Exactly. Things change. That’s why all these predictions based on status quo are nonsensical. It was not long ago that x86 has entered the server space and it took it a while to dominate it. What we are seeming now is that people who don’t need x86 compatibility start migrating to ARM. It doesn’t really seem like x86 has much answer either. Intel can do their little dance with big little architecture and pretend it’s a great win but their fast cores are still inefficient.
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 12900HK is faster than M1 Max for both single and multi cores.
- Thread starter sunny5
- Start date
- Sort by reaction score
Exactly. Things change. That’s why all these predictions based on status quo are nonsensical. It was not long ago that x86 has entered the server space and it took it a while to dominate it. What we are seeming now is that people who don’t need x86 compatibility start migrating to ARM. It doesn’t really seem like x86 has much answer either. Intel can do their little dance with big little architecture and pretend it’s a great win but their fast cores are still inefficient.
Apple has no interest in entering the server market. As to workstations, they will either focus on the market segments that are ok with 512 GB RAM (or whatever Apple is willing to build) or they will have a tiered memory approach with fast on-package RAM acting as a humongous cache.
Firefox (formerly Phoenix and Firebird) was started by a high-school teenager with help from a few Mozilla developers. The main team was four people working part-time, and lots of help in testing and building. The good old days.
There were other companies that used to be incumbents, just like Intel is today. Being the king is no guarantee that you will be tomorrow.
Yeah. But then the question is what that does to overall performance.
I don't think this is necessarily desperation. They know that their own manufacturing process will take a couple more years to catch up, so they will partially fill the gap with TSMC. It all lines up with the roadmap they have laid out. As a side benefit, they can aggravate the supply situation for their rivals, which all depend on TSMC.
That's desperation. They had a two-three year lead and Tick-Tock. Until they didn't. Having to go to TSMC is desperation.
If they were truly desperate they'd have given up on manufacturing on the leading edge and gone fabless like most analysts predicted they would. Instead they are doubling down on their fabs ...
Maybe. But it's also possible that they see an opportunity to turn the tables with the switch to GAA transistors in a couple of years, which is an inflection point for the industry. TSMC appears to be slipping on that (and have also delayed the 3nm FinFET-based process).
Integrating RAM on package doesn’t help with latency … M1 RAM latency is fine but nothing special - middle of the road for lppdr4x/5. It *does* however help with power usage but that’s less important on a desktop or workstation so theoretically they could go DDR5 there. But the power costs of doing so and having an extra wide memory bus would be … substantial. So I don’t know what they’ll do. They surprised me with the M1 Pro/Max. When the M1 first came out I didn’t think they’d redo the memory bus so substantially for the larger chips.
It's not actually. The market also includes pads and mini PCs. The desktop market has stopped growing. The pad market has too. Over 50% of the market is notebooks. And most of that market requires decent battery performance. Intel is wide open. Intel has had stable growth for some time. AMD has been growing rapidly. Its revenues are now half of Intels. AMD's trying to buy ARM. Intel's got some safe markets, but I don't think the fastest growing market sector is one of them. And there are some powerful players looking for opportunities who owe nothing to Intel and may want to write their own history.
If the reports are accurate that TSMC is doing large 3nm orders with Intel I can only surmise one or more of the following is true:
1) TSMC is getting an absurd amount of money from Intel, more than even from Apple, which will cut into Intel’s margins. TSMC knows they’ll be harming current customers and helping a direct rival, who will jump ship the moment their own fabs achieve parity.
2) TSMC knows their 3nm fab isn’t that great and is likely to lose customers to Samsung’s equivalent node anyway. Truthfully building new nodes gets harder from here for everyone and I would take all the different fab’s roadmaps with a mountain of salt. From what I’m reading: Believe it exists when production starts and only believe it’s good when you see it being tested and lots of wafers leaving the factory without faults. Maybe that’s too cynical.
3) The CEO of TSMC is susceptible to Jedi Mind Tricks. Not sure about this last one.
