dullard
Elite Member
- May 21, 2001
- 25,898
- 4,485
- 126
I strongly recommend Pate.Anyone else interested? It's not a bet, these fine folks are just signing up for a gourmet event!
What is Intel's definition of "manufacturing ready"?
TSMC publicizes when mass production starts or will start, so you know the first production wafers will come out ~3 months after that. Intel being "ready" for manufacturing could mean the start of what TSMC terms "risk production". AFAIK I have never heard of Intel refer to "risk production", because when operating their own fab once they were getting enough saleable chips per wafer to be worth dicing/testing they were putting those aside and were likely shipping to customers before they reached a yield good enough for TSMC to enter "mass production".
If Intel's "manufacturing ready" is roughly equal to TSMC's "start of risk production" then a lengthy gap after that date to general availability of chips made in that process would make sense.
Saylick
Diamond Member
- Sep 10, 2012
- 3,923
- 9,142
- 136
Wait. You and @SiliconFly aren't betting against each other? So if Intel uses TSMC for Arrowlake's compute die then both of you get to enjoy some cat food? What happens if they don't use TSMC? Nothing? If so, that's some courage y'all got there. That, or you relish the taste of cat food. 🤔
Saylick
Diamond Member
- Sep 10, 2012
- 3,923
- 9,142
- 136
If @Thunder 57 and @SiliconFly lose that bet...Anyone else interested? It's not a bet, these fine folks are just signing up for a gourmet event!
(can't use the actual quote due to no swearing policy)
Don't underestimate cat food. It may turn out to be delicious.
Statement for what exactly? For Panther Lake? Do you expect Panther Lake coming just a few months after Arrow Lake-S? Do you really think this is realistic? Maybe you should tone down your expectations a bit. I don't think this ever happened on a same platform. If they can put out Panther Lake-S 1 year after Arrow Lake-S that's great.
20A readiness is meaningless in this context. The first 20A chip isn't coming before the end of 2024 or maybe early 2025. They will use a 6+8 die for ARL-S on 20A and it wouldn't be the first time they are launching 6+8 non K a few weeks or months after the big 8+8 or 8+16 version. We may get something on 18A earlier than that but it's not Panther Lake, the lead vehicle is Clearwater Forest on 18A.
Which they can do if they get enough good chips out of them running at sufficient frequency. One could argue TSMC is doing that with N3, based on their KGD deal with Apple instead of selling them wafers as they have always done in the past. Maybe they feel they CAN'T get yields high enough with N3, which N3E is designed to address.
Intel doesn't have to wait for yields to reach 90% or whatever TSMC's traditional mass production benchmark is. They could sell when yields are 20%, just in lower quantities. That's why we often see just a few SKUs from them at first, and they later release more when their production ramps up due to both running more wafers and improving yields.
intel waiting over a year and a half before releasing arrow lake s on desktop from raptor's refresh? what pat gelsinger is going to return jesus again for a third coming next?
IMO, Intel was able to ramp up I4 process in 4Q22 and sell MTL by mid23.
However, delivering EUV scanner had been delayed.
www.reuters.com
In 3Q22 earnings report, Intel said "Intel 4, 3: Continues to progress on schedule".
But in 4Q22 earnings report, Intel cleverly avoided saying such expression.
However, delivering EUV scanner had been delayed.
Intel staff in Ireland offered unpaid leave in cost-cutting drive - Business Post
Up to 2,000 staff at Intel Corp's Irish operation have been offered three months' unpaid leave as part of cost-cutting measures, the Business Post reported on Thursday.
In 3Q22 earnings report, Intel said "Intel 4, 3: Continues to progress on schedule".
But in 4Q22 earnings report, Intel cleverly avoided saying such expression.
Attachments
I'm going to say that was MTL just not being design ready rather than the fabs holding them back (this time at least).IMO, Intel was able to ramp up I4 process in 4Q22 and sell MTL by mid23.
However, delivering EUV scanner had been delayed.
