TSMC Gives NVIDIA Priority for 28 nm Manufacturing

[btarunr]

Relations between NVIDIA and its principal foundry partner, TSMC, have been unpredictable in recent times, with reports of NVIDIA expressing displeasure with it over 28 nm manufacturing capacity, which is denting its competitiveness; and later crediting collaboration with it, for the energy-efficiency of its latest Kepler family of GPUs. With NVIDIA threatening to find other foundry partners for bulk manufacturing, and reports of Samsung already preparing qualification samples for it, TSMC is responding by issuing NVIDIA a priority over other clients (such as Qualcomm, AMD) for manufacturing of 28 nm chips.

While being unsatisfied with TSMC's output, and its new policy of charging for wafers rather than working chips yielded, NVIDIA refuted rumors of it seeking other foundry partners such as Samsung and Global Foundries. When put on high-priority, TSMC will facilitate speedy launch of new NVIDIA GeForce SKUs towards the end of Q2, 2012. Supply prioritization isn't new, TSMC has, in the past, prioritized Qualcomm when it threatened to shift allocations to other foundries. It remains to be seen how AMD responds to the situation, as such a prioritization would come at the expense of its volumes, and could threaten its competitiveness.



View at TechPowerUp Main Site

 

[Crap Daddy]

TSMC and NV, the story of a beautiful love/hate relationship.

 

[TheMailMan78]

Might see some of those unicorn 680's in the story finally.

 

[entropy13]

Might see some of those unicorn 680's in the story finally.

Someone I know just bought a Palit GTX 680 Jetstream today. He said there was someone that was getting a reference design GTX 680 when he got to the store. Before he left the store, another one arrived that also bought a GTX 680 Jetstream.

 

[TheMailMan78]

Someone I know just bought a Palit GTX 680 Jetstream today. He said there was someone that was getting a reference design GTX 680 when he got to the store. Before he left the store, another one arrived that also bought a GTX 680 Jetstream.

Philippines. Tiny market with a 16th of the users of the US/Canada. I am willing to bet their are more 680's in the US/Canada then the Philippines but since our consumer market is like a million times bigger they sell out instantly.

 

[entropy13]

Philippines. Tiny market with a 16th of the users of the US/Canada. I am willing to bet their are more 680's in the US/Canada then the Philippines but since our consumer market is like a million times bigger they sell out instantly.

Nope, the Taiwanese companies usually group us all together as "Southeast Asia", so it's just one "market" for the whole ASEAN. Asus in the Philippines, for example, is merely a "primary" local partner designated by Asus to use it's name, but it doesn't mean that Asus has a presence in the Philippines itself (or in other ASEAN countries for that matter), considering how near we are to Taiwan. Hmm, actually when I think of it the "Asus" store in one mall is owned by a different company compared to another "Asus" store in a different mall. LOL Same with MSI. "MSI" stores with different owners, much like franchising. lol

So you really can't compare the Philippines as a single "market" to US/Canada...but then it's quite weird, you go on and compare one country instead of the whole region to a whole region ("North America") and not one country.


And accepting the size disparity, shouldn't we be the ones who should be out of stock and not the US/Canada since, as you yourself said, you're the bigger market?

 

[TheMailMan78]

Nope, the Taiwanese companies usually group us all together as "Southeast Asia", so it's just one "market" for the whole ASEAN. Asus in the Philippines, for example, is merely a "primary" local partner designated by Asus to use it's name, but it doesn't mean that Asus has a presence in the Philippines itself (or in other ASEAN countries for that matter), considering how near we are to Taiwan. Hmm, actually when I think of it the "Asus" store in one mall is owned by a different company compared to another "Asus" store in a different mall. LOL Same with MSI. "MSI" stores with different owners, much like franchising. lol

So you really can't compare the Philippines as a single "market" to US/Canada...but then it's quite weird, you go on and compare one country instead of the whole region to a whole region ("North America") and not one country.


And accepting the size disparity, shouldn't we be the ones who should be out of stock and not the US/Canada since, as you yourself said, you're the bigger market?

