Discussing Intel's Earnings, TSMC 3nm, and GPU Production Costs with Daniel Nenni

How is Micron outcompeting Samsung and SK Hynix on volume production DRAM density, while they don't have EUV? What exactly is their "computational lithography" that allowed them to scale DRAM without EUV Thag their rivals seem to require?

KarbinCry

Ahhh, didn’t know that. Curious how long their protect Taiwan will last, hopefully nothing happens but it seems something’s going to happen sooner or later.

XTX 999

The US government will do everything they can to support Intel. The chips act is evidence of that. They won’t bailout intel because that would be tremendously unpopular. Other American companies would buy up Intel before that happens. Apple, Amazon, Microsoft, etc all have plenty of cash to do it. Honestly I believe one or more of those companies will make big investments in Intel soon.

John Iracheta

Hypothetically speaking, if Intel were to be in deep financial trouble and they decided to file for bankruptcy, do you think the US govt will come to rescue and bail them out?

ManBearPig

TSMC intentionally keeps their best tech inside their borders to deter their destruction. That's why TSMC is called Taiwan's "Silicon Shield". Sure they'll be happy to build some 6nm fabs in other countries - but it you want 4nm or 3nm....gotta protect Taiwan.

Moore's Law Is Dead

I’ve seen it mentioned many times that TSMC is making Nvidia “pay” for their wafers, to my understanding a big reason is because nvidia decided to use Samsung for ampere. How does nvidia’s current relationship compare to TSMC faithful like AMD or Apple? Do you have any insight into how TSMC decides to treat each individual customer since it doesn’t seem to be a level playing field for those who are involved.

XTX 999

Hello Tom and guest, considering the geopolitical location of TSMC, why have they not built fabs in other locations? I know fabs are big investments but considering what they are charging for their products, I can’t imagine they are low on cash.

XTX 999

Hi guys, first question: does Intel becoming TSMC's second largest customer ahead of AMD say anything about the state of 5 nodes in 4 years or about IFS 2.0? Second question if I may: How comparable are Intel and AMDs datacenter segments and can comparing their respective revenue numbers tell us anything useful about their relative positions in datacenter?

Robert S Barnes

How would you rate the design of Intel’s Sierra Forest and Granite Rapids server processors? While they do have a sort-of chiplet design, I currently expect the main CPUs to be enormous. That will help with performance and efficiency but it will make them extremely expensive to produce. I expect the top configurations to be in limited supply and probably have half the price/performance of AMD’s equivalents. It feels like they’re going for benchmark wins regardless of the design or manufacturing cost.

Chris Rijk

How do EDA tools interface with different foundries, are they an overarching system with "plugins" for different foundries and nodes? Do foundries distribute their own EDA tools/versions for their nodes? Is the EDA side a large part of the challenges intel is facing in getting IFS off the ground?

qhfreddy

How much do you think TSMC is squeezing the market? If Intel gets their fabs to be competitive and starts securing contracts do you think TSMC will be forced to cut prices? Do you think this would effect newer or older nodes more and why?

Samantha Vimes

If you had to give a percentage chance how likely do you think it is that Intel is able to become a successful fab for outside companies in the next 5 years? If you think the chance is low, what do you think the roadblocks are? If you think they can what gives you that optimism?

Samantha Vimes

What has been the impact of the Chips Act so far? Are the new fabs being constructed in the US enough to change the manufacturing landscape?

Samantha Vimes

What do you think of AMD’s MI300 design? What aspects of the design are you most interested in? (In my case, I wonder about the power delivery to the top chiplets and what trade-offs that might have required). Do you think that something like this would make for a nice general purpose compute platform, in the sense that you just have to design your own custom compute chiplets to stack on top of an existing infrastructure? Would you like to see AMD offer the MI300 base as a platform for other companies to innovate on top of?

Chris Rijk

Let’s say for the sake of argument that AMD sells the Ryzen 7700 to distributors for $250-300. Using a simple silicon calculator and rumours of TSMC wafer costs suggests that the silicon cost would be around $40-45, implying that AMD has massive profit margins on these parts. How accurate do you think silicon cost estimate really is? Also, that doesn’t include packaging, testing and a number of other steps. What would you estimate the true manufacturing cost to be, including all factors? What would you estimate the total R&D costs of Zen 4 on desktop to have been? (The x86 and I/O chiplets basically). Side note: I estimate that about 80% of AMD’s consumer processor sales are APUs with very thin margins, given the overall profitability of AMD’s consumer group (which actually made an operating loss in Q4 2022) and the size of the laptop market relative to the desktop market.

Chris Rijk

What is the perspective looking like for FD-SOI?

KarbinCry

What would, in your estimation, be the rough ratio between TSMC's N5 family throughput, and their hybrid bonding throughput?

