ATI partners don't want to reduce prices

[Munky]

Originally posted by: ExarKun333

Originally posted by: error8

Originally posted by: ExarKun333

Many people here on the forums DO care about other features such as CUDA. I find it a great addition to the Nvidia's products. It DOES add value to a product, but may NOT add value for a specific person. Don't confuse the two.

I never said that CUDA is useless . It's only useless for me and probably for other persons too. CUDA is great for those that use it, like you for example and if I would have used it myself, of course that I would have bought a Nvidia card, there is no doubt about that. But as long as you don't need it, you might turn your eyes into the other camp, when buying a videocard, especially if it's cheaper.

I think we agree.

That's why I originally went with a GTX 260 was due to it's price coming in right at the same as the 4870, but I had gotten used to the CUDA functionality. If the 4870 price cuts had hit a few months ago, I would honestly probably picked that up instead. The no-brainer here is the economy too, but the price cuts should help get their products out there.

One could say the same thing about DX10.1 support on Ati cards, which Nvidia cards lack. If someone actually uses the feature, then it's an advantage, otherwise it's just another checkbox on the spec sheet.

 

[wjgollatz]

Reducing prices is just not about sales. The article did not say which of the partners want to resist the prices. Some makers make boards with ATI and Nvidia chipsets also. AMD/ATI is probbaly more worried about brand name.

When someone has a computer and hardware they like, and they think works well, even if it does not, they will want to keep buying the same product. How many people here now buy Western Digital just because its WD? You had to have had a couple drives you were happy about at least. Same with printers. I used HP, and I trust HP. There would have to be a hell of a bargain for me to go to another brand. The more people that own a brand's hardware, increase the chances of a new sale. And these board makers certainly do not want to have to worry about stock from different chipsets in their warehouses. They want it simple, and to make money on the sale, whatever the chipset is, Nvidia or ATI while ATI wants to increase its market share.

 

[Munky]

Originally posted by: error8

Originally posted by: Falloutboy
couldn't this be more ati putting more pressure on nvidia? atis parts are cheaper ot make atm as far as I know, so forcing nvidia to match the lower prices might hurt them more, and would be a wash for ati if they pick up marketshar

I don't think Nvidia cards are more expensive then ATi's. Now GTX 260, 285, 250 are all made on 55 nm, the PCB is also cheaper then it was before, if I'm not mistaken. These cards should be on par at production costs with 4850, 4870 cards.

The gtx 285, 280 and 260 all use a gigantic die, considerably larger and packing more transistors than the rv770, even on 55nm process. Also, the boards are larger and more complex because they utilize a 448 or 512-bit memory bus, as opposed to 256-bit. Unless there is explicit evidence stating otherwise, I can say with 99% confidence that the gt200-based NV cards cost more to make than the rv770 Ati cards.

 

[chizow]

This isn't surprising news given some of the 4870 price bounce backs in the FS/FT forum yesterday. There were some blurbs a few months ago about how AMD was desperate to regain market share and initiated some price cuts to stimulate sales, but Nvidia has matched or exceeded their price cuts on the GTX 260 at every step of the way. I think the AMD board partners are making it clear they're perfectly happy selling fewer cards at higher margins, we'll have to see if Nvidia partners also pull back on their price cuts though, as it seems there's a few 260s on sale for $160-170.

 

[chizow]

Originally posted by: munky
The gtx 285, 280 and 260 all use a gigantic die, considerably larger and packing more transistors than the rv770, even on 55nm process. Also, the boards are larger and more complex because they utilize a 448 or 512-bit memory bus, as opposed to 256-bit. Unless there is explicit evidence stating otherwise, I can say with 99% confidence that the gt200-based NV cards cost more to make than the rv770 Ati cards.

And how much more does 1GB GDDR5 cost than 1GB GDDR3? People like you have been 99% confident you had a clue since G80, when really, you don't.

 

[error8]

Originally posted by: munky

Originally posted by: error8

Originally posted by: Falloutboy
couldn't this be more ati putting more pressure on nvidia? atis parts are cheaper ot make atm as far as I know, so forcing nvidia to match the lower prices might hurt them more, and would be a wash for ati if they pick up marketshar

I don't think Nvidia cards are more expensive then ATi's. Now GTX 260, 285, 250 are all made on 55 nm, the PCB is also cheaper then it was before, if I'm not mistaken. These cards should be on par at production costs with 4850, 4870 cards.