That's not the way consumerism works. Your consumer base isn't as large as the US. That's a fact. Unless you are saying the "Asian market" is now a bigger market then all of the North, Central and South Americas (since we are lumping in entire regions) when China/South Korea is the only real players in your market.

And no. When the demand out weighs the supply prices hikes and supply is depleted. If you have a good on the market shelves in abundance that means there is LESS demand or to much supply. So either NVIDIA is a dumb ass in distribution OR there isn't enough demand in your region which makes more sense as your consumer base is much lower.

 

[SIGSEGV]

so, they've already gave nvidia priority now, this is interesting, let's see what happen next

 

[Mathragh]

Wouldnt this be unfair, and basically go against contracts TSMC already has with other companies like Qualcomm and AMD?

 

[TheMailMan78]

Wouldnt this be unfair, and basically go against contracts TSMC already has with other companies like Qualcomm and AMD?

Depends on the contracts honestly. The "exclusiveness" of the faculties time frame is probably up for AMD. They have their market saturation anyway. No need to stay balls deep in production for AMD. Also if the stay exclusive to one vendor for to long then they risk a "anti-competitive" suit from NVIDIA.

 

[cadaveca]

Wouldnt this be unfair, and basically go against contracts TSMC already has with other companies like Qualcomm and AMD?

It could just mean that TSMC has fulfilled an order with another company(like AMD or Quallcomm), and can now give nVidia more fab time.


Funny how this comes out today.

 

[TheMailMan78]

Funny how this comes out today.

I was thinking the same thing.

 

[alwayssts]

They gotta get GK106 out somehow...seriously, even if through the typical strong-arming tactics. That's very much what this article is implying and I agree.

The pricing structure is going to more-or-less implode when they release the 6 SM GK104 to upset 7870, which they pretty much have to do. That will in-turn cause a release of 7950 1.5gb and/or a 1536sp 1.5GB Tahiti to compensate at a similar price and Pitcairn will become cheaper. If they don't have a product to compete with the cheap-as-chips to make Pitcairn parts when prices are adjusted they will be in a very difficult position, as that is without a doubt the largest market.

When all dies that you see come from TSMC that's logiiiiiiistics.

 

[eddman]

Might see some of those unicorn 680's in the story finally.

680s are in short supply but they do become available once or twice a day, at newegg at least, although they sell so fast that most people can't manage to get any.

so, they've already gave nvidia priority now, this is interesting, let's see what happen next

Charlie tends to pull things out of his arse usually, and if you've noticed his so called articles have tags like "humor", "opinion" and "rumor".

 

[TheoneandonlyMrK]

Probably BS and probably dosnt matter on the whole anyway now nvidias paying by the wafer, something that to me clears up WHY the GK104 became their flagship this round, possible to make and useable yields(given fans will pay the earth), Its starting to look to me like nvidias got some big issues but if any co are good at polishing a turd its nvidia, year after year they amaze me with what they manage to get their gpu's todo despite hicups and tdp and they amaze me moreso that they allways charge the earth comparitively for a products worth

 

[Casecutter]

OK, but so now Mailman is saying that becuase of nVidia's issues, AMD has been given priority @ TSMC.

That was yesterday, today...

It could just mean that TSMC has fulfilled an order with another company(like AMD or Quallcomm), and can now give nVidia more fab time.


Funny how this comes out today.

Interesting to say the least, sounds like Nvidia was still juggling chip design way late into last year or even this year. Now when a manufacturing company sets a delivery base on the date of placing the order, they can juggle some up "in-the-line" as long as earlier orders could remain in a realm of staying consistent. Although companies like I work for will normally require a fee for such expediting. Wonder if TSMC is charging for expediting such orders, it would indicate how friendly the two have remained.

Contractually and legally there's a window to deliver and a must be on the dock date. AMD having missed deliveries is more costly to TSMC, and legally binding than Nvidia stating, a vendor being "anti-competitive". Didn't ask for your wafer start's early enough that nobody’s fault but their own.

 

[Jarman]

Not that I want one, but they are readily available in the UK.

