RDNA3, Chips for Tesla Cars, Samsung Mobile and Steam Deck for Q3, 4 and Q1 2022 at which point they are forecast to break $20 Billion annual revenue, its $16 to $17 Billion for 2021.

AMD are still growing, rapidly, but given its at 2 percentage points margin growth and with all the new products AMD are expanding into i would argue while yes prices have gone up but its more about that expansion, Intel's margins are still falling, they admit its due to tougher competition, from who is that?

Competition is good but its tainted by a lot of craziness right now, as we all know, and that competition is NOT yet secured because while AMD are growing to a $20 Billion revenue company Intel are still racking in $70 Billion and until AMD become a whale of similar size they are still prey.

Intel out supply (and outsell) AMD over 20:1 overall, in all market segments. With their potential accquisition of Global Foundries, make that 50:1.

AMD are just out the picture without their own manufacturing capacity, as it doesn't matter how good a design of CPU/GPU they make, they'll never be able to produce enough to make an impact.

Apple, Nvidia, Intel and others, all have deeper pockets to pay TSMC more for each wafer.

 

[humbug]

The 5800X is a better CPU than the 11900K, and with it the 11700K, the later of which is the same price as the 5800X.

The 5600X, a fantastic CPU is £240, don't forget 4 core 8 thread CPU's used to cost £350.

Yes right now Intel have better sub £175 CPU's, the 11400F is a brilliant £150 CPU, but try and find one, this is what Intel do, its what they did with all the Pentium G CPU's, get them on the bar charts, get everyone talking about Intel and then stop making them, why? Because actually having £150 CPU's that perform like £250 CPU's is really bad for your bottom line.

You know what Intel did to AMD the first time AMD overtook Intel, AMD could not defend them selves from that onslaught and it contributed to their near bankruptcy, AMD need to bulk up before they really go on the attack.

 

[EsaT]

Intel out supply (and outsell) AMD over 20:1 overall, in all market segments. With their potential accquisition of Global Foundries, make that 50:1.

GloFo is second tier, and not even really that, fab company and its acquisition wouldn't boost Intel's capacity really much.
Actually Glofo even doesn't have any top level node, but only old one and even older ones.
But maybe Intel could rebrand those nodes with arbitrary number of + characters in end for one more respin of 2015's 6th gen Skylake...

Again TSMC's 2020 revenue was 48 billion USD and that's all from fab business.
Which is more than all other fabs combined:
https://www.statista.com/statistics/867223/worldwide-semiconductor-foundries-by-market-share/
And suspect TSMC has about whole GloFo worth of modern node capacity under construction or conversion.
They're simply that much bigger fab company.

 

[KompuKare]

Is @Dave2150 still plucking figures out of thin air?
20:1 is 5%, yet AMD have around 20%.
50:1 would be 2%.
Pure fiction.

 

[Purgatory]

The lovelace and rdna3 rumours are pretty crazy.

You mean like with every generations rumours before ? ...

Reality expect the same old 20-30% boost from generation to generation . Business as usual...


Meeh don't believe any of the rumours until I see it come out of the horses mouth. This always ends up with disappointment when the real product is out and in the hands of reviewers. Next generation I expect one thing to be true, the prices will be silly.

 

[humbug]

Is @Dave2150 still plucking figures out of thin air?
20:1 is 5%, yet AMD have around 20%.
50:1 would be 2%.
Pure fiction.

He, just..... i don't know.

 

[Harlequin]

None of my screens support Freesync so what amd sells for a graphics card does not work for me unless they support gsync.

If AMD ever supports VRR then it can work with 1 of my screens but not the others, though AMD has been promising VRR over HDMI for years and stil nothing

Thats your screen supplier not AMD

 

[humbug]

I was a little confused reading that from @Grim5 but let it slip my mind, because, well, i must have misunderstood it, right?

AMD does and has supported VRR over HDMI since its inception.
There are literally 100X more Free-Sync screens now than G-Sync.
Your problem, probably, is that you have an early G-Sync screen which has a silly chunk of hardware in it that locks it to Nvidia.

 

[Wrinkly]

None of my screens support Freesync so what amd sells for a graphics card does not work for me unless they support gsync.

If AMD ever supports VRR then it can work with 1 of my screens but not the others, though AMD has been promising VRR over HDMI for years and stil nothing

The newer Gsync modules support Freesync.

 

[Harlequin]

I hoenstly doubt RDNA3 will be on 5nm - i`ll hedge its on 7nm+. Also GloFo are a major tier 1 supplier for AMD - without them, Ryzen wouldnt exist.

 

[humbug]

The newer Gsync modules support Freesync.

Gsync modules? That's odd, as far as i was aware G-Sync / Free-Sync are the same screens now not because Nvidia's G-Sync Modules support Free-Sync, but because Nvidia did away with the modules to take up Free-Sync, they still call it G-Sync but they have no modules in those screens.

 

[humbug]

I hoenstly doubt RDNA3 will be on 5nm - i`ll hedge its on 7nm+. Also GloFo are a major tier 1 supplier for AMD - without them, Ryzen wouldnt exist.

What makes you say that? 5nm has been in mass production since Q3 last year, The rumour is AMD start initial production of RDNA3 in Q4, so its design complete and they are producing chips but that doesn't mean they will hit the market weeks later, its probably 3 to 6 months to test the final production product, build up enough stock for a launch and turn them in to actual GPU's to sell on mass.

RDNA3 is two 7,680 shader dies, compare that with the 6900XT at 5,120 shders, that's how you get 3X the number of shaders, 15,360, 7nm, + or not is not dense enough to make GPU dies with 50% more shaders, the size and power consumption probably wouldn't allow it, they have to be 5nm.

 

[Harlequin]

What makes you say that? 5nm has been in mass production since Q3 last year, The rumour is AMD start initial production of RDNA3 in Q4, so its design complete and they are producing chips but that doesn't mean they will hit the market weeks later, its probably 3 to 6 months to test the final production product, build up enough stock for a launch and turn them in to actual GPU's to sell on mass.

RDNA3 is two 7,680 shader dies, compare that with the 6900XT at 5,120 shders, that's how you get 3X the number of shaders, 15,360, 7nm, + or not is not dense enough to make GPU dies with 50% more shaders, the size and power consumption probably wouldn't allow it, they have to be 5nm.

Because Apple have N5 tied up till 2022, literally 90% of the entire node. Actually i think AMD will use N6 (5 layer EUV version of N7 - N7P isnt compatible with N7, but N6 is). It`ll be Q3 2022 before anyone gets anything more than early production on N5 once Apple fully moves away to N3 Intel and Apple tipped to be the first to adopt TSMC's N3 node-based chips | TechSpot