POLL: nVidia's Silence on GT300

[HurleyBird]

Even with the volume you aren't going to make up the difference in cost between G200 and RV770 dies. It of course depends on the agreement between TMSC and Nvidia, but as a general rule you shave off 10% each time you double volume on a chip. I also think don't know if Nvidia even has higher volume of G200, since a greater percentage of their volume would be G92, and RV770 competes with both of those parts.

 

[Keysplayr]

Originally posted by: HurleyBird
Even with the volume you aren't going to make up the difference in cost between G200 and RV770 dies. It of course depends on the agreement between TMSC and Nvidia, but as a general rule you shave off 10% each time you double volume on a chip. I also think don't know if Nvidia even has higher volume of G200, since a greater percentage of their volume would be G92, and RV770 competes with both of those parts.

Still means that we don't know. That would be the safest answer. And 55nm is not restricted to G200. A wafer is a wafer and you'd have to apply any and all hypothetical discounts across any 65nm (at launch) and 55nm products.

 

[alyarb]

You can fit about double the number of RV770 dice onto a wafer compared to GT206 and there is greater market penetration of RV770 products than GT206. Yields don't matter too much at that point, but they would have to be cast outrageously in nvidia's favor for both companies to get the same amount of good dice per wafer. Of course we will never know the exact cost of a finished wafer. That is privileged information. Is there evidence to suggest that RV770 is cheaper to produce than GT206? Yes there is.

Is there any evidence to suggest that GT206 is cheaper to produce than RV770? No, there is none.

 

[Keysplayr]

Originally posted by: alyarb
You can fit about double the number of RV770 dice onto a wafer compared to GT206 and there is greater market penetration of RV770 products than GT206. Yields don't matter too much at that point, but they would have to be cast outrageously in nvidia's favor for both companies to get the same amount of good dice per wafer. Of course we will never know the exact cost of a finished wafer. That is privileged information. Is there evidence to suggest that RV770 is cheaper to produce than GT206? Yes there is.

Is there any evidence to suggest that GT206 is cheaper to produce than RV770? No, there is none.

Still means that we don't know. Doesn't matter if double the number of dice onto a wafer if Nvidia is being charged 1/2 as much for their orders as AMD. You can doubt this, but you can't prove or disprove it either. You see where this goes? To infinity. We don't know. Pretty much done.

 

[alyarb]

haha no, by all means. I'm sure TSMC is wholly indebted to nvidia and no one else for their business. In fact, I bet the nvidia wafers are free. that would explain why they can't even profit $800m after $4 billion in sales. i'm sure they spent $3.2b on R&D and left the marketing up to the unpaid interns. Now there is something you may be able to clarify for us.

 

[waffleironhead]

Originally posted by: Keysplayr

Still means that we don't know. Doesn't matter if double the number of dice onto a wafer if Nvidia is being charged 1/2 as much for their orders as AMD. You can doubt this, but you can't prove or disprove it either. You see where this goes? To infinity. We don't know. Pretty much done.

yes, but we can speculate. Which is all we do around here these days. :shrug;

 

[EnzoLT]

none, i just clicked View Results

 

[Obsoleet]

Originally posted by: alyarb
you can't be serious; larrabee is the least likely architecture to succeed in D3D and there is absolutely no performance data available to support this or that outcome.

All that has to be done is pull everyone to Larrabee and NV is done for. Flood the market with it, and get AMD on board who also has an x86 license and it doesn't have to "succeed". It just has to be fast-enough, which it will be.
In the meanwhile, they won't even be able to hold their ground in competition with ATI. It's really not looking good for Nvidia no matter how one wants to look at things. I really wish people would quit pretending it isn't as if Nvidia has a future in this market. Simply because they are on top now and a very boisterous, loud company with loud supporters dazzled like Creative EAX fanboys of old, won't help anymore. Which btw, 3d vision, Physx and the rest of the trinkets will go the way of EAX because no one really cares and everyone in this forum knows that it's true.

 

[Cookie Monster]

Originally posted by: Keysplayr
We've never had any numbers to prove, or disprove this.

Here it is

 

[OCGuy]

Originally posted by: Obsoleet

Originally posted by: alyarb
you can't be serious; larrabee is the least likely architecture to succeed in D3D and there is absolutely no performance data available to support this or that outcome.

All that has to be done is pull everyone to Larrabee and NV is done for. Flood the market with it, and get AMD on board who also has an x86 license and it doesn't have to "succeed". It just has to be fast-enough, which it will be.
In the meanwhile, they won't even be able to hold their ground in competition with ATI. It's really not looking good for Nvidia no matter how one wants to look at things. I really wish people would quit pretending it isn't as if Nvidia has a future in this market. Simply because they are on top now and a very boisterous, loud company with loud supporters dazzled like Creative EAX fanboys of old, won't help anymore. Which btw, 3d vision, Physx and the rest of the trinkets will go the way of EAX because no one really cares and everyone in this forum knows that it's true.

