Soldato
Zen IO dies are already made on a different process node, no reason why they couldnt be made by a different fab.
-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
AMD on the road to recovery.
- Thread starter humbug
- Start date
Zen IO dies are already made on a different process node, no reason why they couldnt be made by a different fab.
The IO dies are already made by a different fab (12/14nm GloFo). What's the relevance?
Not really feasible when there's such a large gap between TSMC and others. You'd have to design your uarch/chips for the worst fab you want to use, it takes away a big advantage that AMD has over Intel. This is also the case in mobile chips right now, ARM can't design cores that would require N5, they have to design it to be built on worse nodes, so even when Mediatek can secure N5 they're not getting the full benefit as they're using uarchs/chips designed for worse nodes.
There's so much competition for TSMC's wafers right now and AMD not being as cash-rich as some of their competitors has definitely limited their bidding power. With Qualcomm planning to come back to TSMC things might get even more difficult. But things can change as AMD just got another credit rating bump (to A-/A3) which should help them raising funds allowing them to at least outbid the likes of Qualcomm and Mediatek.
No thanks
AMD 14nm Chips used half the power with at twice the performance of Intel 14nm chips.
Currently Intel's 10nm + chips use 3X as much power per core vs Zen 3 (7nm) and Intel claim thier 10nm + is "At least as good as TSMC 7nm" even renaming it "Intel 7"
Intel has far too bigger problems for AMD to worry about them.
These arguments that AMD will always be outbit by [Insert name here] at TSMC have been going on since the day AMD switched to TSMC for CPU's and yet TSMC revenue growth has been the result of their partnership with AMD more than all others except Apple.
For too many years so called industry experts have been making far too many perditions about which way AMD will meet its demise or stop them from moving forward, to such an extent its a meme in its self now as these people are increasingly viewed as clowns, AMD continues to defy these people to such an extent it makes them look like a joke, at least some of them now to keep what is left of their shredded reputation have stopped commenting on AMD like this entirly, the rest of them need to do the same before they become the meme.
None of which is reason for AMD to go backwards and use inferior fabs.
But AMD are outbid by others -- this is a fact. TSMC's N5 node has been available on a large scale for more than 18 months and AMD doesn't use it, others are paying more for that capacity. N5 will be almost 2 years old when AMD begins to use it with Zen 4. The reason is simple: others can pay more.
While AMD is TSMC's third largest client (after Apple and Mediatek), they are only 4.4% of TSMC's revenue. Hardly the engine for TSMC's 3x growth in the last 10 years. Up until 2021, the growth was almost entirely Apple and Huawei, together they were over 40% of TSMC's revenues in 2020 while being more or less 0 just 10 years earlier.
https://www.tomshardware.com/uk/news/amd-becomes-tsmc-third-largest-customer
I'm very optimistic about the future of AMD. It's also true that AMD has not been able to secure TSMC's best fab technology due to being outbid by others given that they don't have the cash resources of some of their competitors. This has been and continues to be a struggle for them. These are not contradictory.
Soldato
You saying it's not feasible....
You misunderstood. You can design different chips for different fabs, that happens all the time. And it's irrelevant. Nobody claimed all chips out of a company has to be fabbed identically.
What I said was unfeasible was to design the same chip (e.g. AMD's Zen 4 core chiplets) to be built on a wide range of fab technologies with different characteristics (e.g. TSMC 7nm, or GloFo 14nm). If you want to be able to build it on all of them, you're limited by the characteristics of the worst fab. Your design is always limited by your fabrication capabilities. Reduce those, and your design (i.e. uarch) will suffer.
AMD's supply issues are because they are limited by the production capacity of TSMC's 7nm node. There are no alternatives from other fabs that are as good. AMD can't really go to GloFo to build more Zen 3 chiplets, as GloFo doesn't have a comparable fab technology to TSMC's 7nm, only TSMC does. Even Samsung's 5nm is not comparable to TSMC's 7nm.
No zero on the Net Income
Look at the decimal ........
@HACO Out of interest what would you think about AMD buying their old fab back?
Shut up you
Look at the decimal ........
Shut up you
Big mistake imo. GloFo has nothing to offer to AMD at this time beyond some cheap 12/14nm wafers (which GloFo licenses from Samsung) for IO and motherboard chipsets, and AMD already gets these from GloFo for a good deal due to their partnership (extended a couple of months ago until 2025 now). GloFo abandoned their 7/5/3nm plans so not that useful for AMD.
Buying back GloFo would only make sense if AMD were willing to throw a whole load of money at it to get it back somewhere near the TSMC or even Intel level, and that's the exact reason they sold it in the first place
Sure relying on TSMC limits supply, itself limiting growth, but in some ways that's not the worst problem to have...
