Nvidia Ampere Discussion [2020-05-14]

What would be the point of dual sourcing from both Samsung and TSMC? The cost of doing two different designs and wafer cost increase from TSMC would negate any price advantage they got from Samsung

They will probably release Geforce on TSMC but i dont think it will happen within the next 6 months. This time next year perhaps
 
The only statement made was from Jensen who indicated Samsung would be handling a small portion of Ampere capacity, and TSMC handling the major portion.

Where did he say that? That only makes sense If the 3070 and below is on TSMC, because GA102 on Samsung is clearly higher volume that A100 on TSMC. And I seem to recall that 3070 is on Samsung too.
 
Where did he say that? That only makes sense If the 3070 and below is on TSMC, because GA102 on Samsung is clearly higher volume that A100 on TSMC. And I seem to recall that 3070 is on Samsung too.

It's a statement that keeps getting brought up but I don't know of a source of the actual direct quote. The original report was picked up by the western rumour mill from a Taiwanese site which was reporting on a media interview (not clear on the source of this interview, but presumably to Taiwan media?) that Jensen Huang was asked about the rumours of Nvidia choosing Samsung's 7nm process for it's next generation GPUs, and the reported reply was that "most 7nm process orders will be given to TSMC and that Samsung would receive a small number of orders."

However there's several considerations here -

- As far as I know there isn't a direct quote but interpenetration of his reply, so we don't know specifically what he said and the entire context which can be very important.

- The statement is strictly true as of now as 7nm Ampere is at TSMC (GA100). There is no Samsung 7nm Ampere GPUs that we know of currently. Samsung Ampere is 8nm (GA102 GA104).

- This was reported in Dec 2019. The immediate upcoming GPU then was GA100. The rumours then were specifically regarding Samsung 7nm and the equivalent generation. A lot of things could have changed since then. How long would such a statement be "binding" (in the soft sense) for? I mean it would be insanity if we take that to assume Nvidia will somehow forever be bound to having a specific ratio of 7nm class products that favor TSMC over Samsung.
 
Last edited:
Also just to add it should be kept in mind that the entire Ampere stack almost certainly hasn't been released yet. There have been reports on GA106 and GA107 existing which would not be surprising given past history. There is also reports of a GA103 and to a lesser extent GA105 existing. The processes that would be used for these is not a certainty yet. For all we know the TSMC 7nm rumours for next year are just a result of the above and nothing to do with GA102/104.
 
Where did he say that? That only makes sense If the 3070 and below is on TSMC, because GA102 on Samsung is clearly higher volume that A100 on TSMC. And I seem to recall that 3070 is on Samsung too.

During GTC 2019 in Suzhou, China, NVIDIA's CEO responded to the press that the majority orders of their next-generation 7nm GPU will be handled by TSMC with Samsung only playing a small role than previously reported.
...
NVIDIA's CEO also stated that his company has had close relations with TSMC in the past with TSMC producing their previous 16nm (Pascal) and 12nm (Volta / Turing) GPUs. NVIDIA also highlighted its Turing GPU architecture which is built upon TSMC's 12nm process node and offers better efficiency and performance than rivaling 7nm process based products. NVIDIA's CEO also stated that without TSMC's and their advanced process technologies, they wouldn't have been so successful, hence this partnership between TSMC and NVIDIA is very important for them.
https://wccftech.com/nvidia-tsmc-7nm-next-generation-gpu-ampere/
 
One of their biggest customers has moved over to 5nm already, another is out completely. Is there anything solid saying there's capacity issues at 7nm?

From AMD's Q2 earnings call, question to Lisa Su - "Hi, guys. Thanks for taking my questions. I wanted to follow-up on that capacity point. What is your capacity and supply situation look like? And is any of the full-year raise related to capacity freeing up at your foundry partners? And maybe put another way, are you supply rather than demand-limited at this point? What does that capacity situation look like?"

