Samsung will not make pascal

S

Stoly

Supreme [H]ardness
Joined
Jul 26, 2005
Messages
6,713
So there's a rumor on fudzilla and some other sites that Samsung lost a contract to build pascal.

http://fudzilla.com/news/graphics/38771-samsung-will-not-make-pascal

AFAIK it has always been rumored that pascal is going to be 16nm and samsung only has 14nm, plus samsung has ZERO experience building large chips for the desktop, only mobile.

Besides Pascal was already taped out at TSMC 16nm FinFET process a couple of month ago

Samsung was supposed to build Tegra 6 aka Erista, which would benefit from its low power 14nm process.
 
Hopefully TSMC can actually deliver... sometime before I turn into a skeleton.
 
Making graphics chips is incredibly difficult. Notice how Global Foundries has yet to make AMD GPUs.
 
Stoly said:
So there's a rumor on fudzilla and some other sites that Samsung lost a contract to build pascal.

http://fudzilla.com/news/graphics/38771-samsung-will-not-make-pascal

AFAIK it has always been rumored that pascal is going to be 16nm and samsung only has 14nm, plus samsung has ZERO experience building large chips for the desktop, only mobile.

Besides Pascal was already taped out at TSMC 16nm FinFET process a couple of month ago

Samsung was supposed to build Tegra 6 aka Erista, which would benefit from its low power 14nm process.
Click to expand...

Its not about Samsung experience but the fact that Samsung was not willing to go as low as nvidia wanted.
 
DeadSkull said:
Its not about Samsung experience but the fact that Samsung was not willing to go as low as nvidia wanted.
Click to expand...

DeadSkull said:
TSMC bid lower dummy. Do you always attack anyone who says anything bad about nvidia?

http://www.digitaltrends.com/computing/tsmc-will-build-nvidias-new-gpu-on-a-16-nm-finfet-process/
Click to expand...

From your link: "However, it seems that Samsung’s relative inexperience in the field led Nvidia management to pursue an exclusive contract with TSMC."
 
DeadSkull said:
Its not about Samsung experience but the fact that Samsung was not willing to go as low as nvidia wanted.
Click to expand...

Actually third party manufacturing of chips is very competitive, it probably doesn't have to do with price but more to do with the two totally different processes. Metal layers, libraries, masks etc will be different so jumping over to Samsung would probably only have been viable if TSMC is having yield issues comparative to Samsung, because it would have incurred more work, cost, and time. Its not a straight here is the design and build, there would be quite a bit of work to do prior to even sample manufacturing.

Also TSMC would probably give priority to nV than Samsung would, because Samsung has its own chips to work on and Apple which has switched over to Samsung for manufacturing their A9 chip.

Actually space saved and power savings compartitive with 14nm FF+ and 16nm FF+ are similiar
 
jwcalla said:
From your link: "However, it seems that Samsung’s relative inexperience in the field led Nvidia management to pursue an exclusive contract with TSMC."
Click to expand...

That's what tech blogs like to parrot but the fact is Samsung reached 14nm first and already demonstrated its experience and reliability by making Apple A9 SoC.

"Looks like TSMC’s still not out of the running for Apple’s A9 orders this time around even thought the company has been behind both GlobalFoundries and Samsung when it come to the 14nm FinFet.
 
razor1 said:
Actually third party manufacturing of chips is very competitive, it probably doesn't have to do with price but more to do with the two totally different processes. Metal layers, libraries, masks etc will be different so jumping over to Samsung would probably only have been viable if TSMC is having yield issues comparative to Samsung, because it would have incurred more work, cost, and time. Its not a straight here is the design and build, there would be quite a bit of work to do prior to even sample manufacturing.

Also TSMC would probably give priority to nV than Samsung would, because Samsung has its own chips to work on and Apple which has switched over to Samsung for manufacturing their A9 chip.

Actually space saved and power savings compartitive with 14nm FF+ and 16nm FF+ are similiar
Click to expand...