KarbinCry

Who wins the advanced packaging race? Both in terms of players - ASE+SPIL, Amkor, TSMC, Samsung, Intel... and in terms of technologies - Si interposer vs organic interposer, hybrid bonding vs microbumped stacking...

KarbinCry

What are your thoughts on the use of ML tools in semiconductor EDA, with recent announcements from Nvidia and Synopsys? How much will these impact chip design quality, and development cost for a desired design performance? Lots of interest is on the big HPC leading edge usecases, but could these advances breath new life to trailing nodes like 14nm or even 28nm, perhaps extending their lifetimes in many markets?

KarbinCry

Hey Tom and Dan. AMD and intel have plans to integrate specialized AI hardware in client products soon. What kind of innovation is happening to utilize this hardware on PC? Do you think unequipped products may have an earlier than usual obsolescence?

DeadOfKnight

Hi Tom and Daniel, Jim Keller and Sam Zeloof recently launched a new silicon manufacturing startup called Atomic Semi. Judging from Sam's previous work, they're likely gearing up to use electron-beam lithography for maskless fabrication for low-volume and prototype silicon. There has been a recent rise of open source EDA tools, such as OpenLane, and even open source PDKs for trailing edge processes, such as Skywater 130. Over the coming years could we see silicon design and manufacturing become as accessible to researchers and hobbyists as PCB design and manufacturing is today? Or is it too niche to ever become cheap or sustainable enough for this kind of business to flourish? Thanks for all the hard work.

Kloud

What are realistic options for effective stacked die cooling? Or is there any way to reduce temperature sensitivity for higher clocks on cores with VCache?

FloridaMan

How much do you think TSMC will bump up pricing for 2nm if unchecked by a foundry with similar silicon density? If Intel opens up their foundry for others to use, what is Intel's potential for price competition with TSMC?

FloridaMan

What do you think of TSMC’s 3nm delays? Do you think it was a once-off? A sign of TSMC faltering? An indication of just how hard process design is becoming? What do you think of AMD’s decision to use 4nm for standard Zen 5 and 3nm for Zen 5c due to TSMC's 3nm issues (as leaked by Tom)? Could the same thing happen again with Zen 6 and 2nm?

Chris Rijk

How much do you think AMD spent bringing Zen 4 to market? (Referring to the desktop and server versions only). AMD have said that Zen 4c will be the same ISA as standard Zen 4, so presumably a different physical implementation of the same logical design. How expensive do you think it was to design Zen 4c on top of Zen 4? Do you think this is a better approach than Intel’s big-LITTLE designs?

Chris Rijk

Hi Tom and Daniel, in-memory processing has been talked about for years, but without any mainstream products. What have you heard about it's use?

Gach

Hi Tom and Daniel, I've read that the measurements applied to nodes, such as 4nm, don't really measure transistor size. What do they measure? Or is it just marketing? Are there differences between nodes that claim to be the same size? If so, what are they?

Alex Sloter

Hi Daniel, welcome back. You were last on in Q3 2021. Since then Intel’s revenues have dropped from $19.2Bn to $11.7, their gross margin has dropped from 56% to 34.2% and their net income from $6.8Bn to -$2.8Bn. In your opinion, how much of this is due to Intel’s own mistakes, how much is due to the competition and how much of this is due to the economy?

Chris Rijk

I assume a lot of people did not notice this so I thought I'd point it out. "Before moving to Business Unit results, I will highlight a few changes made within our segment reporting. The Client and Data Center focused products from the former AXG business are now reported within our CCG and DCAI segments, respectively, and will have a dilutive effect on the operating margins of those businesses. Our Silicon Photonics and Foundry Automotive businesses have moved out of NEX and IFS, respectively, and are now reported as part of all other revenue, sharpening our focus on the significant market opportunities available to both NEX and IFS." Not the best look to continue to shift around reporting of business divisions while the company is still going down hill.

Reginald Aryee

"DCAI margins were also diluted by the merge of the AXG business and inventory reserves tied to the exit of our Server System business." ...is AXG no longer competing in the server market? https://seekingalpha.com/article/4597564-intel-corporation-intc-q1-2023-earnings-call-transcript

Reginald Aryee

Hey Daniel, Hey Tom. I'd be interested to know how multi-threading could be improved in the future, both from an architectural and scheduling perspective. I'm asking cause physics simulations will just get more taxing, same es AI and other game-logic, so an effective scheduler and hardware-improvements seem rather necessary. Thanks and take care!

Dave Scholze

How long will it take Intel's oneAPI offerings to be as good as NVIDIA's more mature SW offerings around CUDA? James Hamilton of AWS mentioned how servers are getting smaller and SoC'd and goes on to say that even though it will take time, a server is going to be a system on a chip. It will all come up off the board and land on a chip, he says what happens in mobile ends up happening in servers and just takes five to ten years. Your thoughts on this.

What FPGA will recommend for a beginner and what (maybe other) FPGA you will recommend for prototyping for GDDR7.Also when do you thing new (independent )foundries will pop up .Also what do you thing about the x-ray lithography and can this kind of lithography be applied to smaller scale and be more easy to execute (for example because it is far more easy to manipulate a beam of electrons than beam of photons ).

Valko Milev

How compelling do you think IFS' "lower end" and planar offerings are? Glofo seems to be on a rampage with their profits from that sector, does intel genuinely see that as a market they can push into?

qhfreddy

Do you think AMD is feeling enough competitive pressure from the non-x86 server/HPC landscape? Do you think they should try to be more aggressive on that front so x86 can retain market share?

qhfreddy

How important are partners of TSMC like AMD and Apple when it comes to pushing packaging innovation?

qhfreddy

Do you think "the west" can really ever be self-sufficient when it comes to semiconductors as the politicians seem to want to make it? Is there really a desire to go into passives and PCBs? Can we even make a self sufficient electronics industry in terms of raw materials? Do we have enough young people with the skills or willing to learn the skills?

qhfreddy

What are the similarities and differences between where you see intel going and where IBM has gone?

qhfreddy

Hi Daniel and Tom, With all the attention paid to processors and GPUs, I wanted to shift gears to NAND flash. Given that the majority of NAND production is concentrated into a few companies, are we looking at a future where those companies just keep shifting position in market share? Or are we looking at a situation where one company, like Samsung, comes out miles ahead of everyone, like how TSMC seems to be far ahead of Intel and Global Foundries? Also, are there any exciting developments in NAND flash that we might be looking forward to? (Higher densities, more stacked layers, more efficient controllers, etc) Thanks!

Woody Chang

Hello Tom & Mr Nenni, Mr Nenni mentioned on SemiWiki that TSMC 3N was "a very important node" compared to TSMC 5N. Could you elaborate on why? Is 3N going to last a very long time? Does it have specific impacts on the industry in other ways than its lifecycle?

KingHarkinian

Hey Daniel and Tom. So I realized that yoy quarterly revenu shrinked by 7 million which is approximately what AMD and Nvidia makes in one quarter. I mean this is exciting. Soon we will have 3 competitors that earn +- the same. So the competition will have to be about talent not the money. Is this the bottom for Intel? How will they keep financing their ambitious IDM 2.0 strategy? It's not like they are announcing new partnerships to back it up.

QuickJumper

Also just a treat my dad was an ASIC design Engineer for almost 3 years for the USA government.

MelodicWarrior

Oh and also what do you think of Japan now trying to get back into the ring with Fabs after being on the side for so long? (4 Fabs in 6 years)

MelodicWarrior

Hi Tom and Dan Since Intel is planning to refresh raptorlake again for desktop, will the Intel 4 process (7nm) be ready for it? If not what can we expect from Intel 7 again regarding performance and efficiency? It would be nice to see Intel return to their Tic Tok strategy so they can be more competitive with AMD.

Blake Sasturain

Welcome back to the show sir! Always a pleasure and honor to have you on! I hope you were able to were able to find some peace despite the recent situation. Anyway I remember the last time you were on, you thought AMD would not reach almost 40% server until way later this decade. However thanks to recent events, it seems like AMD is not the really the only one to experience major shifts in market share and mind share. Given where everything is going, HPC hardware wise, I am starting to believe we entering a time when a lot of unpredictably is going to be the norm for the semi-conductor industry. So far it seems, even in the retail space, Companies like AMD and Qualcomm are able to supply OEM's and AiB's with inventory a lot faster than the "traditional ones" these days. What is your opinion on how those kind of companies are going to be prioritizing Inventory supply going forward? Will they need to take serious stock or learn the lessons of Intel and Nvidia of not betting too much?

MelodicWarrior

Hi Tom and Dan! How sustainable do you think Nvidia's insane margins are going forward? On one hand gamers are fed up with the pricing and lack of VRAM and on the other companies like Microsoft are going out of their way to design custom hardware, also how do well do you think Tenstorrent and AMD are positioned to compete with Nvidia in the next few years?

Matheus Duque

Hi Tom and Dan, what kind of changes do you expect will come from the transition to high-NA EUV and the lower reticle limit? Obviously everyone has to get onboard with chiplets for many more segments, but how do you think they might diverge from AMD’s approach?

DeadOfKnight

Hey Tom and Dan! My question is about which foundry's node is more dense when we compare: intel 7 vs tsmc 6/7 Intel 4 vs tsmc 5/4 Intel 3 vs tsmc 3

Falto