Answer: "Yes, sure, Stacy. So look, we have a strong supply chain. There's no question. It's been a very dynamic year if you just think about all the puts and takes over the last four or five months.

I've said before and I'll say again, 7-nanometer is tight, and we continue to partner closely with TSMC to ensure that we can satisfy our customer demand. When you ask about the full-year raise, the full-year raise is because demand has gone up from our initial expectations, and some of that is due to the market, and some of that is due to the strength of our product traction. We are increasing capacity to meet those needs, but it is tight. And I would say that as we continue to increase capacity, we see opportunity there."


TSMC slides at its recent symposium showed that 7nm has had close to 500 tapeouts, with a forecast for more than 200 N7/N7+ new tapeouts (NTOs) in 2020. TSMCs 7nm is clearly in high demand from a variety of customers and even Apple moving to 5nm dosen't mean they stop using 7nm completely. Also given the cycle times involved in wafer production, product manufacturing and distribution, freed up capacity will not be apparent in the market immediately. I expect we'll hear further on the supply situation in AMD's Q3 earnings call on the 27th of this month.

Edit: To add, AMD is likely to prioritize 7nm wafer capacity towards higher margin CPU products than consumer GPU.
 
Last edited:
TSMC slides at its recent symposium showed that 7nm has had close to 500 tapeouts, with a forecast for more than 200 N7/N7+ new tapeouts (NTOs) in 2020. TSMCs 7nm is clearly in high demand from a variety of customers and even Apple moving to 5nm dosen't mean they stop using 7nm completely. Also given the cycle times involved in wafer production, product manufacturing and distribution, freed up capacity will not be apparent in the market immediately. I expect we'll hear further on the supply situation in AMD's Q3 earnings call on the 27th of this month.

Edit: To add, AMD is likely to prioritize 7nm wafer capacity towards higher margin CPU products than consumer GPU.

AMD doubled their 7nm capacity from 1H'20 to 2H'20 and with ~20% of TSMC's 7nm capacity now, they have probably also gained additional capacity in 1H'21 for their "mainstream" releases.
Multiple other customers of 7nm have moved to 5nm and more in 1H'21.
Hisilicon is in limbo until 2H'21 and there have been rumors of Qualcomm moving to Samsung.

Anyone moving to TSMC's EUV is going to free up slightly more capacity as well, as they have a higher throughput than DUV due to less layers/exposures.
 
Another rumored Ti in the works ... stated availability on Nov 30

0bb5b3d7-2cb9-48aa-aaa1-d793d9318081.jpg



2e806682-e5fc-4fd3-acfe-a50cdd787915_resized_400x423.jpg

https://hexus.net/tech/news/graphics/146437-nvidia-geforce-rtx-3060-ti-cards-spotted-sale-china/
 
The Ampere product line seems all over the place at the moment. I'm looking forward to the 3070 announcement (is that today)? But it looks like it might be replaced, or at least made much less relevant within weeks, so I'm be holding out for a while longer until we have a clearer view of the product stack as well as Navi2 performance which itself may disrupt Nvidia pricing.

Also with all the crazy performance jumps we're seemingly getting from every direction at the moment, I'm not sure if I'd actually be satisfied with 2080Ti performance now. Didn't think I'd be saying that a couple of months ago!
 
The Ampere product line seems all over the place at the moment. I'm looking forward to the 3070 announcement (is that today)? But it looks like it might be replaced, or at least made much less relevant within weeks, so I'm be holding out for a while longer until we have a clearer view of the product stack as well as Navi2 performance which itself may disrupt Nvidia pricing.

Also with all the crazy performance jumps we're seemingly getting from every direction at the moment, I'm not sure if I'd actually be satisfied with 2080Ti performance now. Didn't think I'd be saying that a couple of months ago!

It does seem like chaos. However, the same leakers we’re talking about Ampere PCB and die specs 6 months to a year before launch. They seem to be getting info quite early. Maybe these SKUs aren’t launching until next summer.
 
Back
Top