TSMC Secures Orders from NVIDIA for 7nm and 5nm Chips

[btarunr]

TSMC has reportedly secured orders from NVIDIA for chips based on its 7 nm and 5 nm silicon fabrication nodes, sources tell DigiTimes. If true, it could confirm rumors of NVIDIA splitting its next-generation GPU manufacturing between TSMC and Samsung. The Korean semiconductor giant is commencing 5 nm EUV mass production within Q2-2020, and NVIDIA is expected to be one of its customers. NVIDIA is expected to shed light on its next-gen graphics architecture at the GTC 2020 online event held later this month. With its "Turing" architecture approaching six quarters of market presence, it's likely that the decks are being cleared for a new architecture not just in HPC/AI compute product segment, but also GeForce and Quadro consumer graphics cards. Splitting manufacturing between TSMC and Samsung would help NVIDIA disperse any yield issue arriving from either foundry's EUV node, and give it greater bargaining power with both.



View at TechPowerUp Main Site

 

[medi01]

TSMC so desperately needed Huang's orders.
Right?
"Secured", chuckle.

 

[JAB Creations]

With AMD making a real comeback with their CPUs and now having the resources to properly invest in their GPUs I find the rumors that Nvidia is nervous about RDNA2 to be reasonably credible. I'd imagine if they're making gaming GPUs with some on 7nm and others on 8nm that they'll focus on splitting the silicon between ideal/close to ideal with no or few defects to those with moderate amounts of defects. It'd be an interesting approach to see something like a 3080ti on 7nm and a 3060ti on 8nm. If so would they split the higher defect 7nm chips down to a 3070ti or is 7nm not yet matured enough for whatever size Nvidia is going for (since they tend to be larger)? Either way AMD really needs to win on both CPU and GPU fronts this year to sustain their growth if we're to see reasonable prices return (and more importantly, stay) to the market.

 

[ppn]

8nm 3050Ti, all the rest 7nm, 5nm in 4 years. I see no reason to jump ahead, historically 2 generations and their refreshes - one node.
besides EUV is limited to 420mm2, so no more big chips there.

 

[medi01]

besides EUV is limited to 420mm2, so no more big chips there.

AMD has 500 mm2 RDNA2 chip in the pipeline (source is "some Asian twitted it").

 

[Ashtr1x]

it could confirm rumors of NVIDIA splitting its next-generation GPU manufacturing between TSMC and Samsung

This is a very interesting rumor, in 2018-19 Samsung was pitched for Ampere but now we see TSMC which was always the Nvidia partner coming back again for the dual sourcing. Just like Pascal days for TSMC 16nm and Samsung 14nm, and also GloFo getting the 14nm from Samsung for the WSA for AMD.

Is it because of capacity, volume being higher that TSMC has been contracted or any other aspect, also we do remember how Apple did this with TSMC and Samsung back during 14nm for A9 and somehow Samsung node was superior for efficiency but now tables have turned completely, as TSMC is now the leader with N7 and Samsung is second, their 10nm, I only saw Qcomm doing the Snapdragons 835 and 845 on that process while their latest 8nm wasn't that much great with Exynos and the new Exynos 990 is done on 7nm EUV, but the uArch was garbage so couldn't see how good it is to even draw assumptions for the Nvidia Ampere.

Very interesting times ahead, when TSMC is now gearing for 5nm already in pipeline for Apple processors and AMD Ryzen 5000, and now Nvidia too (presumably saw that AMD's new RDNA2 is really worth to pump more into the R&D which they are doing very quickly vs what we have on Pascal and Turing - TSMC 12nm still) and on a completely other side note while Intel is still in the bs rumor land for Alder lake crappy biglittle bs on desktop on 10nm for 2022 ? where is 7nm Intel that they were mentioning on Track ? Samsung is already into 3nm GAAFET...

 

[_Flare]

Maybe some very diversely equipped chips with everything of RT and FP64 could need a very dense node, until now there is no Nvidia chip utilizing halfspeed FP64, Tensorcores and RT-HW-Acceleration.
Dunno how important FP64 still is beside all these mixed-precission possibilities.
The split could be strict between Tesla Volta successor Samsung (denser slower clocked, extreme corecount) and the Quaddro and Geforce successor TSMC (higher clocked and minimal FP64)

 

[AnarchoPrimitiv]

With AMD making a real comeback with their CPUs and now having the resources to properly invest in their GPUs I find the rumors that Nvidia is nervous about RDNA2 to be reasonably credible. I'd imagine if they're making gaming GPUs with some on 7nm and others on 8nm that they'll focus on splitting the silicon between ideal/close to ideal with no or few defects to those with moderate amounts of defects. It'd be an interesting approach to see something like a 3080ti on 7nm and a 3060ti on 8nm. If so would they split the higher defect 7nm chips down to a 3070ti or is 7nm not yet matured enough for whatever size Nvidia is going for (since they tend to be larger)? Either way AMD really needs to win on both CPU and GPU fronts this year to sustain their growth if we're to see reasonable prices return (and more importantly, stay) to the market.