The gtx 285, 280 and 260 all use a gigantic die, considerably larger and packing more transistors than the rv770, even on 55nm process. Also, the boards are larger and more complex because they utilize a 448 or 512-bit memory bus, as opposed to 256-bit. Unless there is explicit evidence stating otherwise, I can say with 99% confidence that the gt200-based NV cards cost more to make than the rv770 Ati cards.

The old GTX 260 and 280 were the 65 nm behemonts and were very expensive to make indeed. The "new" gtx c216 and 285 are made on 55 nm and don't be fooled by the massive heat spreader, since the gpu hiding underneath is smaller.
512 bit or 448 bit memory interface is indeed expensive, but GDDR3 isn't. The 256 bit memory bus is cheaper, but GDDR5 isn't. It's pretty much the same thing, when it comes to costs here.

http://www.hardwarecanucks.com...deo-card-review-8.html

This link provides a comparison between a GTX 260 on 65nm and one on 55 nm. You can see the latter has a cleaner PCB, there aren't any memory chips on the back, the power circuitry is simpler, feeding a less hungry gpu. Their coolers are also different, shorter and cheaper on the 55 nm card.

Who knows which exactly is the more expensive card between ATi offerings and Nvidia's, but I'd say that overall, they're at about the same level.

 

[Munky]

Originally posted by: chizow

Originally posted by: munky
The gtx 285, 280 and 260 all use a gigantic die, considerably larger and packing more transistors than the rv770, even on 55nm process. Also, the boards are larger and more complex because they utilize a 448 or 512-bit memory bus, as opposed to 256-bit. Unless there is explicit evidence stating otherwise, I can say with 99% confidence that the gt200-based NV cards cost more to make than the rv770 Ati cards.

And how much more does 1GB GDDR5 cost than 1GB GDDR3? People like you have been 99% confident you had a clue since G80, when really, you don't.

How much more does 1GB GDDR3 cost compared to 512MB GDDR5? Maybe you should have thought of that option before spewing more nonsense.

 

[chizow]

Originally posted by: munky
How much more does 1GB GDDR3 cost compared to 512MB GDDR5? Maybe you should have thought of that option before spewing more nonsense.

Rofl, it doesn't cost more, that's the whole point. Case in point, I can get a 1GB 4850 for less than a 512MB 4870 when the only differences are going to be a few ICs and the RAM.

As for spewing nonsense lol. Again, you and others have claimed on numerous occasions to have some clue about board pricing, yet historical market pricing has shown you to be wrong time and time again. The GTX 260 is now priced at $160-170 only 9 months after launch, again, proving Nvidia has more flexibility than many thought despite the massive die size on GT200. It probably has something to do with the fact TSMC's fab's are sitting at ~25% utilization and that glass and metal are cheap lol.

To take the comparison further, a consideration I'm sure you didn't take into account was product diversification with regard to die size and wafer distribution. For example, a 300mm RV770 wafer at 55nm with ~292 dice might be distributed like:

• 30% 4830 @ <$100
  40% 4850 @ <$150
  20% 4870 @ <$200
  10% 4870X2 @ <$225 (prorated per core)

In comparison, a 300mm GT200b wafer at 55nm with ~120 dice might be distributed like:

• 60% GTX 260 @ <$200
  20% GTX 285 @ <$350
  20% GTX 295 @ <$250 (prorated per core)

Without doing the exact calculations (the % obviously aren't going to be accurate although they're pretty reasonable) you can clearly see that even though Nvidia will get fewer dice per wafer, they're also selling the majority of chips for the maximum price ATI is selling the minority of their chips for. In any case, I think this clearly demonstrates simple die size comparisons don't tell the whole story with regards to die prices.

 

[Munky]

Originally posted by: error8

Originally posted by: munky

Originally posted by: error8

Originally posted by: Falloutboy
couldn't this be more ati putting more pressure on nvidia? atis parts are cheaper ot make atm as far as I know, so forcing nvidia to match the lower prices might hurt them more, and would be a wash for ati if they pick up marketshar

I don't think Nvidia cards are more expensive then ATi's. Now GTX 260, 285, 250 are all made on 55 nm, the PCB is also cheaper then it was before, if I'm not mistaken. These cards should be on par at production costs with 4850, 4870 cards.

The gtx 285, 280 and 260 all use a gigantic die, considerably larger and packing more transistors than the rv770, even on 55nm process. Also, the boards are larger and more complex because they utilize a 448 or 512-bit memory bus, as opposed to 256-bit. Unless there is explicit evidence stating otherwise, I can say with 99% confidence that the gt200-based NV cards cost more to make than the rv770 Ati cards.

