- cross-posted to:
- amd@hardware.watch
- cross-posted to:
- amd@hardware.watch
You must log in or register to comment.
Is it that easy to move a design from let’s say 4nm TSMC to Samsung? I always assumed if something was designed for one, it would be trouble to move it to another. I mean if the size if cache these days isn’t shrinking, and there is a 5% difference or so in logic density between the two 4nm nodes, would that not screw up a design of logic decreases but cache was the same? Or do they just upscale everything by 5%?
From the article it looks like nothing is being moved really. There is a general architecture that is Zen 5c but both TSMC and Samsung variants will be different designs developed independently.
Contract wise, you will need to hire separate teams. People working with TSMC will not be allowed to work with Samsung and vice verse cause these foundries will not want their trade secrets to be leaked out.
Only the physical design. Also there’s no such requirement, not that strict anyways. Much smaller companies have worked with both, e.g. Tenstorrent. AMD can easily manage that.
They logical design doesn’t change. But they have to use Samsung’s process design kit for the physical design (cell creation, placement, routing etc)
are these the samsung’s new GAA or is that coming with samsung’s 3nm later?
i hear samsung’s GAA method is going very well so I am expecting good result
GAA is for 3nm. And it is pretty much done, there are a few designs in bring up.
GAA is 3nm, and reports are that it’s not going very well
Good for Samsung if this is true. With the news that mediatek is dual sourcing some components with IFS it seems like we may be on the way to diversifying our leading edge semiconductors.
Any hints or clues as to what kind of products could be produced at Samsung? Do we know anything about the intricacies of this Samsung node?
3d vcache is out of the question, so are we talking about budget parts? Laptop/OEM oriented stuff?
Maybe lk-99 was real after all LOL
This is good for AMD and Samsung. AMD can move from relying on one sole foundry and probably a sign that Samsung’s 4nm node can play an important role in Zen5c and other products. Hopefully these Samsung components are still competitive efficiency and performance wise. You would imagine AMD are sufficiently pleased that this will be the case.
Waiting for athlon 3000g successor for am5.
Reports have come out that Samsung’s enhanced 4nm process has better yields and the chips they’ve churned out (so far at least) don’t have any blaring problems.
Not sure how all this works but my biggest hope is that Samsung’s 3nm GAA is a homerun and AMD can jump on it.
I want to believe~
It would be nice if the article told us which flavor of SF4 the chips would allegedly use!
Glaring problems. Not blaring, although blaring sort of works I guess.
One would be visual the other sonic?
This is either going to be good, or a complete shit show. It entirely depends on how good Samsung’s 4nm node is
TSMC gave Samsung this opening. TSMC was strictly better all the way from 14nm until recently.
The first wave of 3nm kind of sucks and it shows up in the mediocre improvements generation over generation for the Iphone 15. First-gen 3nm has terrible yields too.
Wouldn’t be surprised if these are just the IO dies while the rest is on TSMC’s 3nm.
Wouldn’t be surprised if these are just the IO dies while the rest is on TSMC’s 3nm.
or things ya know, can change over the years; and Samsung finally is nearing node quality parity.
or it could be AMD managed to widen the already massive efficiency gap against intel enough on the design end; they deem it not worth it to pay for an absolute bleeding edge node. That being said I could believe either way because they also have to worry about risc competitors too in enterprise.
So many variables so I will wait and see.
Why would they go for more expensive dies when the performance improvement is minor?
AMD is truly desperate