I think RDNA2 will be the Zen2 moment for the GPU division, RDNA1 made impressive strides and was only intended to be a stepping stone. Doesn't RDNA2 have some co-developement from Microsoft/Sony? Anyway, if the PS5 clock speeds are anything to go by as well as what we've discovered from the Xbox Series X reveal, I think RDNA2 will turn some heads and have more success than previous generations. This will be the first big GPU release since Zen2 which has changed a lot of perceptions on AMD. Also, if Lisa Su was able to turn around the CPU division, then there's absolutely no reason to believe she can't do the same with the GPU division.

 

[renz496]

With AMD making a real comeback with their CPUs and now having the resources to properly invest in their GPUs I find the rumors that Nvidia is nervous about RDNA2 to be reasonably credible.

nvidia will coming with faster product regardless AMD is doing well or not. AFAIK they never wait for AMD to "cook something" before they get into gears. nvidia always have certain expectation for AMD but will AMD able to meet that expectation or not is different story.

This is a very interesting rumor, in 2018-19 Samsung was pitched for Ampere but now we see TSMC which was always the Nvidia partner coming back again for the dual sourcing. Just like Pascal days for TSMC 16nm and Samsung 14nm, and also GloFo getting the 14nm from Samsung for the WSA for AMD

to be honest i don't know why some people were obsess about nvidia to put majority of their GPU manufacturing on samsung. if we look at nvidia history and the way they did things it is almost impossible for them to do that. for one they are not the one going to take risk on unproven nodes. this is true even for several tech that are being used in GPU world. we know nvidia really fond of their big die strategy. and for the past two decade the only one with enough experience and expertise to do that is TSMC. suddenly nvidia going all out with samsung that never proven capable of making gigantic die? i'm sure samsung probably get some nvidia business but i dare to bet it won't be nvidia primary product like GPU. maybe for tegra SoC or some low end GPU but for the vast majority of nvidia GPU most likely being manufactured at TSMC.

 

[Kaotik]

Seriously @btarunr ? This was just discussed in the last "nvidia secret 5nm chip" news piece, it was known already that NVIDIA splits 7nm between TSMC and Samsung, it wasn't any rumor or such, it was direct comment from Jensen Huang himself, last December!

 

[Vayra86]

I think RDNA2 will be the Zen2 moment for the GPU division, RDNA1 made impressive strides and was only intended to be a stepping stone. Doesn't RDNA2 have some co-developement from Microsoft/Sony? Anyway, if the PS5 clock speeds are anything to go by as well as what we've discovered from the Xbox Series X reveal, I think RDNA2 will turn some heads and have more success than previous generations. This will be the first big GPU release since Zen2 which has changed a lot of perceptions on AMD. Also, if Lisa Su was able to turn around the CPU division, then there's absolutely no reason to believe she can't do the same with the GPU division.

Its still going to hinge on the software side I think. Nvidia has a lead in more than just performance. Even a 5-10% gap for the same dollar in favor of AMD won't persuade people to switch over. Its similar to Ryzen adoption rate in that regard and let's face it, its not exactly like Navi has been a Zen 1 yet, its cheaper but certainly not better in any way and even lacks features. Let's hope RDNA2 is that, and then AMD can keep it up for a few more generations. For much the same reasons as in CPU. Not everyone upgrades in the same gen. If you bought Turing upper midrange/high end now, how likely is it really going to be?

Also note, the momentum for Navi to truly shine has already passed because they had the node advantage already, but won't have it now. And that's a pretty big one, especially when the chips start getting larger.

 

[chodaboy19]

At least Wccftech runs all the way with their rumors, not some half assed attempt like this...

Report: NVIDIA Might Go With TSMC 7nm EUV For High End 'Ampere' GPUs, And Samsung 7nm/8nm EUV For Low End GPUs

DigiTimes just published a massive report (hidden behind a paywall) that sheds some light on NVIDIA's plan for Ampere GPUs as well as their future plans. It had been rumored that NVIDIA was thinking of going with Samsung's 7nm process but those rumors appeared to have been debunked. As it turns...

wccftech.com

NVIDIA's Hopper Architecture Will Be Made On TSMC's 5nm Process, Launching 2021

Taiwan's leading technology publication dropped quite the scoop on NVIDIA news today: they have confirmed that the company has pre-booked 5nm capacity from TSMC to use for their next-generation Hopper GPUs. According to a report we already covered, AMD's aggressive approach with 7nm caught...

wccftech.com

 

[medi01]

Even a 5-10% gap for the same dollar in favor of AMD won't persuade people to switch over

Yeah, AMD market share going from 18% to 30% in a single quarter once Navi hit was totally nothing.
Oh, and that despite Huagn's price drops and "supers".