14 and 16nm from Samsung and TSMC are both FinFet ...... Speaking of which, TSMC was doing terrible on its adoption/move to 16nm manufacturing process. So bad that Qualcomm decided to jump ship from TSMC over its absurd year long delay with FinFet 16nm move. This is why the statements that "TSMC has more experience at 16nm" don't make any sense.

http://wccftech.com/tsmcs-16nm-finfet-faces-delays-qualcomm-jumps-ship-samsung/
 
Last edited:
DeadSkull said:
That's what tech blogs like to parrot but the fact is Samsung reached 14nm first and already demonstrated its experience and reliability by making Apple A9 SoC.

"Looks like TSMC’s still not out of the running for Apple’s A9 orders this time around even thought the company has been behind both GlobalFoundries and Samsung when it come to the 14nm FinFet.
Click to expand...



http://appleinsider.com/articles/15...tsmc-for-30-of-a9-chip-orders-for-next-iphone

they are using Samsung for most of their production of A9 chips
 
DeadSkull said:
14 and 16nm from Samsung and TSMC are both FinFet ...............
Click to expand...



Doesn't matter if its finfet, finfet is just the transistors, not the metal layers, not the masks, not the libraries used to create the transistor layouts, and what not.
 
jwcalla said:
From your link: "However, it seems that Samsung’s relative inexperience in the field led Nvidia management to pursue an exclusive contract with TSMC."
Click to expand...

Businesses don't usually say "Oh we decided to go with ****** because they're cheaper."
 
NickJames said:
Businesses don't usually say "Oh we decided to go with ****** because they're cheaper."
Click to expand...

NVIDIA doesn't always make the best decisions so if Pascal get's delayed due to TSMC imbecility this would be hilarious.
 
DeadSkull said:
NVIDIA doesn't always make the best decisions so if Pascal get's delayed due to TSMC imbecility this would be hilarious.
Click to expand...


They have had pascal samples from TSMC for over 3 months now (rumored), so would it have been a bad decisions to stay with production at TSMC? I think it would have been a worse decision to switch to Samsung at this point. Also I think at this point nV would have a very good understanding of TSMC's 16 FF + process regarding Pascal. If yields are going as suggested by TSMC, they shouldn't have any problems.

Second note, Samsung will also be making HBM2, so the "exclusivity" for AMD goes out the window, exclusivity is only from Hynix for the one year of creating HBM1.
 
NickJames said:
Businesses don't usually say "Oh we decided to go with ****** because they're cheaper."
Click to expand...

Ummm... riggghhhttt.

Every business I have worked for is always trying to come up with "cost-savings".

They nickle and dime stuff that needs to be taken to the people with the most experience/best product and they overpay for crap that doesn't matter as much.

Oh, and you think they are going to hire the more expensive person for a job if both have the same experience and both know their field really well?

Nope, they are going to hire the cheapest one unless there is somebody already working there that knows them and has recommended them.
 
cyclone3d said:
Ummm... riggghhhttt.

Every business I have worked for is always trying to come up with "cost-savings".

They nickle and dime stuff that needs to be taken to the people with the most experience/best product and they overpay for crap that doesn't matter as much.

Oh, and you think they are going to hire the more expensive person for a job if both have the same experience and both know their field really well?

Nope, they are going to hire the cheapest one unless there is somebody already working there that knows them and has recommended them.
Click to expand...

Highlighted in red, that is something Samsung doesn't have yet, GPU's are much larger pieces of silicon than SOC's. And if a competitive market, Samsung and TSMC will not be that much of a difference with costs. TSMC might be more flexible with their cost structure because of their dealings with nV in the past though.
 
razor1 said:
Highlighted in red, that is something Samsung doesn't have yet, GPU's are much larger pieces of silicon than SOC's. And if a competitive market, Samsung and TSMC will not be that much of a difference with costs. TSMC might be more flexible with their cost structure because of their dealings with nV in the past though.
Click to expand...

Right.. which is why I wrote the response I did.
 
DeadSkull said:
That's what tech blogs like to parrot
Click to expand...

