NVIDIA GF100 & Friends speculation

[leoneazzurro]

- GPU packaging
- GPU testing
- component assembly (the more the components, the more the cost)
- Soldering

Not all of those things go into the cost of the graphics card.

Really? Why? These are cost sustained by the chip maker (GPU packaging, testing and shipping) and board maker (PCB, components, assembly, soldering, etc.). So they are part of the manufacturing cost of one card.

of course lets add more to that cost anyhow lets say $40 instead.

Let´s say that

- Packaging
- Shipping
- Bundle & extras (adapters, software, etc)
- Card testing
- Support
- Marketing & advertising

All this is done by the AIB partners so we can kinda put that aside for now.

Not really, these are also costs that are part of the chain for the product to be sold to the final customer, and Nvidia and its board maker are involved in some of this, i.e. software, card testing, marketing. And I forgot about R&D expenses, tooling, etc. in chip making and board making, too. WIP costs could also be quite high in this industry, so could be the stock values.

So, if $299 are what teh AIB are paying for a GTX470 board w/o thir sticker to add to it, then I can believe that (but about margins - it´s really difficult to tell also because the indirect costs). Final user price will be much higher than that, however.

This is an interesting one, not sure who would do this, just a thought, if Flextronics put these cards together before sent out to AIB partners, they might do the testing and AIB partners will do more testing for the overclocked ones?

If they work like I and my colleagues in one power semiconductor firm, then probably Flextronics does all the testing for both the standard and OC cards, then there is a sampled incoming quality testing at the AIB site.

 

[Razor1]

I missed those, could you dig up the links?

http://www.eetimes.com/news/latest/showArticle.jhtml?articleID=222900754

and the other portion haven't been able to find yet, but it was linked actually in this thread somewhere, so many pages ago .

 

[Razor1]

Really? Why? These are cost sustained by the chip maker (GPU packaging, testing and shipping) and board maker (PCB, components, assembly, soldering, etc.). So they are part of the manufacturing cost of one card.



Let´s say that



Not really, these are also costs that are part of the chain for the product to be sold to the final customer, and Nvidia and its board maker are involved in some of this, i.e. software, card testing, marketing. And I forgot about R&D expenses, tooling, etc. in chip making and board making, too. WIP costs could also be quite high in this industry, so could be the stock values.

So, if $299 are what teh AIB are paying for a GTX470 board w/o thir sticker to add to it, then I can believe that (but about margins - it´s really difficult to tell also because the indirect costs). Final user price will be much higher than that, however.



If they work like I and my colleagues in one power semiconductor firm, then probably Flextronics does all the testing for both the standard and OC cards, then there is a sampled incoming quality testing at the AIB site.

No see actually cost of manufacturing is different then cost to market. I'll see if i have sometime later today or tonight to go more into it, but Mize hit the nail on the head.

 

[LordEC911]

FYI, Nvidia sells kits for the reference designs to AIBs.
So yes, most of the BOM is already passed on to AIBs plus a little extra.

So to think that AIBs are only purchasing the GPU and then purchasing the rest of the components from 3rd parties, at least on reference boards, is false.

FYI- For those looking for the final BOM you should read Charlie and Theo's articles on the matter, other than the yields(they both are wrong), the other costs seem to be pretty accurate.

 

[leoneazzurro]

Of course there are manufacturing costs and costs sustained to market a product, but -usually- a production firm should cover both to make a profit.
Summing it up, if the final price of one board is set lower than the costs of the components, R&D, assembly, marketing, distribution, etc. then someone in the chain is losing money on it and that someone is not happy.

 

[Alexko]

http://www.eetimes.com/news/latest/showArticle.jhtml?articleID=222900754

and the other portion haven't been able to find yet, but it was linked actually in this thread somewhere, so many pages ago .

There's nothing there, except for this:

Nvidia stopped short of slamming TSMC. TSMC has improved their yields and is doing a ''fabulous job,'' the company said.

Which is vague at best. Then again JHH also said that NV's current mainstream offering is "fabulous" so maybe "fabulous" is Nvidian for "mediocre".

 

[Razor1]

There's nothing there, except for this:



Which is vague at best. Then again JHH also said that NV's current mainstream offering is "fabulous" so maybe "fabulous" is Nvidian for "mediocre".

then look back in this thread, someone put up the link, another stated about it, and charlie came in and stated the yield amounts for the gtxxx line, I remember that, but don't remember the page it was on.

 

[MfA]

and the other portion haven't been able to find yet, but it was linked actually in this thread somewhere, so many pages ago .

There's was a statement from ancient antiquity from TSMC saying they had 60% yield, but no one believed it then either ... also yields are not design independent of course. So yeah I'm just going to have to call shenanigans on the previous assertions ... NVIDIA hasn't said the yields are 60% nor have they said that the yields for Fermi are comparable to GT2xx. (Even if they did, what GT2xx? They might just have very bad yields on a 40 nm GT2xx for entirely different reasons.)

I don't really see the problem with launching with low yields ... ATI did it. I'm sure they all prefer not to have to do it, but you go with what you can get.

 

[Razor1]

There's was a statement from ancient antiquity about TSMC saying they had 60% yield, but no one believed it then either ... also yield are not design independent of course.

it wasn't that old of post, actually, it was soon after TSMC announcement , hmm actually maybe it was in the doom and gloom thread. It had nothing to do with the % of yields just that the yields are now as good as 65nm yields.

http://www.digitimes.com/news/a20100120PD204.html

 

[FrameBuffer]

I think you guys are confusing margin with profit. Even if nV's direct costs come in below MSRP that only means that make MSRP-COGS = GM. For many companies a GM% below 60% is loss making. That means you have to sell for COGS*3 basically to make a dime. (for those who don't know GM$ = sale price - COGS and GM% = (sale price - COGS)/sale price. This is because overhead (r&d, marketing, depreciations, etc.) is a huge cost. That's why I asked what is the typical % of MSRP for the raw chip...if anyone knows that will tell.