The old GTX 260 and 280 were the 65 nm behemonts and were very expensive to make indeed. The "new" gtx c216 and 285 are made on 55 nm and don't be fooled by the massive heat spreader, since the gpu hiding underneath is smaller.
512 bit or 448 bit memory interface is indeed expensive, but GDDR3 isn't. The 256 bit memory bus is cheaper, but GDDR5 isn't. It's pretty much the same thing, when it comes to costs here.

http://www.hardwarecanucks.com...deo-card-review-8.html

This link provides a comparison between a GTX 260 on 65nm and one on 55 nm. You can see the latter has a cleaner PCB, there aren't any memory chips on the back, the power circuitry is simpler, feeding a less hungry gpu. Their coolers are also different, shorter and cheaper on the 55 nm card.

Who knows which exactly is the more expensive card between ATi offerings and Nvidia's, but I'd say that overall, they're at about the same level.

Even with both camps on 55nm, you're still comparing a roughly 1.4B transistor gpu to a ~956M transistor part, all those additional trannies don't come for free.

 

[Munky]

Originally posted by: chizow

Originally posted by: munky
How much more does 1GB GDDR3 cost compared to 512MB GDDR5? Maybe you should have thought of that option before spewing more nonsense.

Rofl, it doesn't cost more, that's the whole point. Case in point, I can get a 1GB 4850 for less than a 512MB 4870 when the only differences are going to be a few ICs and the RAM.

As for spewing nonsense lol. Again, you and others have claimed on numerous occasions to have some clue about board pricing, yet historical market pricing has shown you to be wrong time and time again. The GTX 260 is now priced at $160-170 only 9 months after launch, again, proving Nvidia has more flexibility than many thought despite the massive die size on GT200. It probably has something to do with the fact TSMC's fab's are sitting at ~25% utilization and that glass and metal are cheap lol.

To take the comparison further, a consideration I'm sure you didn't take into account was product diversification with regard to die size and wafer distribution. For example, a 300mm RV770 wafer at 55nm with ~292 dice might be distributed like:

• 30% 4830 @ <$100
  40% 4850 @ <$150
  20% 4870 @ <$200
  10% 4870X2 @ <$225 (prorated per core)

In comparison, a 300mm GT200b wafer at 55nm with ~120 dice might be distributed like:

• 60% GTX 260 @ <$200
  20% GTX 285 @ <$350
  20% GTX 295 @ <$250 (prorated per core)

Without doing the exact calculations (the % obviously aren't going to be accurate although they're pretty reasonable) you can clearly see that even though Nvidia will get fewer dice per wafer, they're also selling the majority of chips for the maximum price ATI is selling the minority of their chips for. In any case, I think this clearly demonstrates simple die size comparisons don't tell the whole story with regards to die prices.

LOL, what does the retail cost of the card have to do with how much it costs the board partners to produce, or how much profit NV makes off each gpu? The prices vary according to supply and demand. Or you think that 9 months ago it cost twice as much to manufacture the cards as it does now?

 

[chizow]

Originally posted by: munky
LOL, what does the retail cost of the card have to do with how much it costs the board partners to produce, or how much profit NV makes off each gpu?

Rofl it has everything to do with market prices as actual board costs are fixed for the most part with minor variable cost differences (like RAM type or ICs) based on SKU. The often huge difference in actual market price translates directly into margins.

The prices vary according to supply and demand. Or you think that 9 months ago it cost twice as much to manufacture the cards as it does now?

No it didn't cost 2x as much, that's exactly my point. People like you didn't know what you were talking about 9 months ago when spewing garbage about GT200's chip pricing, just as you don't know what you're talking about now.

 

[Munky]

Originally posted by: chizow

Originally posted by: munky
LOL, what does the retail cost of the card have to do with how much it costs the board partners to produce, or how much profit NV makes off each gpu?

Rofl it has everything to do with market prices as actual board costs are fixed for the most part with minor variable cost differences (like RAM type or ICs) based on SKU. The often huge difference in actual market price translates directly into margins.

So therefore, unlike 9 months ago, neither NV nor the board partners have such fat profit margins, because the cost to produce the card is not affected by the retail price.

The prices vary according to supply and demand. Or you think that 9 months ago it cost twice as much to manufacture the cards as it does now?

