Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Intel Eager to Avoid Clashes With Apple As It Looks to Secure TSMC's 3nm Chip Supply
- Thread starter MacRumors
- Start date
- Sort by reaction score
Wow.
Once upon a time Intel ran from success to success, but as soon as they broke with Moore’s Law, and other technologies broke their monopoly, Intel’s been stumbling from strategic mistake to strategic mistake.
I don’t see this move as a change from this pattern.
Ps I think the article was poorly written as it doesn’t easily come across what intel is actually planning.
It seems like Intel wants to contract TSMC as a production vendor using a process Intel isn’t capable of, but I suspect Intel wants to offer TSMC production capacity using TSMC’s process.
Once upon a time Intel ran from success to success, but as soon as they broke with Moore’s Law, and other technologies broke their monopoly, Intel’s been stumbling from strategic mistake to strategic mistake.
I don’t see this move as a change from this pattern.
Ps I think the article was poorly written as it doesn’t easily come across what intel is actually planning.
It seems like Intel wants to contract TSMC as a production vendor using a process Intel isn’t capable of, but I suspect Intel wants to offer TSMC production capacity using TSMC’s process.
They will do what will earn them the highest amount of money (like all companies), BUT they must also handle a complex geo-political situation.
They are opening new plants abroad, IMO, to earn more and to offload the reliance on Taiwan.
Here’s a better headline for this thread about what Intel trying to do:
Homeless man seeks meeting with Bill Gates, wanting to inherit Bill’s entire fortune
Homeless man seeks meeting with Bill Gates, wanting to inherit Bill’s entire fortune
No they are not. They are using TSMC until their fabs are up and running. In a few years they will be at the same as TSMC.
The worst thing you can do as a company is have your revenue dependent on one major customer.
Not really unless you think its Intel's upcoming GPU is an Apple SoC.
Intel is just making sure that TMSC will have also have the required production capabilities for their GPU in addition to what TMSC need to produce Apple's chips.
TSMC would never let Intel host TSMC’s production process; Intel is a competitor, and TSMC is absolutely deadly serious about their trade secrets. When I was allowed to see their production flow data I had to do it on a machine they provided, I wasn’t allowed to bring any electronics into the room, the room had glass walls and I was observed the entire time.
I'm always surprised at the level of hating on Intel whenever these stories pop up. Intel's only problem is that they made the near fatal mistake that almost all deep technology companies make at least once during their existence: Hiring an MBA to be the CEO (Bob Swan), as opposed to someone with a hard-core engineering background. Fortunately Intel has rectified that with Pat Gelsinger...
... I'm rooting for Intel, and I fully expect that by mid-2020's they'll bypass TSMC/Samsung and regain their leadership as the premier chip company.
... I'm rooting for Intel, and I fully expect that by mid-2020's they'll bypass TSMC/Samsung and regain their leadership as the premier chip company.
Everyone here who thinks Intel is on the ropes needs to realize that Intel will never ever NOT explore every single avenue, financial, political or geopolitical to stay on course and at the top. You think Intel is on the ropes…sorry, you’re view is too narrow. Intel hasn’t been around this long without learning a thing or two about playing the long game. Nothing is out of bounds…nothing. Intel is as unpredictable now as they have ever been. The dragon cornered is the dragon most dangerous. There’s a reason why they’ve courted TSMC. I don’t think anyone at TSMC is dumb enough to fall for it, but Intel is never to be trusted. Their ulterior motives are to always be on top, remember that.
Agree with almost everything you said but ~25% is revenue ( or estimated revenue ) not unit shipment or wafer output.
Pat Gelsinger is a proselytizing tool.
Before Apple, the critical partners for advanced nodes were the gate array companies. That's because gate arrays (although quite complex) are more regular and easier to characterize than SOCs and bleeding-edge processors.
Altera uses TSMC for advanced fab. And Intel purchased Altera a few years back. So Intel is already (via gate array production) a bleeding edge customer for TSMC.
