[Business Korea] Nvidia dumping TSMC in favor of Samsung

[tviceman]

Next up, either Samsung licenses Nvidia IP to incorporate Maxwell (or Pascal) into future Samsung SoC's thus dropping the lawsuit OR Samsung and Nvidia settle out of court and Samsung pays Nvidia licenses fees to continue using Mali. Nvidia getting their GPU name into Samsung phones would be a pretty massive win for them. They couldn't hang with Qualcomm and Mediatek on mainstream chips, but their GPU tech is still second to no one in the mobile world.

 

[lopri]

That would be disgusting to say the least. Did anyone buy Kepler IP after NV announced its intention to sell?

 

[Cloudfire777]

TSMC isnt the issue. The market is. TSMC simply delievrs what the market wants.

Unlike TSMC, Samsung doesnt have any HP process.

Except Nvidia had Maxwell designed for 20nm. They already got 20nm samples from TSMC the day they announced Kepler to the public. In early 2012. They wrote about that in their blog

We got high end Maxwell later than originally planned because Nvidia was forced to redesign their chips to 28nm because TSMC couldnt deliver.
AMD is behind the curve because they couldnt rely on a new node like they did before.
So I think it was more because TSMC had most of their tools ready for Apple while neglecting Nvidia and AMD because they are nowhere near the quantity Apple demands. And because of this, price for 20nm SOI HP was very high, making Nvidia and AMD abandon the process and rather wait it out until 16nm FinFETs are here to get more benefits.

I said Samsung and GlobalFoundries work together. One of them producing chips for graphic cards earlier. With Samsung`s expertise and GlobalFoundries commitment to AMD : They produce the chips in PS4/Xbox One with pretty beefy "graphic cards" but is SOCs because its a one in all solution.
Manufacturing HP chips in GlobalFoundries for Nvidia isn`t out of the question at all. We need more competition between foundries to stop TSMC from slow playing us and catering for the most paying customer

 

[ShintaiDK]

Tegra X1 is 20nm, based on Maxwell.

Not sure why you are so busy making TSMC to the big devil. And then champion 2 companies with an even worse reputation and track record.

You also forget the higher transistor cost at 20nm. Not to mention yield. Specially at GPU sizes.

TSMC doesnt discrimitate. Its simply about who pays the most. So if nVidia for any reason didnt get the x wafers they wanted. Then its because they wasnt willing to pay what others would.

 

[Arachnotronic]

Not sure why you are so busy making TSMC to the big devil. And then champion 2 companies with an even worse reputation.

Too much of that going around. They are by far the best semiconductor foundry in the industry today, as its financials and large customer base illustrate.

 

[tviceman]

Except Nvidia had Maxwell designed for 20nm. They already got 20nm samples from TSMC the day they announced Kepler to the public. In early 2012. They wrote about that in their blog

Could you source this please? I'm on Nvidia's blog site, going through 2012 month by month, and found nothing about 20nm production.

EDIT: Found it. http://blogs.nvidia.com/blog/2012/04/25/the-impact-of-process-technology-on-keplers-efficiency/

All it says is that they did a PQV, doesn't say for what kind of product (GPU or SoC). Chances are Nvidia said that the benefits vs. the cost did not warrant further research at that time if it was a PQV for a GPU project.

 

[Cloudfire777]

More like price was too high because of commitment to Apple and TSMC screwing up.

Its already mentioned on quite a few sites that Nvidia had 20nm Maxwell graphic cards engineered and ready but was forced to redesign them to 28nm because of undisclosed reasons. Tegra comes after the graphic cards, thats their secondary objective.
Getting the low and midrange graphic cards up and running is far far more lucrative than Tegra which only hits a very small market due to Samsung and Qualcomm domination

This isnt the first time rumors on the internet about unhappy clients getting tired of TSMC.
Many are jumping ship to Samsung and GloFo. Even the 28nm process for this generation consoles was better from GloFo than TSMC which is why AMD signed a big contract with them

 

[3DVagabond]

Looks like Samsung/GloFo have 2 fabs in the US that are 14nm mass-production sites; that is something I can get behind.

Pretty sure one is cross state in Saratoga.

 

[tviceman]

More like price was too high because of commitment to Apple and TSMC screwing up.

Its already mentioned on quite a few sites that Nvidia had 20nm Maxwell graphic cards engineered and ready but was forced to redesign them to 28nm because of undisclosed reasons. Tegra comes after the graphic cards, thats their secondary objective.
Getting the low and midrange graphic cards up and running is far far more lucrative than Tegra

This isnt the first time rumors on the internet about unhappy clients getting tired of TSMC.
Many are jumping ship to Samsung and GloFo. Even the 28nm process for this generation consoles was better from GloFo than TSMC which is why AMD signed a big deal with them

Imagine another 30% performance out of GM200 and GM204 right now at the same power levels. Dammn.

 

[3DVagabond]

If I had to guess, NVIDIA will probably start off by having Samsung make Tegra chips and then see how well they can deliver. Then they might trial a few low and middle tier GPUs before making the full switch. They've had a long relationship with TSMC so I don't see them dumping them quite yet. Oh and the guy that screwed TSMC really did a number on them but that's typical Samsung, they're a conglomerate based on ripping other's technology.

I don't think nVidia could care less about the "long relationship with TSMC". I could also see them playing one against the other trying to get TSMC to give them a better deal.

 

[Cloudfire777]

Pretty sure one is cross state in Saratoga.

Yup

Imagine another 30% performance out of GM200 and GM204 right now at the same power levels. Dammn.