But... that's the article that you linked to to support the argument you made, and all it did was completely contravene the point your were making. And then you called the other guy a dummy.

but the fact is Samsung reached 14nm first and already demonstrated its experience and reliability by making Apple A9 SoC.
Click to expand...

Who cares? The SoC processes are completely different than the HP processes used to make GPUs. This is why we never got 20nm GPUs despite the fact that TSMC was cranking out ARM SoCs at that node.

I don't think it's any offense against Samsung, either. It's just that schedules are tight and it could be risky going with a company that has no experience manufacturing GPUs.

I'm sure cost was one of the deciding factors, but most likely not the only one.
 
more like besides intel, no one is delivering 14/16nm chips at 300mm2 die sizes at reasonable yields.
 
^ Bingo. The level of process control to make these chips simply boggles the mind. It really is Clarke's 3rd law.
 
and that's why people that laugh at async shouldnt be taken seriously.

if there is any modicum of delay, nvidia is in a huge pinch, while AMD can still rebrand and pump out even cheaper 28nm cards in 2016, when basically AMD is guaranteed to have 2 of the top 5 biggest games for 2016-- battelfront and battlefield5.

now understand that pascal/greenland is rumored to be 600mm2.
 
^ Battlefront is 2016? And what does that game have to do with AMD? It's DX11, honey.
 
DPI said:
^ Battlefront is 2016? And what does that game have to do with AMD? It's DX11, honey.
Click to expand...

I think Battlefront is going to be a Gaming Evolved title judging by how much AMD is involved with marketing and game enhancements.

Afterall AMD does power XBONE and PS4 where Battlefront will be one of the biggest sellers and will probably outpace PC sales.
 
cyclone3d said:
Ummm... riggghhhttt.

Every business I have worked for is always trying to come up with "cost-savings".

They nickle and dime stuff that needs to be taken to the people with the most experience/best product and they overpay for crap that doesn't matter as much.

Oh, and you think they are going to hire the more expensive person for a job if both have the same experience and both know their field really well?

Nope, they are going to hire the cheapest one unless there is somebody already working there that knows them and has recommended them.
Click to expand...

I think you missed the point. Of course companies will go with the cheapest bidder, they just don't say it.
 
I'm very certain that far more due diligence factors are considered in choosing a fab than just how "cheaply" it can be done. Timing, track record, yields and volume capability. Some of those are far more important than simply being cheaper than the next fab.
 
DPI said:
I'm very certain that far more due diligence factors are considered in choosing a fab than just how "cheaply" it can be done. Timing, track record, yields and volume capability. Some of those are far more important than simply being cheaper than the next fab.
Click to expand...

You would be surprised, that's exactly all the former ceo of Home Depot that came in to run Chrysler cared about. He ordered 15% cuts to all vendors or he would ship the production to China. He didn't care the quality was garbage, it was cheaper and that looked great on paper which is all most ceo's actually know. This is why Chrysler almost went the way of the dodo.
 
Gideon said:
You would be surprised, that's exactly all the former ceo of Home Depot that came in to run Chrysler cared about. He ordered 15% cuts to all vendors or he would ship the production to China. He didn't care the quality was garbage, it was cheaper and that looked great on paper which is all most ceo's actually know. This is why Chrysler almost went the way of the dodo.
Click to expand...

Yeah, companies that intend to commit suicide will look at costs only.

Nvidia doesn't strike me as one of those companies, and Jen-Hsun doesn't strike me as a fly-by-night only-looking-for-a-golden-parachute CEO.
 
DPI said:
I'm very certain that far more due diligence factors are considered in choosing a fab than just how "cheaply" it can be done. Timing, track record, yields and volume capability. Some of those are far more important than simply being cheaper than the next fab.
Click to expand...
Citation needed. Or did you just pull that out of your .. ?
 