BTW, gate array backlogs are one of big causes of the worldwide semiconductor shortage affecting all products.
Wait, wasn’t Intel going to get Apple back as a customer by producing their chips?
Why would TSMC prioritize Intel over Apple when Apple has a symbiotic relationship with them and is funding the new nodes, while Intel has openly stated their intentions to compete with TSMC and is only a short-term customer?
If it seems odd of Intel to claim they want Apple back and also run ads against them, the way they are treating TSMC is hard to justify, relying on them while also writing op-eds like this: https://www.politico.com/sponsor-co...in-chip-production-must-support-us-priorities
Here Gelsinger asks the government to support chips in the USA, but only give money to Intel and specifically not TSMC. Sounds like a great partner.
If it seems odd of Intel to claim they want Apple back and also run ads against them, the way they are treating TSMC is hard to justify, relying on them while also writing op-eds like this: https://www.politico.com/sponsor-co...in-chip-production-must-support-us-priorities
Here Gelsinger asks the government to support chips in the USA, but only give money to Intel and specifically not TSMC. Sounds like a great partner.
Last edited:
There are a couple of rumors/leaks that Meteor Lake "processors" are not a single monolithic die solution. More like the Meteor Lake multiple "tile" ( chiplet) package that servers as a "Processor".
Going to TSMC isn't necessarily for x86 cores.
Intel's Xe-HPC Ponte Vecchio chip is composed of :
" While the compute tile is based on TSMC N5, the Ponte Vecchio Base Tile is on Intel 7....
...We discussed Xe link but this is based on TSMC N7. Typically we see SerDes and switch tiles on older nodes as they can be harder to scale...."
Intel Ponte Vecchio is a Spaceship of a GPU
The Intel Ponte Vecchio is a spaceship of a GPU that required Intel's chipbuilders to innovate across many vectors
The Xe-HPG , Arc Alchemist GPUs are based on TSMC 6. [ basically done at this point. Drivers are holding back the large scale launch at this point. CES 2022 should present some movement there. ] . It is a larger monothilic chip. However, Intel's GPU line up is seriously looped into TSMC advance processes. It has already happened; there is little to no "rumor" there.
If Intel is looking to do a GPU 'tile' for the Meteor Like package , the huge burning question would be why would it NOT be on TSMC????????? The new Xe-core that Intel has developed already has two implementations on TSMC. One in TSMC N6 and another in TSMC N5 . After N6 and N5 going to N3 is a relatively standard ( almost mundane) progression. A second (or third) generation 'new' Xe Core probably would not get much from going to N4. And timeline wise 2023 is a better window to iterate on GPUs for Intel..... so ... N3 . What else would they pick?
A week old 'Rumor"
https://videocardz.com/newz/intel-m...umored-to-utilize-tsmc-3nm-process-technology
If Intel is trying to play "catch up" in iGPU 'horsepower' , then the GPU tile is the one that has the greater need to move forward to pack more "compute" and cache onto the die. Not the CPU cores. In 2023 Intel should have their Intel 4 ( old '7nm') process up and running.
There is a Xe-HPG graphics core generational change coming in 2024-2025 . Maybe that will switch back to Intel fab if they have worked out a way to get back on steady updates, but before then GPU tile is likely sticking to TSMC.
This negotiation is probably more so about timing during the calendar year and Intel not wanting to get into a bidding war over fab capacity. Intel doesn't want to over spend here, but also probably not looking to just eat table scraps that Apple dribbles over the edge either. There is some balance of pre-pay and guarantees that Intel needs to set up to secure the number of starts they need.