We will see.

 

[HurleyBird]

Originally posted by: KeysplayrStill means that we don't know. That would be the safest answer. And 55nm is not restricted to G200. A wafer is a wafer and you'd have to apply any and all hypothetical discounts across any 65nm (at launch) and 55nm products.

Sure, there is some economy of scale stuff working in TSMCs favor in terms of the number of 55nm wafers it produces, but that applies to everyone. If AMD, Nvidia, or anyone else increases the number of 55nm wafer orders, costs go down for 55nm wafers across the board and TMSC could theoretically charge everyone less. The main economy of scale is still going to be the number of wafers for a specific part you order though. Extreme example: If Nvidia were to order a single wafer for a new g200b sized part it would be several times more expensive for TMSC to make than a g200b wafer.

You're right that we can't know costs for sure, but it's easy enough to tell where one company has the advantage. ATI is almost certainly paying more for memory, Nvidia is paying more for large G200b dies vs. much smaller RV770 dies, Nvidia is probably paying more for more complex circuit boards, etc.

Originally posted by: Cookie Monster

Originally posted by: Keysplayr
We've never had any numbers to prove, or disprove this.

Here it is

Memory prices seem WAY off.

 

[Qbah]

Originally posted by: Cookie Monster

Originally posted by: Keysplayr
We've never had any numbers to prove, or disprove this.

Here it is

Based on that neither nVidia nor AMD sold below manufacturing costs and nVidia had some pretty amazing profits - no idea how accurate it is, but looks legit. So the argument about nVidia selling with a loss and AMD cards being a lot cheaper to produce can be pretty much dropped

And wow, the GTX280 costs 160$ to make. Even assuming initial costs were higher, that's some amazing margins on a card (it was like 650$ at launch!). Earning money at its finest

Also, I wonder if those numbers include yields? That would only influence the GPU cost though, but still.

 

[Cookie Monster]

You can question the source, although it is straight from the full q2 mercury research graphics report.

GTX280 uses 16 memory chips compared to 8 memory chips used on the HD4890/4870 for instance. If you divide the cost by two (to compare the cost of 8 GDDR3 memory chips to 8 GDDR5), it does indeed show GDDR3 being cheaper than GDDR5.

 

[Idontcare]

Originally posted by: HurleyBird
Memory prices seem WAY off.

Yeah its pretty clear they are just using numbers that are representative of the ASP's at some point in time.

ASP's of course change over time as well as being contract specific, something Mercury research guys would not be lawfully knowledgeable about.

No doubt they do unofficially know the rather intimate details of contract prices for each (as a static snapshot of APSs at the time the report is generated), so their reported numbers are going to be bound by some level of rational.

Originally posted by: Qbah
Also, I wonder if those numbers include yields? That would only influence the GPU cost though, but still.

Yields, like ASPs, are not static, they change over time so even if Mercury researchers had the yield data to 5-9's certainty the data was no longer valid to that degree of certainty the day after the report was published.

But yes they do factor in ball-park yield impacts into the GPU costs and for all practical purposes it is as good as you'll ever get in terms of actionable data (albeit static) regarding BOM structure between these two companies.

 

[Keysplayr]

Originally posted by: HurleyBird

Originally posted by: KeysplayrStill means that we don't know. That would be the safest answer. And 55nm is not restricted to G200. A wafer is a wafer and you'd have to apply any and all hypothetical discounts across any 65nm (at launch) and 55nm products.

Sure, there is some economy of scale stuff working in TSMCs favor in terms of the number of 55nm wafers it produces, but that applies to everyone. If AMD, Nvidia, or anyone else increases the number of 55nm wafer orders, costs go down for 55nm wafers across the board and TMSC could theoretically charge everyone less. The main economy of scale is still going to be the number of wafers for a specific part you order though. Extreme example: If Nvidia were to order a single wafer for a new g200b sized part it would be several times more expensive for TMSC to make than a g200b wafer.

You're right that we can't know costs for sure, but it's easy enough to tell where one company has the advantage. ATI is almost certainly paying more for memory, Nvidia is paying more for large G200b dies vs. much smaller RV770 dies, Nvidia is probably paying more for more complex circuit boards, etc.

Originally posted by: Cookie Monster

Originally posted by: Keysplayr
We've never had any numbers to prove, or disprove this.

Here it is

Memory prices seem WAY off.