No it didn't cost 2x as much, that's exactly my point. People like you didn't know what you were talking about 9 months ago when spewing garbage about GT200's chip pricing, just as you don't know what you're talking about now.

That's only your opinion, which means jack. Pulling arbitrary numbers from thin air doesn't prove your point either, if you even have one.

 

[chizow]

Originally posted by: munky
So therefore, unlike 9 months ago, neither NV nor the board partners have such fat profit margins, because the cost to produce the card is not affected by the retail price.

Of course profit margins are lower, that much is obvious from all their financials and quarterlies. However, high margins aren't the only path to profitability, higher volume sales on lower margins get there as well. I think in this case its the path both AMD and Nvida are actually hoping to take to reach profitability although its unlikely either will get there.

That's only your opinion, which means jack. Pulling arbitrary numbers from thin air doesn't prove your point either, if you even have one.

No you've already confirmed my opinion that the numbers you and others pulled out of thin air 9 months ago, 18 months ago, 38 months ago were all bogus and based on nothing. Just as you and others claimed Nvidia couldn't drop prices on GT200 from $450 to $300, $300 to $250, $250 to $200, $200 to $170 etc etc.

 

[WelshBloke]

Originally posted by: chizow

Originally posted by: munky
How much more does 1GB GDDR3 cost compared to 512MB GDDR5? Maybe you should have thought of that option before spewing more nonsense.

Rofl, it doesn't cost more, that's the whole point. Case in point, I can get a 1GB 4850 for less than a 512MB 4870 when the only differences are going to be a few ICs and the RAM.

As for spewing nonsense lol. Again, you and others have claimed on numerous occasions to have some clue about board pricing, yet historical market pricing has shown you to be wrong time and time again. The GTX 260 is now priced at $160-170 only 9 months after launch, again, proving Nvidia has more flexibility than many thought despite the massive die size on GT200. It probably has something to do with the fact TSMC's fab's are sitting at ~25% utilization and that glass and metal are cheap lol.

To take the comparison further, a consideration I'm sure you didn't take into account was product diversification with regard to die size and wafer distribution. For example, a 300mm RV770 wafer at 55nm with ~292 dice might be distributed like:

• 30% 4830 @ <$100
  40% 4850 @ <$150
  20% 4870 @ <$200
  10% 4870X2 @ <$225 (prorated per core)

In comparison, a 300mm GT200b wafer at 55nm with ~120 dice might be distributed like:

• 60% GTX 260 @ <$200
  20% GTX 285 @ <$350
  20% GTX 295 @ <$250 (prorated per core)

Without doing the exact calculations (the % obviously aren't going to be accurate although they're pretty reasonable) you can clearly see that even though Nvidia will get fewer dice per wafer, they're also selling the majority of chips for the maximum price ATI is selling the minority of their chips for. In any case, I think this clearly demonstrates simple die size comparisons don't tell the whole story with regards to die prices.

Apart from the fact that you pulled all those numbers from your arse. :roll:

 

[Munky]

Originally posted by: chizow

Originally posted by: munky
So therefore, unlike 9 months ago, neither NV nor the board partners have such fat profit margins, because the cost to produce the card is not affected by the retail price.

Of course profit margins are lower, that much is obvious from all their financials and quarterlies. However, high margins aren't the only path to profitability, higher volume sales on lower margins get there as well. I think in this case its the path both AMD and Nvida are actually hoping to take to reach profitability although its unlikely either will get there.

That's only your opinion, which means jack. Pulling arbitrary numbers from thin air doesn't prove your point either, if you even have one.

No you've already confirmed my opinion that the numbers you and others pulled out of thin air 9 months ago, 18 months ago, 38 months ago were all bogus and based on nothing. Just as you and others claimed Nvidia couldn't drop prices on GT200 from $450 to $300, $300 to $250, $250 to $200, $200 to $170 etc etc.

If those prices can't maintain profitability, then it doesn't matter how low they drop them.

 

[chizow]

Originally posted by: WelshBloke
Apart from the fact that you pulled all those numbers from your arse. :roll:

No, not really. The only numbers I guessed on were the % per SKU, but I still based that off of distributions I've seen in various places like Steam survey or Yougamers/3DMark. Market prices are what they are and directly translate into margins. Number of dice per 300mm wafer are quoted from various websites presumably based on die size and verified by counting dice on press pictures of complete wafers.

Originally posted by: munky
If those prices can't maintain profitability, then it doesn't matter how low they drop them.