By doing the compute (and possibly I/O ) tiles with Intel 4 (and Intel 7 or TSMC N5 ) Intel will need far fewer N3 wafers than Apple would with dies that had both CPU and GPU (and cache for both) on them. Chucking the CPU cores gets more GPU tiles built with a limited number of wafers. The "problem" Intel is likely juggling here is that they don't want to have large mismatch in the supply of CPU and GPU tiles for assembly. They are going to need enough wafers to pair up with the wafers that are going through the Intel fabs. [ If the base/SOC tile is N5 then might be leveraging TSMC packaging. If majority are intel then Foveros/EMIB ]
Somewhat skeptical that Intel is going to ship the bleeding edge 86_64 implementation off the TSMC if they don't have to. Additionally Meteor Lake has been in flight for over a year at this point. It has probably been tagged for Intel 4 since the baseline design started. It would be odd to switch late in the game for such a performance tuned piece.
If they were super hard pressed to do a monolithic die variant then those would be better odds. With tiles they can ship out a subset of the work.
They switched back?
" ... Today however that changes, as Intel is announcing its first fully Intel-designed FPGA, built upon its own internal 10nm process, with the Agilex brand name. This new range of products is set to roll out later this year for sampling, and offer a mix of analog, digital, memory, custom IO, and eASIC variations within a singular platform. ... "
Intel Agilex: 10nm FPGAs with PCIe 5.0, DDR5, and CXL
But part of the whole thing of buying Altera was to turn them into the "in house" fab partner. ( did the same thing with Modems when they bought them. Pushed those also onto the in house fab process. )
I think Mobileye and some of the new AI acquisitions are on TSMC. The discrete GPU products certainly are ( Xe-HPC compute tile is on TSMC N5 )
Intel has farmed out some I/O.
The do have a leading edge. IPU/SmartNic product. with 16 Neoverse cores and I/O that likely is either fully or tiled with TSMC N5 (or better) contirubtions.
Intel® Infrastructure Processing Unit
Intel IPU ES2000 ASIC enables cloud and communication service providers to reduce overhead and free up performance for central processing units.
www.intel.com
You could be correct, but there's also a chance that Intel has survived so long because they have been a partner in what amounts to a locked monopoly of sorts; WIntel. That neither Intel nor Microsoft can seem to shake loose. AMD has been chipping away at it, but only just chipping it. Apple's Macs have also been making a tiny move forward as well.
The only real push away from that monopoly was outside the WIntel PC market... mobile and servers.
This probably isn't about prioritizing as much as not trying to needlessly sync up demand bubbles. Outside of the demand bubble for a Fall rollout Apple probably isn't consuming everything on N3. The A16 SoC probably is not N3 because volume want ramp until 2nd half of 2022 which is too late for September. Takes 70-100 days to walk wafers through the whole process. If want to have volume in August then have to start volume push in April-May. If volume isn't coming until. June-July that is too late. A16 is more likely on some N4 variant.
Some M-series variant and likely not the base , 'plain' M2 is probably going to be first to N3 (also a pretty good candidate for N4 ) . M3 Max or M3 Max2/4 . Those could be a substantively lower volume wafer consumer than the A17 would eventually be late Spring/ Early Summer. Intel could fit in there if ready to go first half of 2023 before M3 and/or A17 fully get going.
The issue for TSMC is how much more N3 capacity to do they bring online after going to "full volume". The "full volume" wafter starts aren't stuck as a single number forever. After a while TSMC will be able to do high volume for multiple customers, it is mainly timing.
There is a notion that Apple just buys up all the wafer starts to completely 'starve out' the competition. That wasn't likely before ( Huawei had N5 wafer allocation before frozen out in the trade-war) and even less so now (fab capacity shortage .... so wasting wafers will draw political carnage for TSMC. )
And as stated in another reply the other likely flawed notion here is that Intel is going to shift there entire processor production needs over to TSMC. Meteor Lake is a tiles/chiplet product. The whole thing probably isn't being done on TSMC. Intel probably just needs a relatively ( to M-series and monolithic old school Intel die) small chiplet. The chiplet and the Apple dies are probably substantially different size dies. Which means a different wafer demand. ( smaller die more product per wafer ... less wafers needed. )