Did not really think of that. Sounds interesting. That could potentially drive them under though if they keep paying for others processes. Sure while the original fab is booked up GF can also supply the same process but that has to cost them money somewhere and should the original fab be under used at any point why not go directly to them rather than GF.
They just know their position in the market. They know they don’t have the money to compete with Samsung and TSMC that’s the main reason I think they are changing their “strategy”. Also yes they are gonna be paying licenses however as they have secured their 7nm production with AMD, ( based on the link provided by @BookrV ) that means they will be still making money.
And with the reduced expenses from R&D they might be in a pretty good spot.
That wont happen soon. There are AMD and nVIDIA on the one side all producing CPU’s and GPU’s on the new 7nm process and the Mobile chip makers Qualcomm, Mediatek etc. producing chips for the biggest market right now. What will most likely happen is Glo. Fo. will be probably only having a small production for the ryzen processors and TSMC wont have enough producing power for all the company’s.
2 Likes
4 posts were merged into an existing topic: /dev/null
Post were removed due to being off-topic, please PM me if you would like to address this issue.
TLDR:
AMD Abandoned GloFo due to performance and power problems on their 7nm (& preceding 14nm) node.
Yield and quality issues with Ryzen 1 where part of that situation (You can blame low clocks, segfault bug and early Ryzen lockup issues on GloFo’s process)
Zen 2 (Ryzen/Threadripper 3000 & EPYC Rome) where designed for TSMC’s 7NM CFF node from the start.
The Wafer Supply Agreement is set to expire in 2020 and AMD has accelerated that timeline.
Once GloFo lost their primary revenue and driving R&D force, they have decided to scale back from cutting edge node development.
Most of the R&D IP will be sold to who knows who.
Henceforth GloFo will continue producing less R&D intensive designs on proven 28nm, 14nm & probably 12nm which it will seek to improve.
The End
5 Likes
Seems reasonable
Extremely disappointing. That the technology has reached the point that it is too hard to come to completion on the new fabs does not bode well for tech geeks.
If this is a foreshadowing we may have to wait for a new process/material before we see big improvements again.
1 Like
I don’t think so.
Global Foundries has been producing subpar silicon for at least a decade. Not to say that I’m happy to see them stop focusing on smaller processes, but I’m not upset by it in the slightest, not that I’ve had some time to digest.
Keep in mind that we will be seeing improvements in their current nodes and that these so called big improvements you’re looking for are misguided, I believe. Firstly, GlobalFoundries is one of four major fabs that was working on 7nm, so we’re still going to get 7nm, but more importantly, the silicon only limits two things: complexity and speed. Typically, you can’t clock past a certain point on silicon of a specific type. We saw that on Ryzen and Pascal. Ryzen being ~4.25GHz, Pascal being ~2.1GHz. The major improvements we saw from AMD came not from a die shrink, but from an architectural redesign. Look at Intel for example; they kept making small improvements to their architecture, shrinking the die, and each change brought just under 15% performance in the best case.
Don’t tell me that die shrinks are a source of big improvements.
To be clear, I’m not concerned about a lack of competition. If we run out of competition, we’ll see the market correct for it. The tech industry is one of the largest cap out there. If we need more fabs, we’ll make more fabs.
From my read on the article.
Global foundries realizes it is behind in 7nm and Customers like AMD have road maps down to 5nm then 3nm.
By the time Global foundries gets the kinks out of 7nm and can make a profit making chips. AMD will be transitioning to 5nm. That would mean global foundries make no money because as they hit profitability production AMD is wanting 5nm then 3nm etc.
I can understand GF point of view if they dont have a big customer using 7nm on a longer term to recoup the R&D getting it working well.
The upside is as Consumers AMD will be puting out smaller / faster chips over the next serveral years and Intel will be racing to compete. All good for us.
Would be nice to have more foundries but this tech is so expensive. If you not on the wave surfing the profit. Your behind the wave paddling by burning money. A bit like Intel is burning money now trying to get there next fab working and it looks like late next year. Which is so far away.
i see 2 outcomes from this:
-
They are going to work with us agencies, and screw all (since IBM moved to GloFo; this may be meaning something…)
-
They are out of CPU/GPU race.
I talked to a PhD semiconductor something about this and he said that a die shrink in itself doesn’t really increase performance per se.
What we usually get is lower power usage and higher clocks, this was evident with Fiji vs Vega from AMD.
1 Like
This leaves AMD (and IBM, which sold its manufacturing capabilities to GloFo in 2014) without its preferred manufacturing partner. Accordingly, AMD has switched to TSMC for the “breadth” of its 7nm products. This arrangement will cover both CPUs and GPUs, with AMD previously having announced that Vega GPUs (due later this year) and future Epyc processors, codenamed Rome (due 2019), will be built by TSMC.
As was mentioned before, there are lots of semi-custom and fully custom chips being made for 3 letter agencies, as well as specialized things. My company contracts with global foundries for a couple things. I don’t know the specifics, but I do know it’s a line item in the budget.
I have now heard some crazy rumours about AMD’s 7nm TSMC fabbed chips related to node improvements. In particular Vega’s power consumption reductions and some unbelievable clockspeeds for Ryzen. 
2 Likes
Don’t do this to me!
Don’t be a tease.
You bastard!
Do we know the theoretical limits for TSMC’s 7nm silicon?
1 Like
1 Like
Well, 14nm to 7nm is half, so I get the density and can see the power efficiency. I guess performance comes from clock speed, right? I don’t imagine they’re optimizing the architecture that much.
I guess with those numbers, it would be safe to assume we can hit 5GHz on a 3700x or whatever they call it.
#MyPaycheckInTheHypeJar
What’s neat is how close WCCFTech was on this back in June
But yeah
January 8th to 11th
2 Likes
If we assume that the 1.35x scales all the way, that’s potentially a ~5.8GHz boost (considering achievable 4.35GHz boost on 2700X right now).
That’d be insane.
1 Like
Maybe we don’t do finfet anymore eh?