Something something, momentum, truly shine, chuckle.

 

[Fluffmeister]

Yeah, AMD market share going from 18% to 30% in a single quarter once Navi hit was totally nothing.
Oh, and that despite Huagn's price drops and "supers".

Something something, momentum, truly shine, chuckle.

We get it, you hate the ex-AMD employee who set up a hugely successful company. I can see why he picked Nvidia as a name too.

We all hope you are enjoying your Navi card.

 

[Vayra86]

Yeah, AMD market share going from 18% to 30% in a single quarter once Navi hit was totally nothing.
Oh, and that despite Huagn's price drops and "supers".

Something something, momentum, truly shine, chuckle.

It will be nice if AMD can get those who switched as returning customers. So they'll need a longer breath even than RDNA2. That was the point of my post.

 

[renz496]

Yeah, AMD market share going from 18% to 30% in a single quarter once Navi hit was totally nothing.
Oh, and that despite Huagn's price drops and "supers".

Something something, momentum, truly shine, chuckle.

It did not happen in a single quarter. In fact for what i can remember AMD launch Navi in Q2 last year and they lost some market share to nvidia in the same quarter.

 

[watzupken]

"Splitting manufacturing between TSMC and Samsung would help NVIDIA disperse any yield issue arriving from either foundry's EUV node, and give it greater bargaining power with both. "

This is somewhat true, but going exclusive with a single supply should also give AMD an advantage in terms of pricing. Dual sourcing from my perspective is more to diversify risk of issue with any of the supplier. So in this case the flipside is that if TSMC run into issues with their fab, then it will affect all of AMD's product lines. If Nvidia is tapping TSMC for their top end/ flagship product, it will nonetheless affect them.

 

[klf]

amd ?? like X399? my 2990WX was dead platform after one year!! 2500usd go to hell.. willl never again buy amd.
zen 2 have no problem by pcie-4 but amd decide block and make "tr40" chipset .. HATE AMD!
my 2080Ti is solid invest after years , like
)

 

[medi01]

It did not happen in a single quarter. In fact for what i can remember AMD launch Navi in Q2 last year and they lost some market share to nvidia in the same quarter.

Citation needed, and, given how extraordinary the statement is, an extraordinary one.

t will be nice if AMD can get those who switched as returning customers.

In other words, your statement is unfalsifiable: regardless if AMD's market share stays the same, is cut in half, or nearly doubles, it's still right.

A nice "true scottsman".

ex-AMD employee who set up a hugely successful company

Someone: AMD market share went from 18% to 30%
Someone else: It's because you hate Huang

Mind boggling.

 

[renz496]

Citation needed, and, given how extraordinary the statement is, an extraordinary one.

AMD market share in Q2 2019 is 32.1% (the quarter AMD launch RX5700)

Jon Peddie Research releases its Q2, 2019 global add-in board report

AMD increases market share again in the quarter

www.jonpeddie.com

AMD market share in Q3 2019 is 27.08% (the quarter we should start seeing how RX5700 affecting AMD market share)

Global Q3’19 add-in board market soars led by Nvidia, reports Jon Peddie Research

Nvidia significantly increases market share in the quarter

www.jonpeddie.com

AMD market share in Q4 2019 is 31.08%

Add-in Board report – a report on the Graphic Add-in Board market

Add-in Board report – a report on the Graphic Add-in Board market The Add-in Board report is a quarterly report that focuses on the market activity of PC graphics controllers for mobile and desktop computing. The report provides an in-depth look at the PC graphics market and includes unit...

www.jonpeddie.com

 

[Vayra86]

Mind boggling.

Yes, the endless twisting of facts and figures to suit your sad narrative is indeed mind boggling. See source above. There is no Navi effect, but I knew this already.

 

[medi01]

There is no Navi effect

That does't matter, twisty one, you have already justified AMD going from 18 to 30% as "returning customers".

 

[Vayra86]

That does't matter, twisty one, you have already justified AMD going from 18 to 30% as "returning customers".

View attachment 154135

No that was your figure, not mine

 

[medi01]

No that was your figure, not mine

That exactly the embarrassing part: even THAT figure was not enough for you to change the narrative, you are that twisty.

 

[Vayra86]

That exactly the embarrassing part: even THAT figure was not enough for you to change the narrative, you are that twisty.

Why? For all I know you pulled 18% out of your ass, because as usual, your sources are either missing or skewed to favor your narrative.

Trust then quickly becomes zero. Its too bad, damage is done, I've moved on. Sometimes it seems you're doing better, but then you manage a bunch of posts that completely destroy all credit once more. Something you might want to reflect on. Or not, whichever floats your boat...

I'm always open to a fresh discussion on neutral grounds with sound facts, as you might have noticed before, and I will also admit when I'm wrong. Can you, or is the narrative more important than people on this forum? Its a choice you will have to make.