The 28nm SHP process sure is very interesting. It could be the joker that fuels the upcoming 300 series from AMD. BitsAndChips recently wrote an article where they said they spoke to several AIBs, and the performance for 390X was much higher than whats been posted on the internet lately. You know, the ones that put 390X equal to Titan X.

Its already officially confirmed that AMD is using the 28nm SHP process for their products in 2015. GPUs are mentioned twice (hint hint)

 

[Arachnotronic]

I don't think nVidia could care less about the "long relationship with TSMC". I could also see them playing one against the other trying to get TSMC to give them a better deal.

NVIDIA don't have that kind of clout in this new mobile world.

When GPUs were the big drivers of leading edge silicon at the foundries, AMD and NVIDIA both could have significant influence at TSMC.

Today? Qualcomm and Apple are the fabless big dogs.

 

[ThatBuzzkiller]

People are seriously doubting Samsung's ability to bring out a high performance processor ?

Any chip designer could easily make a high performance chip from a low power version of the process node ...

Here's something to chew on for everyone, TSMC already DETERMINED that there was no noticeable difference in performance between the "low power" and the "high performance" versions so why are people hoping that TSMC would push out a "high performance" version for 16nm ?

 

[Vesku]

NVIDIA don't have that kind of clout in this new mobile world.

When GPUs were the big drivers of leading edge silicon at the foundries, AMD and NVIDIA both could have significant influence at TSMC.

Today? Qualcomm and Apple are the fabless big dogs.

Indeed, Nvidia started publicly complaining about the per transistor costs with 28nm. The companies making chips for phones and tablets are willing to pay more because they have better yields with their relatively tiny chips. AFAIK TSMC was still willing to reserve wafers for Nvidia it's just they weren't also offering a significant discount versus what the ARM SoC companies were willing to pay.

 

[ShintaiDK]

Indeed, Nvidia started publicly complaining about the per transistor costs with 28nm. The companies making chips for phones and tablets are willing to pay more because they have better yields with their relatively tiny chips. AFAIK TSMC was still willing to reserve wafers for Nvidia it's just they weren't also offering a significant discount versus what the ARM SoC companies were willing to pay.

Exactly. A few nVidia slides on the issue. As said, nVidia and AMD for that matter, just isnt willing to pay the same price as others. In the "old days" it was the other way around.

 

[96Firebird]

Pretty sure one is cross state in Saratoga.

Yeah I actually looked into getting a job there when I was about to graduate college (when the fab first opened), but it was odd shifts and clean room attire. Still wouldn't mind a tour of the facility, would be cool to see the process.

 

[Headfoot]

nvm, delete

 

[soccerballtux]

1. Samsung are making 14ff stuff for a few big companies, NV included. What they are making, is less clear.

2. Nobody should take Samsung lightly, this is why TSMC will pay dearly for their arrogance. GloFo is already onboard with Samsung's 14ff, so whenever they are ready, expect it for AMD products.

what arrogance was it?

 

[Headfoot]

^ Did you read about the suit? NV say everyone is infringing. ARM's Mali, Qualcomm's Adreno, and Imagination's PowerVR are all infringing products. So that covers not only Samsung and Qualcomm, but ARM, Imagination, HTC, Motorola, Sony, LG, Hwawei, ZTE, Shaomi, and of course Apple. Basically anyone who does not have cross-licensing agreement with NVIDIA is allegedly infringing on NV's patents.

NV state they are suing Samsung first to set a precedent, or set an example, or something of that nature, and will go after the rest as they see fit.

Legit, no doubt. ^_^

If Nvidia has a patent on a particular graphics technology... and all of those companies use or make that technology without licensing it... that's infringement. Infringement is making, using AND importing -- so technology vendors (=Imagination, ARM, Qualcomm), pure manufacturers (=Sony, LG), and manufacturers who design their own tech (Apple) all are within the reach of that...

Trolling is: you make no products, you have no research and development budget, you employ no engineers, you sue everyone (as in ACTUALLY everyone, like hundreds of defendants literally pulled from the Fortune 500 list with no relation whatsoever). Typically always on business method and business method "on a computer" disguised as technical software type patents. Intellectual Ventures is a troll. Acacia Research Corporation is a troll.

nVidia is absolutely, 100%, undeniably NOT a patent troll.

More on topic --
I have no doubt that Samsung can make a high performance version of a low-power optimized process. If they can do one, they can do the other. The real issue is whether there is sufficient demand (profits) to be made from doing a high performance 16nmFF process given the capital costs. Less of "can they," more of "should they?"

It's very possible that that litigation between nVidia and Samsung is an important chip on the bargaining table. A reasonable royalty (the most likely measure of damages in an integrated device patent infringement suit) would be in the hundreds of millions in the US alone.

 

[Madpacket]

And yet.. none of this involves Intel. Can't wait to see them take a hit as everything computing shifts to mobile. Let's see their process node advantage save them this time.

 

[lopri]

I find it amusing that people try to define the meaning of troll/trolling on the Internet of all place.

 

[exar333]

And yet.. none of this involves Intel. Can't wait to see them take a hit as everything computing shifts to mobile. Let's see their process node advantage save them this time.

Exactly. I can work in my phone and don't even need a laptop or desktop.

 

[Madpacket]

Exactly. I can work in my phone and don't even need a laptop or desktop.

I can on my phone. Is it ideal? Not really yet but give it a few years. Actually I can do a lot more on my Note 4 than my PC in general. The trend is clearly going in the direction of smartphones. Once SOC /ASIC technology reaches parity of "Good enough for the desktop" why is is it so hard to fathom desktops will die a slow death?

You think in a vaccum my friend.

Intel trying to break into the smart phone market reminds me of them trying to build a good GPU. It's taking them way too long or they continually fail to make anything of consideration.