 

[TheoneandonlyMrK]

Not that I want one, but they are readily available in the UK.

yeah theres four at aria MSI's flavour the other 10 brands are preorder only, they are a reasonable(now) 449 uk quids though

oc uk seem to have a few though, not entirely surpriseing

Its a good job nvidia might have sorted supply out as 5xx series card prices are on the up wtf , miss a date did they (GK106-107) ,looks like retailers are takeing the urine again

 

[nikko]

Previous nvidia gen was introduced at 15% tsmc production lines allocated to 40nm. Now they pulled a 1% paper launch of $499 product like never before. 8800 came out at 10% and 199$, GF104 at 16% and $229. Scarcity drives the price trough the roof. This card should cost no more than $129-139 for the 1152-1344, the full crystal being a premium product with not exactly the best bang for the buck.

 

[cadaveca]

That was yesterday, today...


Interesting to say the least, sounds like Nvidia was still juggling chip design way late into last year or even this year. Now when a manufacturing company sets a delivery base on the date of placing the order, they can juggle some up "in-the-line" as long as earlier orders could remain in a realm of staying consistent. Although companies like I work for will normally require a fee for such expediting. Wonder if TSMC is charging for expediting such orders, it would indicate how friendly the two have remained.

Contractually and legally there's a window to deliver and a must be on the dock date. AMD having missed deliveries is more costly to TSMC, and legally binding than Nvidia stating, a vendor being "anti-competitive". Didn't ask for your wafer start's early enough that nobody’s fault but their own.

Very interesting, isn't it? The fact the word "priority" was even used, in this context, has me rolling around clutching mah belly against the laughter.

It was MY thought that AMD was given priority, as they were the only partner that seemed to be happy with the number of wafers they secured from TSMC, although they were clearly indicating that if they could have had more wafers, they could have used it. The other partners simply said they could not get enough, with Quallcomm openly stating they were looking for other 28 nm suppliers.

What even more interesting ot me is how it's today, and not a couple of weeks ago, that this news came out, after the GTX690 launch. I think it might have worked out better for both parties if they coudl ahve pulled this news out just before the launch, although I guess that might strain things between AMD and TSMC, maybe.

AMD has several different products using the HP 28NM TSMC process, and all of them are on shelves today, with what seems to be good supply, so it does makes sense that TSMC has a bit more fab time now, and to me, it makes sense that it's not AMD that really plays a roll here in this news; rather, it might be those other partners that TSMC has given nVidia priority over, and I almost want to say that Qualcomm made a deal with Samsung for 28nm that made this possible.


Before today, TSMC was stating that they would not be able to meet 28nm demand until q4 2012, into Q1 2013, and this demand made them push up 20nm development by a whole year. As well ,it;s rumoured that Apple is interested in using TSMC for either 20nm or 28nm products, and if TSMC does secure that contract, the added investment would ensure that they are caapble of meeting demand in the future.


If it wasn't for Nvidia's claims of yield issues, we could say that the industry is rockin', and demand is so high that they are incapable of meeting it....which would have been a good thing. According to AMD's financial report, TSMC met their needs well, if only barely.

 

[HumanSmoke]

What even more interesting ot me is how it's today, and not a couple of weeks ago, that this news came out, after the GTX690 launch. I think it might have worked out better for both parties if they coudl ahve pulled this news out just before the launch...

Generally speaking, by the time the press release arrives, the deal is signed, sealed and delivered already. The actual mechanics/accounting of revised wafer allocation probably took place some time ago.

When put on high-priority, TSMC will facilitate speedy launch of new NVIDIA GeForce SKUs towards the end of Q2, 2012

We're seven weeks from the end of Q2, if they stated today, a wafer start would need all that time to go from deposition to having the dice cut and packaged...and then you'd need to factor in transhipping of GPU's, AIB binning, board assembly, packaging and shipping of completed retail product.

AMD has several different products using the HP 28NM TSMC process, and all of them are on shelves today.