The lowest bid for something like this is not that simple, it is actually rather complicated. It is not going to be a straight forward bid process in terms of Samsung offering to supply GP100 (hypothetical) chips at $1.00 each and TSMC offering to supply them at $0.90 each. The "quality" (as a broad term) of each respective fab is very much related to the actual cost. Costs would also not be simply limited to the actual supply of chips from a fab.
 
DPI said:
I'm very certain that far more due diligence factors are considered in choosing a fab than just how "cheaply" it can be done. Timing, track record, yields and volume capability. Some of those are far more important than simply being cheaper than the next fab.
Click to expand...

Thing is though, when it comes to timing, track record, and yields (at least initially), TSMC fails on all three accounts.

Not saying Samsung is any better (because I know nothing about their fabs), but TSMC's track record can only be described as "very bad" in the past 5 years.
 
n=1 said:
Thing is though, when it comes to timing, track record, and yields (at least initially), TSMC fails on all three accounts.

Not saying Samsung is any better (because I know nothing about their fabs), but TSMC's track record can only be described as "very bad" in the past 5 years.
Click to expand...

Why would you single out TSMC with the latter statement? Even a company the size of Samsung hasn't perfected 14nm/16nm yet and that was with the benefit of stealing TSMC technology. Then there's also GloFlo who is perpetually behind the curve. Apart from Intel, TSMC is one of the fabs that actually has a really good track record.
 
Gideon said:
You would be surprised, that's exactly all the former ceo of Home Depot that came in to run Chrysler cared about. He ordered 15% cuts to all vendors or he would ship the production to China. He didn't care the quality was garbage, it was cheaper and that looked great on paper which is all most ceo's actually know. This is why Chrysler almost went the way of the dodo.
Click to expand...

Well the difference is that this CEO first of all does not care about the company he is employed with at all looking from a long-term perspective. This is not the case with Nvidia and JHH. It's his baby and as long as he is on board and in charge i highly doubt that price is his major concern. Nvidia can perfectly settle this off with corresponding GPU pricing.

I think the biggest concern for them in that case is volume, track record and past experience. So far TSMC out of all of them has the biggest experience with large GPUs.
 
5150Joker said:
Why would you single out TSMC with the latter statement? Even a company the size of Samsung hasn't perfected 14nm/16nm yet and that was with the benefit of stealing TSMC technology. Then there's also GloFlo who is perpetually behind the curve. Apart from Intel, TSMC is one of the fabs that actually has a really good track record.
Click to expand...

Uhh because apart from Intel, no other foundry on this planet can even begin to compare to TSMC? TSMC literally had half the market share and revenue for the entire foundry market in 2014. The other 9 of the top 10 foundries combined don't reach the market share or revenue of TSMC. Likewise for the period from 2011-2013. So I don't see how comparing TSMC against GloFo or any other foundry that has 1/4 or less its revenue/market share is proper. Intel is the only benchmark that TSMC should be compared against IMO, not to mention since TSMC fancy themselves as being superior to Intel in certain segments, this comparison is all the more fair.

For a company of TSMC's size and status, naturally one expects more from them. So when they run into issues with every node starting with 40nm (they canned 32nm HKMG), coupled with them denying they were having issues with 40nm and 28nm despite there really being issues (at least initially), that doesn't speak well for their track record.
 
Last edited:
cyclone3d said:
Ummm... riggghhhttt.

Every business I have worked for is always trying to come up with "cost-savings".

They nickle and dime stuff that needs to be taken to the people with the most experience/best product and they overpay for crap that doesn't matter as much.
Click to expand...

I've seen this too. And on the overpay side, what is it with companies who insist on buying desktop memory from "approved vendors"? A few years ago I convinced management to get more ram for our desktops. They paid more for 2GB (or was it 4?) than i paid for 8GB of OC ram. I said, "you know we could just order this from Newegg or Amazon for 1/3 - 1/2 the price. If it doesn't work, exchange it.

But that doesn't stop them from pushing out a product months early to save money at the expense of employee support hours (which are often free to them, if you don't count all the people that bail after every single project because of mismanagement)
 
You must log in or register to reply here.
Back
Top