AMD Ryzen 4000 desktop CPUs to be fabbed on TSMC N5P?

[CAT-THE-FIFTH]

Lisa Su was talking about mobile,so I think this more to do with the collaboration they have with Samsung,or a mobile part. Roadmaps,even published this year show Zen3 is on 7NM,and not even 7NM EUV.

[watercooled]

Yep, I could be wrong but I think this could be the latest 'official' information regarding the nodes used: https://www.anandtech.com/show/15589...-not-specified

[Oberoth]

Is Zen 3 on 5nm+? At first you think there is no way this is true simply because we have all had 7nm cemented in your mines for the last 4 years! But there are things that we know that could back it up things like:
- 7nm is hugely over subscribed with a long lead time for production. We have RDNA 2, Ampere, PS5, XSX, current Zen 2 chips, plus a load of other stuff all fighting aggressively for capacity. If TSMC thinks 5nm is ready and it has capacity it would make sense to move some customers over to this node so TSMC can maximise production.
- So then the question is, does 5nm have spare capacity? We also know the answer to this and it's a massive YES after Trump's demanded that TSMC stopped working with Huawei which until this happened was TSMC's second largest customer so we know that TSMC must have a huge gaping hole in 5nm orders with Huawei gone and so AMD could easily and happily step in here.
- So then it's a question of, is it possible to got from 7nm to 5nm and how long would it take? This requires a more technical breakdown than I can offer but we do know that many products get a 'die shrink' from one node to another so it certainly possible, as for how long? Well you would think it would take a good 6 months. So have AMD done anything recently to indicate to us that they might actually be stalling for a little more time? Well the answer is another big YES once again, they are bringing an out of the blue but half-hearted Zen 2 refresh on a few key processors to seemingly 'tie us over for a while'. This refresh would look very odd if we were going to get Zen 3 in September but not so odd if it now wasn't coming until early 2021 as they need time to move Zen 3 from 7nm to 5nm.
- Finally, we know from TSMC orders that AMD has ordered a lot of 5nm wafers for production this year and so clearly been working with 5nm for quite a while and full expected to be using 5nm this year, if it's not Zen 3 then what is it?
As for the Warhol and Raphael leaked slide, this could easily just be an old slide that hadn't been updated since Huawei got kicked out and TSMC needed a big 5nm order to replace them.

- On a separate note, it's really important to emphasise that we are only talking about Vermeer being 5nm+ and NOT Milan. This helps AMD make enough chiplets for both the server market and desktop market plus it also means none of AMD's slides or recent presentations were wrong or inaccurate as Zen 3 'Milan' is still 7nm and crucially that Milan won't be delayed which would get shareholders annoyed simply because they don't like last minute changes even when they are possibly for the best.

- One additional point which doesn't seem to be getting any coverage is that it looks like Zen 4 has been delayed pretty badly, Zen 4 was going to be the first 5nm CPU and required AM5. But if Zen 4 isn't coming until well into 2022 or later then AMD needed to change it's 5nm plan anyway. There appears to be mounting evidence that we are going to get a Zen 3+ before Zen 4. Why do I believe this?
- Firstly, the Gigabyte leaked motherboard support slide shows "future CPUs being Vermeer & ...." which is clearly indicates there is another CPU line coming that will be compatible with x570 mobos.
- Secondly, when AMD back-tracked on ditching 400 series compatibility with Zen 3 they issued a statement that said "This is the final pathway AMD can enable for 400 Series motherboards to add new CPU support. CPU releases beyond Zen 3 will require a newer motherboard." Which again, unless there is another AM4 line of CPUs coming after Vermeer then AMD didn't need to make this statement as we would then be on AM5 and even the most die-hard AMD fans can't complain about no comparability here!

[msroadkill612]

TSM are also getting a nice financial leg up from Trump, to fund their promptly agreed, stateside fab.

[watercooled]

If TSMC thinks 5nm is ready and it has capacity it would make sense to move some customers over to this node so TSMC can maximise production.

You do make some good points but this is the crux of the matter really. TSMC can't just move customers to a new node as we've mentioned earlier in the thread (and you have mentioned yourself). It would have to be a decision made by both AMD and TSMC, and would not be without risk and cost. With the increased complexity and design rules of current nodes it's not quite the same of the half-node type die shrinks we used to see quite frequently on mid-cycle product refreshes. Moving from 7nm to 5nm at TSMC would require significant reworking of the processor and would need to bring some significant advantages to be worthwhile. That may well be the case, but it does seem like a long shot?