Just to be pedantic; technically AMD has one GPU (Tahiti) comprising two SKU's (7970 and 7950) using 28nm HP. The other AMD cards are using HPL

 

[Casecutter]

Was there something about TSMC buy tools to add capacity?
http://www.eetimes.com/electronics-news/4371204/TSMC-s-Chang---The-worst-is-behind-us--on-28-nm-

While yes by the time we catch any wind of what's gone on at TSMS it can be weeks past. Normal product production from a wafer start to in the channel can be 9 weeks if all is managed and falls in place. If TSMC started really hitting on all cylinders mid-March the full push should be happening right now!

Just to be pedantic; technically AMD has one GPU (Tahiti) comprising two SKU's (7970 and 7950) using 28nm HP. The other AMD cards are using HPL

Now I only question this because I've read a bunch of different stuff on this and never heard what process which and who is using. I understood AMD went with more tadition and vetted 28nm High-K process, while Nvidia went with TSMC HP HkMP.

"Tahiti is... based on the GCN architecture and built on TSMC’s 28nm High-K process"
http://www.forums.anandtech.com/show/5261/amd-radeon-hd-7970-review/4
http://www.guru3d.com/news/nvidia-illustrates-power-savings-of-28nm-vs-40nm-/

Here's a chart and more about the various manufacturing processes.
http://www.xbitlabs.com/news/other/...Wafers_than_Expected_Projects_Rapid_Ramp.html

While this is much earlier, but shows TSMC exuberance back July, 2011 and is telling of AMD trepidation.
http://www.drivershouse.com/drivers...e-south-nvidia-kepler-tsmc-28nmhkmghphpl.html


I don't know if AMD and Nvidia are make wafers on exactly the same process. Anyone?

 

[HumanSmoke]

afaik, Both TSMC's 28nm High Performance processes are HKMG (HP, HPM and now joined by HPL -Low Power HKMG). Low Power was originally SiON -both for risk sharing and because TSMC wanted to be able to qualify 28nm quickly at relatively low risk on the established process. TSMC have switched (or are switching) Low Power (LP) to Low Power HkMG (HPL) now that HkMG has proven itself with the HP process. I guess LP (SiON) will remain for chipsets etc. for as long as the tooling is available.

Tahiti and GK104 are HP (High Performance HKMG), and Pitcairn/Cape Verde are HPL ( Low power HKMG) from what I understand.

Bit of a blurb from TSMC

 

[newtekie1]

TSMC and nVidia: Still a better love story than Twilight.

 

[Casecutter]

afaik, Both TSMC's 28nm High Performance processes are HKMG (HP, HPM and now joined by HPL -Low Power HKMG). Low Power was originally SiON -both for risk sharing and because TSMC wanted to be able to qualify 28nm quickly at relatively low risk on the established process. TSMC have switched (or are switching) Low Power (LP) to Low Power HkMG (HPL) now that HkMG has proven itself with the HP process. I guess LP (SiON) will remain for chipsets etc. for as long as the tooling is available.

Tahiti and GK104 are HP (High Performance HKMG), and Pitcairn/Cape Verde are HPL ( Low power HKMG) from what I understand.

Bit of a blurb from TSMC

I swear I read that AMD wasn't hot on the HP as they where so far out in front back in Sept. They didn't want to chance their launch on the unproven, and figured they could build Tahiti on HPL which didn't differ from 40Nm. Could that explain the now 1Ghz 7970... the originals witg the lower clocks, and Nvidia saying, "expecting more from the HD 7900". Could AMD have now moved to "HP" since TSMC stopped production and cleaned up the manufacturing. Is there a way to know what a chip was made with, other than what we hear?

But Guru was fairly explicite back in July 2011.
http://www.guru3d.com/news/amd-southern-islands-and-nvidia-kepler-may-use-different-28nm-processes/
Here it is at pcper.com
http://pcper.com/news/General-Tech/Southern-Island-ahead-pack-it-set-low-power-now

A reason for this might be that TSMC probably have their third Gigafab running... Fab 15, that was to come on line first quarter of 2012 and focus on 28nm products on 12in wafer. The April 26th Earnings Call is telling you kind of have understand and read between the share-holder gobbly gook.
http://seekingalpha.com/article/531...sses-q1-2012-results-earnings-call-transcript

While if you start treading these tags together it begins to follow what we've come to know... and a little telling.
http://news.softpedia.com/newsTag/28nm,2