Intel staff in Ireland offered unpaid leave in cost-cutting drive - Business Post
Up to 2,000 staff at Intel Corp's Irish operation have been offered three months' unpaid leave as part of cost-cutting measures, the Business Post reported on Thursday.
In 3Q22 earnings report, Intel said "Intel 4, 3: Continues to progress on schedule".
But in 4Q22 earnings report, Intel cleverly avoided saying such expression.
I think the desktop market should be maintaining status quo until Zen 5 is released. Only then, the clock starts really ticking and Intel could be in a world of hurt.
I doubt there is any sort of Zen 4 refresh in the cards, only some lower end models + a potential 7600x3d.
that depends is what q amd decide to release. i don't see it happening in first half of 24 like i did a coule months ago. q3 possbility, it'll be be before arrow lake. pricing and performance matter here but if they are necked with each other in many fronts it may be a pricing war where consumerism flourishes. board prices will still suck
I have no evidence that MTL design got tangled up.
BTW, in generally, tiled arch tends to show one digit fewer bugs than monolithic arch.
Intel has announced tape in, tape out, power on dates of MTL. Line those up with the RPL dev timeline, which was publicly released, and you will see that MTL on the design side was unable to launch late 2022 or even likely early 2023.
For example, Pat announced MTL power on April 29 2022. RPL powered on 15 months before launch. April 29 + 15 months = July 29 2023 (Q3 2023). Add in a couple months for a longer design process, more time in Post Si Validation, etc etc (since MTL is a larger change than RPL was) and you would see a late 2023 launch for MTL, exactly where it looks to be launching (late Q3, though that might not be 'real' volume).
Do you have the corresponding dates for ARL, to get some range of expected ARL launch dates?
Intel Moving to Chiplets: ‘Client 2.0’ for 7nm
In theory, on the one hand, ADL needs "3-4 years of Dev time", but on the other hand, MTL needs "2-3 years of Dev time".
RPL diverted ADL design. So, it is'nt an instance appropriate for this debate.
Intel stated in the Q2 2023 earnings call that they have 20A ARL silicon running in the labs, whatever that means. Not going to attempt guestimating with that bcuz we have something much better to go off of- LNL tape out in late 2022. LNL is essentially a specialized ARL, and both are claimed to launch by 2024 according to Pat Gelsinger.
For some reason, Intel hates us so they decide to be extra difficult by mixing up the terms tape out and tape in. They did this before as well, so it's not like they started confusing tape in and tape out starting with LNL. It's kinda annoying.
Pat almost certainly meant they taped in LNL in late 2022, which ye, RPL taped in ~2 years before launch as well. On track for late 2024.
Since LNL is arguably more complex than ARL, if LNL is on track, I don't doubt ARL is as well.
RPL took 2.5 years of dev time. ADL took 3 years.Intel Moving to Chiplets: ‘Client 2.0’ for 7nm
www.anandtech.com
In theory, on the one hand, ADL needs "3-4 years of Dev time", but on the other hand, MTL needs "2-3 years of Dev time".
RPL diverted ADL design. So, it is'nt an instance appropriate for this debate.
MTL taking 2-3 years of dev time is essentially the same time it took for RPL from development to launch. Using RPL timeline as a template for this debate is perfectly acceptable.
Also, MTL has been confirmed to launch to Q3 this year. MTL's tape in, tape out, and power on dates have all been announced by Intel themselves. Its timeline fits. I'm not making this stuff up, literally search up the individual events and dates if you want to yourself. Sure, chiplets in theory might shrink down design timelines, but with MTL, it appears to have taken a similar amount of time as RPL/ADL.
MTL literally was not design ready for a late 2022/early 2023 launch date. Idk what to tell you. Pat also mentioned that MTL will PRQ in Q3, so that would mean the time between PRQ and power on would be still ~15 months. That means MTL spent around the same amount of time in post-si validation as RPL did (slightly more actually). Intel has been announcing engineering milestones, it all lines up.