They’re talking about specifically servers and supercomputers not the general computing market.
Not directly, but having huge bandwidth avoids latencies caused by congestion when multiple cores (and the GPU) compete for access to unified memory.
Yep.
Aye that’s latency hiding and theoretically I believe they could do that with DDR5, it’s just the power would be huge compared to lpddr5. There’s a reason GPUs go with GDDR for bandwidth even if latency suffers.
3 nm in Q2 2022? This can‘t be right - what am I missing?An interesting twist is that by placing large orders for TSMC 3nm chips, Intel will take away manufacturing capacities from rivals, including AMD and Apple, which are already struggling with supply constraints.
Intel Grabs Majority of TSMC's 3nm Capacity, 4 Products Including A GPU & 3 Server Chips In The Works With First Delivery In Q2 2022
TSMC has reportedly won a huge order of its 3nm process node from Intel who will be using the new technology to develop its next-gen chips.wccftech.com
I wonder if TSMC really does favour Intel over Apple.
For 1) I‘d assume Apple has longer term contracts in place.
2) - why helping out a company that‘s a rival by its very nature, a rival that just uses TSMC to win time to eventually beat them.
3) - why let down Apple by doing so; a humongously big customer without any ambitions of the sort Intel has.
Something does not seem right. Where‘s the error in my line of thought?
I think you're overthinking this. TSMC will take orders from whoever is willing to pay. Given the huge investments they have announced over the next few years they need the capital. Another issue that has probably factored into this is that due to the delay of the TSMC 3nm node it will be too late for the 2022 iPhones.
Intel 12900HK is faster than M1 Max for both single and multi cores.
If the reports are accurate that TSMC is doing large 3nm orders with Intel I can only surmise one or more of the following is true: 1) TSMC is getting an absurd amount of money from Intel, more than even from Apple, which will cut into Intel’s margins. TSMC knows they’ll be harming current...
forums.macrumors.com
I am aware.Intel 12900HK is faster than M1 Max for both single and multi cores.
If the reports are accurate that TSMC is doing large 3nm orders with Intel I can only surmise one or more of the following is true: 1) TSMC is getting an absurd amount of money from Intel, more than even from Apple, which will cut into Intel’s margins. TSMC knows they’ll be harming current...
But again: is 3 nm really ready by Q2 2022? That‘s just half a year away.
And if so, Apple surely has secured its share quite some time ago. How could Intel block Apple‘s manufacturing capacities?
It just doesn‘t add up
It’s true they probably lost the 2022 iPhones but companies still tend to be strategic. TSMC is basically fabrication and packaging. They’re not like Intel or Samsung in that regard who acting as a foundry for others naturally conflicts with our aspects of their business. But that means helping a rival bridge the gap has a lot worse implications for them. You laid those out yourself.
So no I think either Intel had to write them a blank cheque or they were worried about their own node losing customers anyway. As you say the 2022 iPhones will already have to be manufactured on their 4nm process if Apple decides to maintain its roadmaps and still use them (which is the most likely future).
I thought that it was nVidia that was trying to buy ARM.
Correct. Authorities did not yet approve though - and there seems to be quite some resistance
Oh no yeah that part is also suspect. I mean it was *supposed* to be ready by then. But the reports are it won’t be. Which is why Apple is reportedly going to have to use 4nm rather than 3nm for its 2022 iPhones.
Like there’s a lot suspect with these reports of massive order wins from Intel on TSMC 3nm especially given what we know about the relationship between all the companies involved.
Your right ... typing too quick. Their revenue is I think, double that of Intels. And Apple hates them, so posters here say.
But yes - the future is quite open IMO.
Edit - I should add too, that AMD's revenue growth has been quite substantial. Unlike Intel's. I don't know why though - maybe they are buying companies, which is often a false way to appear to be growing. Unless you pick the right company.
Last edited:
We've had something like 7 motherboard replacements over nVidia graphics issues. I can understand why Apple was upset but that was back in 2013 and earlier. Surely they could let it go by now.