Permabanned
I predict from March onwards 
-
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 Zen 3 (5000 Series), rumored 17% IPC gain.
- Thread starter AStaley
- Start date
- Status
Soldato
You got problems....he died last month.
Soldato
Some more patents behind tech which will apparently appear in Zen3:
https://www.reddit.com/r/Amd/comments/f26yr0/underfox_on_twitter_finally_im_able_to_publish/
https://www.reddit.com/r/Amd/comments/f26yr0/underfox_on_twitter_finally_im_able_to_publish/
Associate
- Joined
- 7 Apr 2017
- Posts
- 1,762
So I'm guessing that the new 4k series will have a 7nm IO die instead now? If so, whats the chances of being able to run memory faster, or will it make no difference?
Zen 4 maybe. Zen 3, imo not a chance. Whilst they have a much bigger wafer allocation now on both 7nm EUV and DUV than last year, they're not going to 'waste' it on I/O dies when that cannibalises wafer space for CPU and GPU cores.
Also, see below. Not sure how accurate it is, but that's another factor to bear in mind.
Soldato
It's not just the desktop parts though, it is the new laptop U series parts that are all 7nm monolithic dies, and as they ramp up there will be less 7nm capacity left for alternate parts. However as the likes of Apple etc.move to 5nm this year 7nm capacity will be less crowded from the big players.
Deleted member 209350
D
Deleted member 209350
How many other players are currently using 7nm? I thought Apple and AMD would have been the 2 biggest customers
Soldato
HiSilcon and Qualcomm currently use more than AMD, but it is due to change. Also TSMC are targeting over 140,000 wafers in H2, well up from the 110,000 in H1 '20
Soldato
Good thing Zen 3 is moving to 7nm EUV then. Just at the right time too
Pretty ambiguous. But packaging changes to Zen 3 mentioned by TSMC & AMD at a joint event.
Perhaps I'm wrong about it sticking with the GloFo I/O die. Maybe it'll all be on 7nm+
If so, that likely means AMD are confident about wafer supply at 7nm+.
https://www.pcgamesn.com/amd/zen-3-tsmc-chiplet-packaging
Perhaps I'm wrong about it sticking with the GloFo I/O die. Maybe it'll all be on 7nm+
If so, that likely means AMD are confident about wafer supply at 7nm+.
https://www.pcgamesn.com/amd/zen-3-tsmc-chiplet-packaging
Not much, if anything, we haven't heard before really.. AMD still have the WSA with GF, so using the I/O die is a good way to eat up what would otherwise be useless supply of 14nm chips.
Eventually AMD will be able to wriggle out of that contract as GF hasn't gone 7nm yet, or have I heard anything about the near future, so GF won't be able to force AMD to use what will be useless chips in the very near future. If GF somehow announce 7nm, from memory they all(GF/TSMC/Samsung?) colluded on 14nm so I don't see why they wouldn't go 7nm with them too, but it won't be for a few years as I read that the lithography equipment for 7nm+ is sold out for years - even Intel is having to wait for some of these massive lithography machines.
It would be good if AMD made two different designs, one for desktop(8 core I/O included) and one for everything else(8 core no I/O) and have the 14nm I/O chiplet from GF. That I/O chiplet from GF doesn't cost much, I'd guess under $10 each, but to save that on millions of chips may be worth the extra mask set. With 7nm+, I assume it's 7nm with EUV, will save more space and increase speeds a grade or two at the same power, you could use that extra space for the I/O - I don't think it'll happen this time, but it will sooner rather than later.
Soldato
That's exactly what we have right now.
I meant desktop CPU's without the need for the IO die. Yes we have 2xxx and 1xxx CPU's but 3xxx split it into IO and compute chiplets. The only reason to do this I think, or main one, is the GF WSA, now that GF isn't even trying to compete they should let AMD use the best fab for the job - that being TSMC at the moment.
The IO chiplet is a very cool idea, and still is needed for the larger, Threadripper & Epyc, and the top of the line 12/16core CPU's processors. But for everything else 8core CPU's is a large enough market on their own to warrent it's own design - with IO incuded on the same die. As I said, I don't think it'll happen for Ryzen 4xxx CPU's, but in the future, with more market share, the IO die is a lot of silicon, not to say anything about the extra expense with chiplets/packaging, going to waste.
Soldato
No, the point is to move away from monolithic dies and the inherent costs and wastage that comes with it.
I'm not quite sure how you think 16/14/12nm yields over 95% and 7nm yields at 94% and using those same wafers across the full product stack of 3 different target markets is even remotely as wasteful as building monolithic dies that have nowhere near the same yields. AMD can make anything and everything from a 6 core home computer to a 64 core dual socket server behemoth using exactly the same base components from GloFo and TSMC. Not sure what "extra expense going to waste" you're talking about.
What you're suggesting is for AMD actually regress their technology and financial flexibility.
Now, having the IO die made by GloFo may well be down to the Wafer agreement, but their 16nm is bulletproof and there is much more improvement and gains to be had refining the chiplets than changing the IO die, and certainly it's been said there was no benefit at all putting the IO die on 7nm at the beginning. Zen 3 might see GloFo's 12nm come into play.
I'm not quite sure how you think 16/14/12nm yields over 95% and 7nm yields at 94% and using those same wafers across the full product stack of 3 different target markets is even remotely as wasteful as building monolithic dies that have nowhere near the same yields. AMD can make anything and everything from a 6 core home computer to a 64 core dual socket server behemoth using exactly the same base components from GloFo and TSMC. Not sure what "extra expense going to waste" you're talking about.
What you're suggesting is for AMD actually regress their technology and financial flexibility.
Now, having the IO die made by GloFo may well be down to the Wafer agreement, but their 16nm is bulletproof and there is much more improvement and gains to be had refining the chiplets than changing the IO die, and certainly it's been said there was no benefit at all putting the IO die on 7nm at the beginning. Zen 3 might see GloFo's 12nm come into play.
Soldato
Matisse I/O is already 12nm LP, it's Rome that uses 14nm GF
Soldato
Matisse is monolithic. Would be interesting to see Zen 3 move to a 12nm IO die.
Soldato
Matisse is the code name Zen2 based desktop, and so on.
Soldato
Brain fart moment, I thought Matisse was just the APUs. Silly arse.
Soldato
We going to have a faster 7nm I/O with Zen 3. It was on the AMD video posted back in October but now is deleted. If anyone knows how to find archive of YT videos let me know. I have the link.
8 core may look like sufficient and gonna last "for years" right now, but with the current pace, in 2 -3 years it may look like entry level stuff.
- Status