Plus Exist50 (god bless his banned account lol) has been complaining about MTL's design woes for a while on this thread. It apparently wasn't smooth sailing.
Intel announced MTL tape-in (= fabless company's tape-out) May21.
Also, Intel announced MTL power-on in 3Q21 earnings.
I think that they had enough time for debug&validation.
wccftech.com
wccftech.com
www.cnet.com
Also, Intel announced MTL power-on in 3Q21 earnings.
I think that they had enough time for debug&validation.
Intel Hits A New Milestone In Development of 7nm Meteor Lake CPUs, Tapes-In Meteor Lake Compute Tile
Intel announced this week that it has achieved a crucial milestone in the development of its next-generation Meteor Lake 7nm CPUs.
wccftech.com
Intel Achieves Power-On For Next-Gen Meteor Lake CPU Compute Tile, Delivers Outstanding Performance According To CEO
Intel reported earlier this week that the company was successful in powering a single compute tile on the new Meteor Lake CPU.
wccftech.com
A look inside Intel's mammoth Arizona chipmaking fab
Take a tour inside one of the highest-tech factories in the world, to see how Intel manufactures microprocessors.
www.cnet.com
Attachments
Last edited:
Just for fun, based on the discussion back over at the Zen 5 thread, I was interested in how GLC effected clocks in relation to WLC.
Because of how rare power limits/scaling tests for non-high end parts are, and how GLC P-core only CPUs are rare, I had to do some digging and ended up comparing the 65watt 12400f vs a power-limit removed 11400h and found the 12400f to clock slightly higher (Jarrods Tech), while also having significantly higher IPC (18%). Sure, Intel 7 is supposed to be a 10-15% jump in perf too, but I still think it's pretty nice. Maybe I shouldn't bully GLC so hard online lol (I say this while typing on a 12900h laptop).
Because of how rare power limits/scaling tests for non-high end parts are, and how GLC P-core only CPUs are rare, I had to do some digging and ended up comparing the 65watt 12400f vs a power-limit removed 11400h and found the 12400f to clock slightly higher (Jarrods Tech), while also having significantly higher IPC (18%). Sure, Intel 7 is supposed to be a 10-15% jump in perf too, but I still think it's pretty nice. Maybe I shouldn't bully GLC so hard online lol (I say this while typing on a 12900h laptop).
MTL tape in in mid 2021 would mean ~27 months would have passed between tape in and launch. RPL took <24 months, and while ADL isn't known specifically, even if we assume that the full extra 6 months of dev time ADL took went between tape in and launch, ADL would have taken <30 months.Intel announced MTL tape-in (= fabless company's tape-out) May21.
Also, Intel announced MTL power-on in 3Q21 earnings.
I think that they had enough time for debug&validation.
Intel Hits A New Milestone In Development of 7nm Meteor Lake CPUs, Tapes-In Meteor Lake Compute Tile
Intel announced this week that it has achieved a crucial milestone in the development of its next-generation Meteor Lake 7nm CPUs.A look inside Intel's mammoth Arizona chipmaking fab
Take a tour inside one of the highest-tech factories in the world, to see how Intel manufactures microprocessors.
So ye, MTL's schedule fits very similarly with ADL/RPL, which should be pretty disappointing considering Intel was claiming that by disaggregating their tiles in chiplets they would be able to shrink their dev time by ~1 year. GG go next lol.
It might have been possible for MTL to launch early 2023 if their design teams had it in them too, but they couldn't. I wouldn't blame the node guys on this one.
dullard
Elite Member
- May 21, 2001
- 25,898
- 4,485
- 126
I both quoted and underlined the specific statement that I asked you to back up (and then specifically mentioned that I wanted more on the underlined section). You are correct, I should tone down my expectations for people to have some sort of evidence instead of just making up what they want (as if it were fact just because the thought crossed their mind and/or was something they hoped was true).
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 24K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes Discussion Threads
- Started by Tigerick
- Replies: 21K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter