I still think 18AP will be competitive with N3P on engineering grounds. The problem appears to be time to market. By the time 18AP is out, N2 will have ramped. Project risk seems to be too high. Frankly, I was very surprised by Reuters article. Sounds like people are telling LIp-Bu Tan very negative things. It's quite sad.
First, sorry to hear about your beavers 😀. Second, thanks for posting this.
Third, I believe we should differentiate between Intel 18 Angstrom being the hot next node for external customers of Intel Foundry Services (it's not, not anymore) and IFS having to get the node shipshape for its internal customer, especially the next generation Intel CPUs for servers. Just canning the entire node and moving on to "14 Angstrom" before fabbing at least one Intel CPU line in 18 Angstrom is simply not an option. A foundry can only announce and then abort an entire node so often before it loses all credibility, including with its owners/shareholders. Intel has to show that they can produce at least one of their own products in a node beyond Intel 3, even if it doesn't beat TSMCs upcoming "2nm" process.
According to this leak https://www.thelayoff.com/t/1k01v9n9t from a now former Intel employee, Intel Foundy has reached ~ 70% yield for Panther Lake in their 18 A Node, and is producing at high volume. Has anyone heard or seen anything that either confirms or disputes that?
It's unclear to me if you're completely unaware or you're just trying to be a controversial figure but your writing and therefore analysis is riddled with bias.
I am indeed very biased. That is why all my positions (including average price) are regularly disclosed. Own a ton of NVDA options, NVDA stock, and AVGO stock. It is up to the reader to disregard what I write if they wish. Since you are here, let's talk about SerDes firmware. Multiple contacts have told me that Marvell has significantly improved the performance of both the 200G optical-side and 200G KR electrical SerDes via firmware updates. In my experience, firmware improvements don't have such a dramatic impact for performance. What specifically did the new firmware improve? Common mode calibration? ADC cal? Skew cal? CDR INL correction? Maybe optimize a LUT somewhere to deal with unexpected process variation?
If one wants to be cynical about it, maybe the firmware update made their 5 nm Chips perform WORSE, which would help make their lackluster 3 nm stuff look better in comparison. Again, just me being a cynic.
Welcome back! We ve missed you 😆
Not back. Still have 2 months left in my break. Hot Chips will be the return to full schedule.
The Beavers! 23 hours to the beavers, and one hour to the stock talk = much happier soul.
He's BACK
Not back. Still have 2 months left in my break. Hot Chips will be the return to full schedule.
Take good rest!
Well, the benefits form new node keep going down, and the cost goes up.
Can Intel be competitive for not topp of the line with 18AP if they manage to fix it?
I still think 18AP will be competitive with N3P on engineering grounds. The problem appears to be time to market. By the time 18AP is out, N2 will have ramped. Project risk seems to be too high. Frankly, I was very surprised by Reuters article. Sounds like people are telling LIp-Bu Tan very negative things. It's quite sad.
Couldn't they compete in cost vs N2?
Dam, should not have sold TSMC stock. TSMC is now a complete monoply for 4-5 years.
First, sorry to hear about your beavers 😀. Second, thanks for posting this.
Third, I believe we should differentiate between Intel 18 Angstrom being the hot next node for external customers of Intel Foundry Services (it's not, not anymore) and IFS having to get the node shipshape for its internal customer, especially the next generation Intel CPUs for servers. Just canning the entire node and moving on to "14 Angstrom" before fabbing at least one Intel CPU line in 18 Angstrom is simply not an option. A foundry can only announce and then abort an entire node so often before it loses all credibility, including with its owners/shareholders. Intel has to show that they can produce at least one of their own products in a node beyond Intel 3, even if it doesn't beat TSMCs upcoming "2nm" process.
According to this leak https://www.thelayoff.com/t/1k01v9n9t from a now former Intel employee, Intel Foundy has reached ~ 70% yield for Panther Lake in their 18 A Node, and is producing at high volume. Has anyone heard or seen anything that either confirms or disputes that?
Do you think Marvell is beyond saving or is there a path to recovery?
If you can tell the Marvell execs some things and they would listen to you, what would you ask them to do?
> Broadcom MAX is better.
Something better than MAX3 PHY?
Noooo
Always love your articles
It's unclear to me if you're completely unaware or you're just trying to be a controversial figure but your writing and therefore analysis is riddled with bias.
I am indeed very biased. That is why all my positions (including average price) are regularly disclosed. Own a ton of NVDA options, NVDA stock, and AVGO stock. It is up to the reader to disregard what I write if they wish. Since you are here, let's talk about SerDes firmware. Multiple contacts have told me that Marvell has significantly improved the performance of both the 200G optical-side and 200G KR electrical SerDes via firmware updates. In my experience, firmware improvements don't have such a dramatic impact for performance. What specifically did the new firmware improve? Common mode calibration? ADC cal? Skew cal? CDR INL correction? Maybe optimize a LUT somewhere to deal with unexpected process variation?
If one wants to be cynical about it, maybe the firmware update made their 5 nm Chips perform WORSE, which would help make their lackluster 3 nm stuff look better in comparison. Again, just me being a cynic.
Xs brain is like a straight line, only one dimension