Sure relying on TSMC limits supply, itself limiting growth, but in some ways that's not the worst problem to have...
Buying back GloFo would only make sense if AMD were willing to throw a whole load of money at it to get it back somewhere near the TSMC or even Intel level, and that's the exact reason they sold it in the first place
Sure relying on TSMC limits supply, itself limiting growth, but in some ways that's not the worst problem to have...
GloFo abandoned their R&D of newer nodes, their business model is now to rely on licensing outgoing node technology from Samsung and TSMC. That $30 billion would be better spent for AMD to secure capacity at TSMC.
Samsung's 3nm GAA (due end of this year, if not delayed) could be at the same level as TSMC's N5/N5P. AMD could multi-source their Zen 4 fab Samsung ends up delivering, but that's a big if. Their 5nm was supposed to rival TSMC's and we saw how that turned out.
Last edited:
That was my thinking, AMD could make a mixture of products, buy up Samsung's / TSMC's old 7nm / 8nm node and make lower end products with them, use TSMC 5nm for higher end products.
They could sell what capacity they aren't using, GloFo are still making a profit even on 14nm, AMD could slowly build up their own foundry business, again. There is demand for it.
They could sell what capacity they aren't using, GloFo are still making a profit even on 14nm, AMD could slowly build up their own foundry business, again. There is demand for it.
So you believe Intel's Atom (& chipsets) strategy of using older nodes for budget chips was always doomed?
What we currently have is AMD totally abandoning the budget sector, with nothing under 5600G aside from some hard to find Athlon 200's. That's one way of getting away from the budget brand, but that's lot of money they are leaving. Not having a strategy on what to do with older and cheaper nodes is a massive failure.
Yes, it would cost money to part Zen2/3 to 12nm in a quad APU but it could sell in huge numbers.
After all, what Wall Street constantly forgets is that profit is volume * margin not just margins.
Just as everyone else if there is money to be made AMD will sell anything, the problem is capacity, they don't have any so they are using what they do have for higher margin stuff.
That should change with new nodes, 6nm and 5nm opening up to them is new capacity, Zen 3 is still competitive even vs ADL, the 5600G and 5700G are prime examples of what could become more budget orientated chips, just re-brand them and sell them for $149 and $199 once Zen 4 is out.
I don't know about sub $100 CPU's, that would probably need a much smaller chip which AMD don't have in their current line-up.
That should change with new nodes, 6nm and 5nm opening up to them is new capacity, Zen 3 is still competitive even vs ADL, the 5600G and 5700G are prime examples of what could become more budget orientated chips, just re-brand them and sell them for $149 and $199 once Zen 4 is out.
I don't know about sub $100 CPU's, that would probably need a much smaller chip which AMD don't have in their current line-up.
But if GloFo licenses modern nodes from Samsung/TSMC then AMD can use those, no need to buy it.
Reusing old nodes is useful, but not for porting new designs to old nodes. If AMD decides to do entirely low-end microarchitectures (like Intel did with Atom), they could aim for cheaper nodes. But the Atom approach hasn't been successful for Intel either.
It's not about the cost of porting it, it's probably not even possible to port Zen 2/3 to a node that is 1/3 the density and 1/4 the efficiency of TSMC's 7nm, even if possible to run that chip at reasonable clocks that die size will be huge and even a quad core would be a 300w+ chip, making it useless. That chip could cost AMD more than Zen 3 chiplet on TSMC. AMD are not idiots, if it was that simple, they'd have already done it. They already have as much GloFo capacity as they would ever want given their deals and agreements.
Things are falling into place for AMD to solve their scale problem this year.
TSMC is increasing N5 capacity continuously, and if TMSC manages to deliver N3 for release in Q4 for new iPhones, that also frees up N5 even further which will be great for AMD and Zen 4 supply. With lots of other mobile chips and AMD itself moving to N5 there will be lots of extra N7 capacity as well which AMD can use for Zen 3 mobile, workstation and server SKUs.
The risk here is other players might also wanting a piece of that extra capacity, especially Nvidia and Qualcomm.
You're right that is a risk, but i do think AMD are looking to diversify their manufacturing as a way to increase capacity, greatly, it doesn't need to have everything on the best node available to them (5nm) or even the second best node (6nm) AMD have the technology to make competitive products on nodes that are multiple generations old, like 7nm if they are not in the high end.
No, at least not before about a week ago when AMD overtook Intel initially.
only revenues and profits matter
Soldato
No zero on the Net Income
isn't 0.316 billion, the 316 million?
full stop one more place over?