So do the cores. 65.00 for a GTX280 65nm core, and 90.00 for a GTX285 55nm core.
How does that compute?
They don't even list the correct memory amount for GTX280 and GTX275. Wonder how many other things they got wrong.

 

[Idontcare]

Originally posted by: Keysplayr
They don't even list the correct memory amount for GTX280 and GTX275. Wonder how many other things they got wrong.

You know what a consultant does, right? A consultant takes your watch off your arm, reads it and tells you the time, and then puts your watch back on your arm.

You know what a market researcher does, right? They take the watch off a consultants arm, read it...well you know the rest of the joke by now. :laugh:


(preventative misinterpretation disclaimer: this was a joke, no consultants or market researchers were harmed or otherwise injured in any way in the creation of this joke, the author of this post was himself at one time an overpaid consultant who enjoys, to this day, the seemingly lost art of self-denigration)

 

[SlowSpyder]

Originally posted by: Keysplayr

Originally posted by: HurleyBird

Originally posted by: KeysplayrStill means that we don't know. That would be the safest answer. And 55nm is not restricted to G200. A wafer is a wafer and you'd have to apply any and all hypothetical discounts across any 65nm (at launch) and 55nm products.

Sure, there is some economy of scale stuff working in TSMCs favor in terms of the number of 55nm wafers it produces, but that applies to everyone. If AMD, Nvidia, or anyone else increases the number of 55nm wafer orders, costs go down for 55nm wafers across the board and TMSC could theoretically charge everyone less. The main economy of scale is still going to be the number of wafers for a specific part you order though. Extreme example: If Nvidia were to order a single wafer for a new g200b sized part it would be several times more expensive for TMSC to make than a g200b wafer.

You're right that we can't know costs for sure, but it's easy enough to tell where one company has the advantage. ATI is almost certainly paying more for memory, Nvidia is paying more for large G200b dies vs. much smaller RV770 dies, Nvidia is probably paying more for more complex circuit boards, etc.

Originally posted by: Cookie Monster

Originally posted by: Keysplayr
We've never had any numbers to prove, or disprove this.

Here it is

Memory prices seem WAY off.

So do the cores. 65.00 for a GTX280 65nm core, and 90.00 for a GTX285 55nm core.
How does that compute?
They don't even list the correct memory amount for GTX280 and GTX275. Wonder how many other things they got wrong.

My *guess* is they just got the numbers switched on the GTX275/280... just a typo. As far as the core prices go, does TSMC charge more for 55nm than 65nm? Though that seems like a pretty heafty premium.

I wouldn't take those prices as absolute, but I would imagine they are very likely in the ballpark. Looking at the chart some of the cards from AMD and Nvidia are pretty close in costs, take the 4890 vs. GTX275 for example. While others in the current generation do appear to be cheaper for AMD, such as the 4870 vs. the GTX260... if you use the 4890's ram price to guesstimate the cost for the 4870 @ 1GB there is a ~$20 difference, multiply that times thousands to tens of thousands of cards and that is fairly substantial.

 

[Avalon]

Probably just silent because they are a little later than expected, but from the rumor mill it sounds like they have a great GPU on their hands.

 

[Idontcare]

Originally posted by: SlowSpyder
As far as the core prices go, does TSMC charge more for 55nm than 65nm?

Yes.

 

[exar333]

I agree with some other posters that they will wait to release a marketing blitz to steal some of AMD's thunder, as soon as the 58xx is about to get available. This accomplishes a few things:

-Minimizes canibalization of current products.
-Steals of AMD's release thunder.
-Attempts to make people re-think their AMD purchase and wait for the NV release to make their purchasing decision.

Edit: I couldn't resist the first option in the poll, it was too fun!

 

[Martimus]

I picked other, because I just think they are trying to avoid canabalizing current sales with news of the next generation product - regardless of whether they believe it will be competitive with AMD or not.

 

[Tempered81]

Im still laughing at 19 people choosing the "I'm an idiot" choice in the poll!


HAHAHA Great!

 

[error8]

Originally posted by: kreacher
With their marketing department's past record, being silent means the situation is so bad right now that they can't even think of a way to spin it into anything positive.

Something like that. When the GT200 was about to be launched, you heard nothing more than just Nvidia, all over the place.
Now, it's so much silence that it blows my ear drums. It's becoming scary, because I really want these cards to create competition to ATi's. Otherwise, ATi will inflate the prices to their new generation as much as they want.

 

[Jaskalas]

Due to their words on DX11, I would guess that they are behind.

 

[OCGuy]

Originally posted by: Jaskalas
Due to their words on DX11, I would guess that they are behind.

They never bashed DX11, they just said it wont be the only reason to buy their next gen cards.

Unless I missed a quote where the bashed it?