AMD to Shift Some of its 4 nm CPU Silicon-fabrication to Samsung from TSMC

[RedBear]

There shouldn't be much difference given they won't be clocking to 6GHz or come insane level, most of them would be tuned for best efficiency so there shouldn't be too much difference between the chips unless 4nm Samsung is really really bad.

Laptop chips also downclock incredibly quickly so there's that.

Mobile phones don't clock to crazy levels and they don't really operate at high frequencies for extended time, but when Qualcomm moved to TSMC 4nm for their Snapdragon 8 Plus Gen 1 they gained both in performance and efficiency. So I wouldn't really hold my breath for any miracle here, probably they're just confident that Intel is sinking and they can safely get a performance handicap (which they're taking for whichever reason, it might be cost, bargaining power, geopolitical tensions, etc.).

 

[Kaotik]

You do realize that ARM isn’t actually a foundry customer. They are not fabbing even a single chip.

Currently not, but according to Financial Times they have put together a team to build them full blown SoCs (they already have IP for most if not all parts from in-house development).
This goes in line with the Intel+Arm announcement regarding Intel 18A, which specifically mentions building reference hardware.
The reference hardware will supposedly be used for marketing Arm IP and Intel fab capabilities rather than to compete with Arm SoC manufacturers.

 

[Denver]

If you look at AMD's non GAAP Q4 2022 earnings (2023 Q1 has yet to be published as of writing), which excludes factors that may impair an investor's view of a company's profitability (like large one time purchases like Xilinx), AMD reported net income of 1.113 billion:

View attachment 294134

In addition, your 3.4 billion figure is off to put it lightly. Looking at the regular GAAP quarterly figures, AMD had a net income of $21 million for Q4 2022. That's not a loss, let alone a loss of 3.4 billion.

View attachment 294135
Maybe AMD is doing this to increase profits or maybe not but the numbers your logic is based on are inaccurate. A majority of the reason AMD took a hit in net profit last quarter is due to it's acquisition of Xilinx, as the difference between non-GAAP and GAAP data demonstrates.



If history is any teacher, I'd say no given that Nvidia failed to pass any savings along to customers when it used Samsung. Companies will only reduce prices when they are force to and no sooner.

I think he was referring to one of Samsung's divisions, not AMD.

 

[Minus Infinity]

You do realize that ARM isn’t actually a foundry customer. They are not fabbing even a single chip.

But is that about to change? They are said to be designing their own SoC for the first time.

 

[trsttte]

Taking a wild guess but the 5 nm + 6nm approach is not going well so they went to Samsung for a unified 4 nm part.

5nm + 6nm approach is going great, just happens to be desktop parts, now rebadged for laptops too. They easily kick raptor lake's ass hands down, but aren't fit for the low power ultra book segment that's much more popular now, they need Phoenix for that. Another year and another generation were AMD laptops are missing in action

but the numbers your logic is based on are inaccurate

The losses he was mentioning were from Samsung, not from AMD

 

[Daven]

But is that about to change? They are said to be designing their own SoC for the first time.

I think its just a partnership for now. ARM’s business model doesnt easily support buying millions of fab chips, marketing and distributing them worldwide.

The idea of the partnership is to provide ARM current and potential licensees with a one stop shop for designing and manufacturing their products.

 

[evernessince]

I think he was referring to one of Samsung's divisions, not AMD.

The losses he was mentioning were from Samsung, not from AMD

Yep, that makes sense. I'll go ahead and delete the prior comment as it no longer applies.

 

[Punkenjoy]

I wonder if the fact that Nvidia, after going to Samsung for the 8nm was back to TSMC and got some 4nm room over AMD 5nm is one of the reason why AMD is moving some of their production to Samsung now.

TSMC used to give privileged access to their new nodes to loyal customers but it look like money talk and they changed their minds.

With Intel getting competitive again and wanting to offer fab services and Samsung looking like they are fixing their yeild issue, we might have some nice competition there. Finally something good for the consumers. (Unless chip maker just boost their margin).

 

[prtskg]

It maybe less about capacity and more about samsung offering a notably better price. With their profits in the toilet, losing 3.4 billion last quarter, they will likely be *incentivising* customers, and AMD will want tot ake advantage of that to raise margins on things like mobile chips.

If TSMC 4nm is under-utilized, then this maybe for next generation mendencino replacement chips.

 

[Count von Schwalbe]

TSMC used to give privileged access to their new nodes to loyal customers but it look like money talk and they changed their minds.

To be fair, Nvidia also had fab space at TSMC since at least the Tesla days. Ampere and Pascal were the only deviations from TSMC, and only part of the product stack were fabricated at Samsung.

 

[watzupken]

I feel TSMC's 4nm is booked out because all the big clients have been stuck on the 4/5nm node for a long time. With 3nm not really taking off and also Apple being the only client for it, the reliance on 4nm will not go away anytime soon. Moving to Samsung may alleviate the problem, but the results may not be desirable. Big clients mostly moved their cutting edge chips away from Samsung to TSMC for good reasons since they are surely paying a premium using TSMC's fab. But it seems like TSMC 3nm may start to become like their 20nm, which is not doing well.

