- Jan 23, 2007
- 1,240
- 290
- 136
I just heard the news that Intel may be buying GlobalFoundries. I find myself wondering: if Intel ends up owning the factory that makes AMD chips, what would keep Intel from prioritizing their own interests?
AMD would likely just end up buying wafers from Samsung if they didn't want to be solely reliant on TSMC.
Samsung does not have the fabs to make anything for AMD.
It would take samsung too long to build fabs and be too expensive to facilitate AMD also, that samsung would probably not find it profitable.
What do you mean they don't have the fabs? They not only produce their own mobile SoCs but are also doing Nvidia's GPUs.
Im sorry but almost the entirety of your post is wrong and your speculation on intel would only make sense if this wasn't gelsingers intel. AMD isn't on 5nm yet and even if they were, samsungs cutting edge nodes are still better than plain old 12nm as nvidias ampere on 8nm has demonstrated. And given that samsung is literally making a SOC with AMD graphics... yeah, samsung indeed does have the facilities to make more than just dram.Fabs and nodes are completely different.
You have a FAB, and you want to make a node, you can not just use the same FAB on a different node.
Samsung can not make a complicated chip outside ARM.
They have not needed to, and china has showed us how difficult it is with there sorry excuse of a processor.
AMD is on a 5nm node, samsung does not have the facility to make something for AMD outside GDDR.
Not even Intel could facilitate AMD.
This is why i am saying intel taking over glofo feels like revenge in losing market share by some salty Intel board members.
What do they need from GloFo? AMD even gave up on GloFo. It seems also like a roadblock to stop AMD from getting back there FABs as well.
AMD isn't on 5nm yet and even if they were
For the record, I don't even think TSMC's N5 and Samsung's 5LPE node are even comparable, at least not on a density basis. TSMC's N5 node is one node ahead of Samsung's 5LPE, which is based on their 7LPP node, and Samsung likely won't achieve density parity until their 3GAE node. Long story short: AMD cannot simply hop from TSMC N5 to Samsung's 5LPE because they are not equivalent. AMD would have to ask Samsung to use their 3GAE/GAP/whatever it's called these days node, which we all know ain't ready yet, and AMD is not going to wait for Samsung to catch up to TSMC.AMD's 5nm Next-Generation Zen 4 Ryzen & EPYC CPUs Rumored To Feature Over 25% IPC Increase, 40% Overall Performance Boost Over Zen 3
With Zen 3 released, all eyes are set on AMD's next-generation 5nm CPU architecture for Ryzen and EPYC CPUs which is codenamed Zen 4.wccftech.com
AMD is on 5nm... if we were speaking hypothetical in lets ask samsung, it would be Zen4 they make, and that is 5nm.
Samsung 5nm node is a LPE
Samsung’s 5nm LPE Node in Production, With 20 Percent Power Efficiency Said in Tow; New Mobile SoCs Expected Very Soon
Mass production of the Samsung’s 5nm LPE has started, with the company’s partners expected to see new mobile chipsets very soonwccftech.com
its not anywhere near designed to make a Complicated A64 processors.
There is that small part of reality being that AMD would prefer this so they're not attached to GloFlo for the next 3 years until their last stage agreement comes to an end. In other words, some old executive who hasn't been following the news or simply not capable of making the conclusion that AMD would prefer not having to deal with GloFlo ever again.Intel securing GloFo, it sort of feels like someone from intel tenure board member wanted to get back at AMD one last time before retiring and decided that assimilating AMD's old fabs is a sweet revenge before retirement.
I think the original goal was to use Samsung's 7nm family of nodes, but it wasn't ready in time so Nvidia used what I presume is Samsung's 8LPP node (with some tweaks so that JHH can fuel his marketing team's claim of a custom node), which was more mature at the time. Rumors are that the next generation of Nvidia GPUs (the one after Ampere Refresh) will finally use Samsung's 5nm nodes, either 5LPE or 5LPP with Nvidia's usual tweaks of course.I'm not sure why Samsung would even be considered in this instance. Didn't they fumble 8nm? There's some rumor that Samsung completely missed their productions goals and came up way short of delivering product to NVidia and are still having trouble delivering enough workable GPUs. I don't know how true that is but it's something I've been told a few times now. This was before last month's news that Samsung was having yield issues.