TSM are also getting a nice financial leg up from Trump, to fund their promptly agreed, stateside fab.

Really? Granted I've not been looking for it, but I've not read anything since the very early, preliminary plans to build a relatively small (by TSMC standards) 5nm fab due to be finished some time around 2024. Not sure why that would impact this story given TSMC are already a very wealthy company?

[Gentle Viking]

subsidization

What the rich people use to get richer and have the tax payer fund it that way too.
Much used in my country and it is one of the main reasons i am embarrassed to be a Dane.

They will also use other names for it, but it is what it is and you the little guy will be the one paying for it, and little Danes pay for a lot of stuff, and they pay a huge price for it too.
And the very worst part is when companies from another country come her and grab money too, some do that a lot and dont pay a dime in tax here.

[albert89]

For the 2 people above Intel would just have to close fabs then... can't see that happening as they have always stated they will never be a fabless company.
They have loads of capacity just on a process that's years behind

That 'load of Capacity' that Intel has at 14nm......++ etc,
is becoming more and more pointless as their TDP's start
to skyrocket for ever shrinking performance gains.

[3dcandy]

That 'load of Capacity' that Intel has at 14nm......++ etc,
is becoming more and more pointless as their TDP's start
to skyrocket for ever shrinking performance gains.

Indeed but that's what they've got to work with...
They will be on 10nm sometime soon(ish) which actually should give them a boost. To be fair gamers don't care so much about the extra TDP's do they which is really the only place I see Intel winning overall when you take off price/performance/lower tdp from AMD

[tygrus]

Intel could outsource non-CPU chips eg. MB chipsets.

[watercooled]

Why would they want to do that?

It's not to say Intel haven't outsourced some things e.g. when they were making some mobile phone SoCs, but moving designs to another fab costs a lot of money - there would need to be a good reason to do it. Capacity is one possible reason and I believe there were some rumours about this, but I believe they ended up going back to 22nm for some of their entry level parts instead. Not certain on that though.

[ik9000]

That 'load of Capacity' that Intel has at 14nm......++ etc,
is becoming more and more pointless as their TDP's start
to skyrocket for ever shrinking performance gains.

Intel have spare capacity? I'd heard they were in meltdown.

sorry couldn't resist. I'll get my coat...

[DanceswithUnix]

They will be on 10nm sometime soon(ish) which actually should give them a boost.

I'm still not seeing any evidence that 10nm will be any use for desktop, only for laptop parts.

The prediction at SemiAccurate ages ago that Intel would only leap forward on server & desktop at 7nm is looking spot on.

[3dcandy]

I'm still not seeing any evidence that 10nm will be any use for desktop, only for laptop parts.

The prediction at SemiAccurate ages ago that Intel would only leap forward on server & desktop at 7nm is looking spot on.

Nope that is true - but then again they are years behind on 7nm as well...
This has all the beginnings of a perfect bad storm for Intel

[DanceswithUnix]

Nope that is true - but then again they are years behind on 7nm as well...
This has all the beginnings of a perfect bad storm for Intel

We can expect Intel's 7nm to be more like TSMC's 5nm, so there is a chance at that point for Intel to catch up. Still a while off, but soon enough to make trying to fix 10nm for performance parts not worthwhile as it would take longer and just divert resources.

[CAT-THE-FIFTH]

Why would they want to do that?

It's not to say Intel haven't outsourced some things e.g. when they were making some mobile phone SoCs, but moving designs to another fab costs a lot of money - there would need to be a good reason to do it. Capacity is one possible reason and I believe there were some rumours about this, but I believe they ended up going back to 22nm for some of their entry level parts instead. Not certain on that though.

I remember this about Intel outsourcing production to TSMC:

https://www.techpowerup.com/249039/i...cation-to-tsmc


However,did it ever happen??

[watercooled]

I've no idea to be honest, but if I had to guess I'd say probably not. I reckon CPUs would have been the last thing they would have considered moving if anything - while porting a chipset still seems quite anti-Intel, it wouldn't be nearly as complex or costly, and would avoid having potentially differently-performing products.

Edit: Wow 2018! I didn't realise it was that long ago!