 

[TumbleGeorge]

I, on the other hand, feel that everything is advanced enough with the nodes, and there is an arrangement to empty our bank accounts the maximum number of times before they deliver us hardware with the next level of lithography.

 

[Vya Domus]

Bad news, Samsung's nodes have notoriously underperformed compared to TSMC's.

 

[Denver]

I feel TSMC's 4nm is booked out because all the big clients have been stuck on the 4/5nm node for a long time. With 3nm not really taking off and also Apple being the only client for it, the reliance on 4nm will not go away anytime soon. Moving to Samsung may alleviate the problem, but the results may not be desirable. Big clients mostly moved their cutting edge chips away from Samsung to TSMC for good reasons since they are surely paying a premium using TSMC's fab. But it seems like TSMC 3nm may start to become like their 20nm, which is not doing well.

The main reason was the very low yield (30%) even manufacturing relatively small chips. However, this is already Samsung's third or fourth improvement (5nm > 4nm >4nm>4nm LPP) over 5nm and yields should be minimally decent (>60%) + better efficiency.

Samsung's prices also tend to be lower which can benefit AMD's margins, in the long run.

 

[Chrispy_]

Ugh, this is the problem with free-market capitalism sometimes.

The underdog that offers the competition needed to regulate pricing in the market is being outbid by the wealthier megacorps who made their billions by being monopolistic, manipulative, anti-consumer, and in the case of both Intel, Apple, and Nvidia, fined by various international governing bodies for anti-competitive legislation.

Let's just hope the shift to Samsung doesn't hurt AMD too much...

 

[Tek-Check]

Then it must just come down to pricing, must have a better deal at Samsung.

There's not enough capacity at TSMC for what they need.

so they'll basically make phoenix on a 7nm TSMC like process. Good luck with it amd. We know for a fact what this results in, from qualcomm. But maybe you can do some magic and remove the crapiness from samsung fab process.

Why such nonsense?

 

[Denver]

There's not enough capacity at TSMC for what they need.


Why such nonsense?

People often speak based on guesswork and feelings rather than facts.

In the beginning, Samsung's 4nm process had very low yields (35%) and also the efficiency was about 20% lower than TSMC's 5nm, so it lost an important customer: Qualcomm. But currently Samsung is already on the third or fourth improved variation of its 4nm process, and it is already quite similar in all aspects including performance to the competitor's 5nm.

 

[Tek-Check]

This also means that AMD has been aware for many months that they wouldn't be able to get enough wafers from TSMC. If they are smart, they designed the Phoenix chip with two foundries in mind so they need not waste an entire year now, or more, to port the design to Samsung's foundry.
(To be clear, I'm not saying that a single design can simply be copied to another stack of floppies and mailed to Korea; I'm saying AMD has had enough time to prepare.)

We do not know any details of AMD's negotiations with Samsung. Your "conclusion" is uninformed.

 

[Wirko]

We do not know any details of AMD's negotiations with Samsung. Your "conclusion" is uninformed.

Indeed we don't know many details of AMD's negotiations and current contracts with TSMC, either. We aren't a public forum of industrial spies here, we're just trying to make some logical conclusions from the little information we have (and that information can be wrong).

The assumptions and conclusions I made:
1. Wafer manufacturing takes about 3 months. Production planning must be longer term than that, for example, a foundry should have a detailed procurement, manufacturing and sales plan for the next ... 6 or 9 months? They know, with good certainty, who will order how many wafers of each process node in that period.
2. A foundry shares some important details with their strategic customers, and AMD absolutely is one of them. I mean details about the current state of a node, expected yields and available quantities for a similar period of several months in advance.
3. So AMD knew about the shortage well in advance, and had time to prepare and design Phoenix silicon in two versions, for TSMC and for Samsung, early in the development process. Which is better than having to hurry and make a chip that's less optimised, larger, and more buggy. (Is it about Phoenix at all? Probably, but that's just an assumption, too.)

 

[Tek-Check]

Indeed we don't know many details of AMD's negotiations and current contracts with TSMC, either. We aren't a public forum of industrial spies here, we're just trying to make some logical conclusions from the little information we have (and that information can be wrong).

The assumptions and conclusions I made:
1. Wafer manufacturing takes about 3 months. Production planning must be longer term than that, for example, a foundry should have a detailed procurement, manufacturing and sales plan for the next ... 6 or 9 months? They know, with good certainty, who will order how many wafers of each process node in that period.
2. A foundry shares some important details with their strategic customers, and AMD absolutely is one of them. I mean details about the current state of a node, expected yields and available quantities for a similar period of several months in advance.
3. So AMD knew about the shortage well in advance, and had time to prepare and design Phoenix silicon in two versions, for TSMC and for Samsung, early in the development process. Which is better than having to hurry and make a chip that's less optimised, larger, and more buggy. (Is it about Phoenix at all? Probably, but that's just an assumption, too.)

We don't know whether Samsung's yields were good enough at the point of finalising chip design. We know that Qualcomm left them precisely because their initial 4nm was not that good.