Source: https://www.hwcooling.net/en/ampere-gpu-deep-dive-whats-new-in-geforce-rtx-3000-architecture/The process is called 8N because it is specifically designed for Nvidia’s use and should include various unspecified improvements and modifications. It should therefore be a better technology than the 8LPP process used by some mobile phone SoCs. According to some sources, modifications done for Nvidia result in up to 10% higher performance (meaning higher frequency could be achieved) than the original version of the process that was focused more on mobile SoCs. However, this process probably achieves lower energy efficiency and transistor density than the real 7nm process, which in the case of Samsung is employs EUV lithography. It’s likely that the 8N process is also inferior to TSMC’s 7nm node.
I'm quoting this and understanding what you said, but you triggered a memory of an article I recall reading about just this and how NVidia was supposedly playing games with the foundries and got screwed. This still doesn't bode well because Samsung is reportedly having yield issues across the node lineups. Companies aren't using them because they have a good product, according to what I've read and been told, but forced to because TSMC has limited free space and Samsung is offering deep discounts.I think the original goal was to use Samsung's 7nm family of nodes, but it wasn't ready in time so Nvidia used what I presume is Samsung's 8LPP node (with some tweaks so that JHH can fuel his marketing team's claim of a custom node), which was more mature at the time. Rumors are that the next generation of Nvidia GPUs (the one after Ampere Refresh) will finally use Samsung's 5nm nodes, either 5LPE or 5LPP with Nvidia's usual tweaks of course.
Source: https://www.hwcooling.net/en/ampere-gpu-deep-dive-whats-new-in-geforce-rtx-3000-architecture/
Yes, I am aware of that article/story. See bullet point #4 below.I'm quoting this and understanding what you said, but you triggered a memory of an article I recall reading about just this and how NVidia was supposedly playing games with the foundries and got screwed. This still doesn't bode well because Samsung is reportedly having yield issues across the node lineups. Companies aren't using them because they have a good product, according to what I've read and been told, but forced to because TSMC has limited free space and Samsung is offering deep discounts.
its not anywhere near designed to make a Complicated A64 processors.
Oh brother, this crap again. I remember when people were saying that TSMC can only make mobile CPUs, and could not make high power desktop or server CPUs. Where did all those people go when TSMC started making high power desktop and server CPUs for AMD?
You're going to be in for a surprise soon as Samsung has been delivering POWER10 CPUs to IBM for RISC servers that will ship later this year. Is the ability to fab a 600 mm^2 die with 18 metal layers and 180 watt TDP enough in your book to make a "complicated A64 processor", whatever that is?
For the record, I don't even think TSMC's N5 and Samsung's 5LPE node are even comparable, at least not on a density basis. TSMC's N5 node is one node ahead of Samsung's 5LPE, which is based on their 7LPP node, and Samsung likely won't achieve density parity until their 3GAE node. Long story short: AMD cannot simply hop from TSMC N5 to Samsung's 5LPE because they are not equivalent. AMD would have to ask Samsung to use their 3GAE/GAP/whatever it's called these days node, which we all know ain't ready yet, and AMD is not going to wait for Samsung to catch up to TSMC.
This article is two years old but it highlights the differences between TSMC and Samsung.
sigh... no comment...
Im just laughing inside at how many guys here think a FAB is like a large scale 3D printer, able to make anything as long as there is a working template.
OK if you guys think that...
Did you completely miss this guys post? From your comment apparently you did.
Samsung currently as of now and most likely until the end of this year can not make Ryzen cpu's unless they overhall the 5nm FABs they have over to ~N5 and not LPE...
And even then, making a new fab is probably cheaper, and it still takes 2 yrs+ to construct.
LPE incase u didnt know is even below LPP, so its not even a single step away for Samsung to be able to produce similar silicon as TSMC.
I honestly love how many of you guys are really underestimating how advance TSMC is, and why China wants it so badly.
But here read this guys comments again, he nails it without getting too complicated and technical on why i said that comment.
Fabs and nodes are completely different.
You have a FAB, and you want to make a node, you can not just use the same FAB on a different node.
Samsung can not make a complicated chip outside ARM.
They have not needed to, and china has showed us how difficult it is with there sorry excuse of a processor.
AMD is on a 5nm node, samsung does not have the facility to make something for AMD outside GDDR.
Not even Intel could facilitate AMD.