Unless I totally failed at accounting 101 wouldn't MSRP be out of NV's reign as the actual sales of the GPU/Cards to the IHVs (Foxconn, Asus, MSI etc) then subtract COGS would yield GM. In the case of MSRP, it would have little to no difference on NV's balance sheet or income statement, it would however play on said hardware vendors financials. I would assume the numbers used would imply use of GAAP and using absorption cost model and not variable where fixed costs where they would be then applied as period costs. This would also mean that while part yields have a larger contribution factor (than say materials), many of the other not dealing directly with the product's production etc would be applied.

As far as yield numbers, it's been a bit since I heard from a friend or 3 but I heard at best they were not even close to GT2X0 numbers, which questions whether the rate was in reference to the GF200 (GTX 260-285) or the GT21X (GT 205-240) where 40nm has been doing well enough for some time.. but then again like I said.. it's been a while (about 5 weeks). I'd take the 60% with a truck load of salt personally.

 

[Alexko]

then look back in this thread, someone put up the link, another stated about it, and charlie came in and stated the yield amounts for the gtxxx line, I remember that, but don't remember the page it was on.

I do remember Charlie stating that the yields for GT200 eventually reached about 60%, and I kinda sorta recall NV saying something kinda sorta concrete about 40nm yields in general, but I don't remember them ever saying anything even remotely concrete about GF100's yields.

From time to time we do get statements from various sites saying stuff like "TSMC's yields are 40%", "no wait, they are 60%", "well maybe they're (still) 40% (again) after all"... which really doesn't mean much anyway, but that's about it.

it wasn't that old of post, actually, it was soon after TSMC announcement , hmm actually maybe it was in the doom and gloom thread. It had nothing to do with the % of yields just that the yields are now as good as 65nm yields.

http://www.digitimes.com/news/a20100120PD204.html

OK. Well that really doesn't mean much about GF100.

And for what it's worth, I don't even think it's true as far as Cypress is concerned, because if it were, there would be no need for the HD 5830, or at least it wouldn't have that many disabled units. The HD 5750 and 5550 are further proofs of this, and those are much smaller dies.

 

[MfA]

It had nothing to do with the % of yields just that the yields are now as good as 65nm yields.

That one didn't have the 60% number though which had to come from somewhere. Yields being as good as 65nm can mean a lot of things, for instance it can mean that the yield of the "same" design is identical on 40nm ...

 

[Razor1]

That one didn't have the 60% number though which had to come from somewhere. Yields being as good as 65nm can mean a lot of things, for instance it can mean that the yield of an optically shrunk design is identical on 40nm ...

That's true but this is going to be a hard launch, at least for the 470, not sure about the 480 though.

 

[MfA]

That's true but this is going to be a hard launch, at least for the 470, not sure about the 480 though.

It's going to be a launch with product, no one from NVIDIA has said that it won't be a launch with shortages though ... otherwise why the hedging about Q2?

 

[Razor1]

It's going to be a launch with product, no one from NVIDIA has said that it won't be a launch with shortages though ... otherwise why the hedging about Q2?

April is one week away from the launch that is q2

 

[MfA]

Assuming financial quarters, you know the routine ... regardless, no one from NVIDIA has said that it won't be a launch with shortages.

 

[Mize]

Unless I totally failed at accounting 101 wouldn't MSRP be out of NV's reign as the actual sales of the GPU/Cards to the IHVs (Foxconn, Asus, MSI etc) then subtract COGS would yield GM.

You're absolutely correct. I should have written "SP" instead of MSRP for sale price (to the reseller). In my business we don't have "retail prices" as we sell industrial controls direct to end-users.

 

[Silent_Buddha]

no I'm talking about keeping on the selves, they won't be able to produce enough, fast enough to keep inventory at acceptable levels, its possible if the chip was smaller but that isn't the case here.

So basically same situation as AMD was in, although in this case possibly worse.

Pricing is a way to manage supply and demand. If you have low supply it behooves the company to price it high enough that demand is just higher than your ability to supply product.

Thus, Nvidia quite possibly launching at 499 and 649 USD as a way to manage demand for a product that is going to be in very short supply, even if benchmarks end up having them only having a slight lead in perf.

Regards,
SB

 

[KimB]

I think you guys are confusing margin with profit. Even if nV's direct costs come in below MSRP that only means that make MSRP-COGS = GM. For many companies a GM% below 60% is loss making. That means you have to sell for COGS*3 basically to make a dime. (for those who don't know GM$ = sale price - COGS and GM% = (sale price - COGS)/sale price. This is because overhead (r&d, marketing, depreciations, etc.) is a huge cost. That's why I asked what is the typical % of MSRP for the raw chip...if anyone knows that will tell.

I should mention that many of those costs are independent of the number of units sold, so it's not quite as cut-and-dried as this. The lower the price, the more units they can sell, and thus the lower their profit margin needs to be to earn an overall profit. Typically companies go through quite a bit of analysis to attempt to determine the optimum price point to maximize profits. Lower the price too much, and what few additional units are purchased don't make up for the costs. Increase the price too much, and too few are purchased to overcome the fixed costs (e.g. R&D).

As I said earlier, I think the expected price point should be slightly higher than ATI's cards given their average performance level (whatever that happens to be).

 

[Mize]

Absolutely Chalnoth, I was just trying to point out the fallacy of trying to link direct materials cost of the main component to final sale price. Heck there are industries where you can loose money with a COGS multiplier of 10 thanks to NRE and low volumes.