I think the CPU Competition is fine except in gaming segment the GPU not so much.
It is the number I have heard from quite a number of sources. Also, your assertion that N3E prices are at 20K/wafer may be incorrect as TSMC has announced price increases across the board recently.Bruh the translation is not good I think he meant $725 Million to tape out a chip on 2nm process technology.
For 30K Wafer Price it is 1.5X increase vs N3E for a 15-20% PPA Improvements lets say TSMC charges relatively more due to being a Monopoly maybe 25% more it would be still 25K Wafer assuming N3E is 20K Wafer.
The market will shrink if they do.I think the CPU Competition is fine except in gaming segment the GPU not so much.
Also everyone will increase prices if the source increase prices.
The market will shrink if they do.
They will buy the cheaper stuff with the lowest PITA.Most of the market is Corpos and their Corpo machines. They will just buy older stuff instead.
TSMC wafer was 18-19K at one time and than it increases again thanks to no competition.It is the number I have heard from quite a number of sources. Also, your assertion that N3E prices are at 20K/wafer may be incorrect as TSMC has announced price increases across the board recently.
Intel is not hell bent on going out of business can you tell me with a single Tile in their entire product stack using N2 external in 2026. They have already spent a massive amount of Money on 18A R&D and the equipment and stuff in 2024. This year the the costs will be related to ramping the fabs if anything Intel will be in a comfortable position in 2026 assuming Products don't f** up.The market will shrink if they do.
This is the point I have been making all along. AMD isn't likely to go "all in" on N2 and damage their bottom line. Intel seems hell bent on going out of business through this approach and I think AMD is more than happy to let them do it.
18A is much more expensive than N2 if costs are correctly accounted for IMO.Intel is not hell bent on going out of business can you tell me with a single Tile in their entire product stack using N2 external in 2026. They have already spent a massive amount of Money on 18A R&D and the equipment and stuff in 2024. This year the the costs will be related to ramping the fabs if anything Intel will be in a comfortable position in 2026 assuming Products don't f** up.
If you are accounting for Cost to customer maybe but for Intel I don't think so and the cost to produce cause N2 pitches are more smaller than 18A also the BSPDN is helping them to direct print.18A is much more expensive than N2 if costs are correctly accounted for IMO.
This is true but you are also forgetting TSMC is doing this with 60% margins with those shrink this also contributes to the cost.The entire industry needs to realign to the fact that costs are going up exponentially for die shrinks and the benefit is slowing drastically.
Fabs by nature is a volume biz the more the better.If ANY company relies on cutting edge lithography to sustain their lead, it will do so at the expense of its profitability. The market price for a computer will NOT go up simply because cost does. People will opt for a less expensive product.
it's impossible lol 45K will translate to worse PPA/$ it's just fake rumors.If you think the 30K is bad, word on the street is that A14 will be 45K/wafer.
Keep in mind demand keeps increasingThis is true but you are also forgetting TSMC is doing this with 60% margins with those shrink this also contributes to the cost.
Yes and also the fact there is not another choice for them even Intel pays premium and on this Topic TIL today Intel was the first US customer of TSMC . They helped them initially 🤣Keep in mind demand keeps increasing
Apple, qualcomm, nvidia & broadcomm get first dibs as they are ready to pay a premium
AMD has used premium nodes only for the dense cores so far
The cost to the consumer is all that matters IF you are talking about market share.If you are accounting for Cost to customer maybe but for Intel I don't think so and the cost to produce cause N2 pitches are more smaller than 18A also the BSPDN is helping them to direct print.
HP Cell 180nm cell height - 36nm and 5 Tracks
HD Cell 160nm cell height - 32nm and 5 Tracks
For N2/N2P
HD Cell 130nm we don't know the track but assuming it's 5 we get a pitch of 26nm which is not possible without double patterning.
HP Cell 156nm - this can be done with single patterning if it's 5 tracks if it's 6 it will require double patterning 26nm MMP in case of 6 tracks.
This!Keep in mind demand keeps increasing
Apple, qualcomm, nvidia & broadcomm get first dibs as they are ready to pay a premium
AMD has used premium nodes only for the dense cores so far
I currently see no reason to believe that this trend will change over the next 5 years as there is no competitor that can do what TSMC is doing ..... at any price. I don't like it any more than you, but it doesn't help the analysis look any better.This is true but you are also forgetting TSMC is doing this with 60% margins with those shrink this also contributes to the cost.
I don't see any economic theory that suggests that PPA/$ must remain constant.it's impossible lol 45K will translate to worse PPA/$ it's just fake rumors.
No it's not. Your original thought process was that AND wouldn't use N2 for anything, then you got to a place where you could agree they would for server but only dense cores.This is the point I have been making all along. AMD isn't likely to go "all in" on N2 and damage their bottom line. Intel seems hell bent on going out of business through this approach and I think AMD is more than happy to let them do it.
I agree 100%. Always bet on profit maximization. Even when the result is better performance the underlying goal is profit maximization - you can charge more for better performance. This is why Intel took their foot off the gas when AMD was lost in the Bulldozer wilderness. Their only competition was their own last generation products, so better performance beyond "just enough so you can see some daylight between the new and previous generation" did not meaningfully increase revenue so it wasn't worth effort to do better than "just enough".
AMD seems to be gaining market share hand over fist in DC where the profits are very high. I am betting they continue this march and design Zen 6 for DC. It might even be that AMD gives up the high performance desktop (other than gaming) and laptop for DC margins. If Zen 6 breaks into the OEM business in a big way (due to price/performance) and DC feeds the profit machine, this seems like a pretty decent strategy.
My point is that AMD could decide to lose to Nova Lake for client in the interest of profit. After all, if AMD holds out long enough and remains profitable, Intel might take itself out of the game. An Intel without fabrication might well take a round or two of design to get its ducks back in a row (as an example).
Link please. Perhaps you are confusing me with someone else?No it's not. Your original thought process was that AND wouldn't use N2 for anything, then you got to a place where you could agree they would for server but only dense cores.
Ok, perhaps that is poor wording."Gives up" why?
Link please?Wait, gives up profit (in much bigger client market)? In interest of smaller profit in
a much smaller datacenter CPU market?
I think you are overthinking this. Or you may be trolling us.
Which is a function of manufacturing cost you are forgetting even if Intel 18A cost more than N2 considering multi patterning which I doubt, they are taking all the fab margin with them along that in products.The cost to the consumer is all that matters IF you are talking about market share.
The cost to the OEM is all that matters IF you are talking about profit.
If you can't cover your costs, you eventually go out of business. The technical details no longer matter.
This!
Yeah this is something I agree on.I currently see no reason to believe that this trend will change over the next 5 years as there is no competitor that can do what TSMC is doing ..... at any price. I don't like it any more than you, but it doesn't help the analysis look any better.
Exactly you are believing rumors of a process that is in R&D phase it can be the cost of test wafer but not actual Wafer Cost in HVM.I don't see any economic theory that suggests that PPA/$ must remain constant.
Having said that, this rumor evidentially came out of China which would benefit from TSMC advanced nodes being considered "out of reach" globally.
They only raised prices for 5nm and below afaik.On the other hand, the jump from N5 tech to N3 Tech to N2 suggests that TSMC is willing to charge increasingly more for something only it can do today. They are also charging more each year for older nodes.
Link please. Perhaps you are confusing me with someone else?
Ok, perhaps that is poor wording.
Link please?
Global x86 server market 120bn/yr
Global x86 client market 35bn/yr
Growth of server market is vastly superior to that of client market .... especially in $/yr.
They might as well be doing a Billion$ in GPU salesKey qualifier: datacenter CPU market.
Datacenter CPU market is ~$5 billion per quarter
Client CPU market is ~ $10 billion per quarter
A lot of "conventional wisdom" in this area is just retardation. You will see this from many otherwise knowledgeable posters. Saying that datacenter profits are somehow superior. But it is just falling for retardation.
For example, AMD datacenter profit was $932 million but it's possible that nearly a half of it was from GPU. So perhaps $500m from server CPU.
Client CPU profit was $496m, nearly equal to datacenter CPU.
This is just quick illustration that "giving up" client "in interest of" server is nonsense.
They might as well be doing a Billion$ in GPU sales
Yeah I don't even consider Intel's AI/GPU DC sale to be anything relevant but the CPU for the 90% of the part of the revenue.The breakdowns in revenue are not always available, so there is some guesswork. Lots of non-CPU stuff is mixed in to datacenter revenue.
In case of AMD, it may be somewhere between 1.7 and 2.1 billion in quarterly sales.
For Intel, which added another level of obfuscation by merging another division into "datacenter", CPU may account to ~2.5 to 3 billion.
For Intel datacenter GPU is effectively zero (maybe $125 million). For AMD it may be between 1.3 billion and 1.9 billion. Just rough, ballpark numbers.
If leakers are telling you "no they're using N2 on everything except budget/mainstream stuff" then what's the point in attempting to deduce the accuracy of this leak? Either they're right or they're wrong. Nobody arrived at that conclusion (that N2 would be used in server/workstation/premium consumer) through deduction.This is the point I have been making all along. AMD isn't likely to go "all in" on N2 and damage their bottom line. Intel seems hell bent on going out of business through this approach and I think AMD is more than happy to let them do it.
If leakers are telling you "no they're using N2 on everything except budget/mainstream stuff" then what's the point in attempting to deduce the accuracy of this leak? Either they're right or they're wrong. Nobody arrived at that conclusion (that N2 would be used in server/workstation/premium consumer) through deduction.
Either the leaks are accurate or they aren't, accept it or don't.
Not really got do with cash, Apples hardware Sillicon engineering VP just wants to be number 1 all things chip related. On the other hand look at Apple AI performance, they suck at that cause the VP for AI is a weak person.I'm like, who cares whether it's N2 or N3. Apple is still going to have the fastest single threaded SoC on the planet, short of some rare miracle. That's what you get when you have mountains of cash.
lol who will dethrone them as the 1t king?? Nobody else is currently even in the same ballpark.Not really got do with cash, Apples hardware Sillicon engineering VP just wants to be number 1 all things chip related. On the other hand look at Apple AI performance, they suck at that cause the VP for AI is a weak person.
It’s all to with a goal and people that demand that change. 2026 is year where Apple might get dethroned as 1t king cause the competition is really high. We got like 5-6 CPU architectures and all on N2.
Nope. Intel had mountains of cash and here we are. It is doing the disservice to the design teams to say it's all thanks to the process.That's what you get when you have mountains of cash.