Now we're getting somewhere! That's the whole point, the fact they are still dropping prices proves the speculation with regard to GT200's chip prices was clearly bogus and overstated. It clearly shows the part is maintaining profitability even at $160 to 170, as it wouldn't make any sense to sell these parts at a loss if the goal is to be profitable overall as a company.

 

[BFG10K]

Originally posted by: chizow

as it wouldn't make any sense to sell these parts at a loss if the goal is to be profitable overall as a company.

What if the goal is to stop ATi gaining market share by undercutting their prices?

 

[exar333]

Nvidia makes money, AMD loses money...enough said about wafer size and profit margins.

 

[WelshBloke]

Originally posted by: chizow

Originally posted by: WelshBloke
Apart from the fact that you pulled all those numbers from your arse. :roll:

No, not really. The only numbers I guessed on were the % per SKU, but I still based that off of distributions I've seen in various places like Steam survey or Yougamers/3DMark. Market prices are what they are and directly translate into margins. Number of dice per 300mm wafer are quoted from various websites presumably based on die size and verified by counting dice on press pictures of complete wafers.

Your making deductions on die prices from the steam survey?

And WTF does 'Market prices are what they are and directly translate into margins' mean?

The only thing youve got is the number of dice per wafer.

 

[LOUISSSSS]

Originally posted by: ExarKun333
Nvidia makes money, AMD loses money...enough said about wafer size and profit margins.

actually, what i heard was that AMD makes more money/chip sold than nvidia...

 

[Lonyo]

Originally posted by: ExarKun333
Nvidia makes money, AMD loses money...enough said about wafer size and profit margins.

Q2 FY09 (May - July 2008) NVDA Net loss $120mn (most recent thing I found on the NV website)
Q2 FY09 (May - July 2008) AMD Net loss from graphics $38mn. (Direct comparison)
Q3 AMD $47mn profit, Q4 $10mn loss. (Subsequent quarters)

Let's talk more.

 

[chizow]

Originally posted by: BFG10K
What if the goal is to stop ATi gaining market share by undercutting their prices?

Market share might be a consideration for sacrificing margins, up until the point you're taking losses. Unless a part is going EOL it never makes sense to take a loss in the hopes of increasing market share as each % gained translates into a loss that further erodes profits and decreases margins.

Its never good business practice or a sustainable business model to sell current SKUs in volume at a loss, as decreasing price and increasing market share only translates into bigger losses. Some recent examples are the US automakers that can't turn a profit on their cars. Market share means very little to them at this point.

In this case however, its highly unlikely that Nvidia would need to undercut AMD, as they're in a position of strength and actually regained the market share in Q4 08 according to the latest JPR info that they lost in Q3, the height of RV770's popularity. It should also be noted that Nvidia posted a profit in Q3 08 despite this reduction in market share and price cuts on its high-end GT200 chips.

Throw in some blurbs from Fudzilla that confirm Nvidia's happy right where they are and have no need to initiate price cuts:
January was good for Nvidia partners
GTX 260, 285 outselling HD 48x0
Nvidia doesn?t plan a price war

 

[thilanliyan]

Originally posted by: chizow
Nvidia doesn?t plan a price war

Isn't that what they're doing right now though...since the 260 was again reduced in price?

 

[chizow]

Originally posted by: WelshBloke
Your making deductions on die prices from the steam survey?

And WTF does 'Market prices are what they are and directly translate into margins' mean?

The only thing youve got is the number of dice per wafer.

No, I'm making deductions about SKU distribution based on sales and market share, which translates directly into a % of die per wafer used for those respective parts. If you sell two 4850s for every 4870, you'd expect ~66% of the dice in a wafer to go toward 4850s to meet demand regardless of whether or not they could be validated and sold as 4870s.

Based on that point, that RV770 dice are more or less equal on any wafer, market prices are what you see and what a part is sold for, meaning your margin per chip is directly determined by the market price of whatever SKU that RV770 is sold as. This means that AMD would absolutely love to sell every RV770 as a 4870 for 2x as much as a 4830 as it has much higher margins, but the demand is simply not there.

Again, I think I've clearly proven simple die size comparisons simply aren't accurate when you look at SKU distribution and overall pricing.

 

[chizow]

Originally posted by: thilan29

Originally posted by: chizow
Nvidia doesn?t plan a price war

Isn't that what they're doing right now though...since the 260 was again reduced in price?

No, sounds like it was more reactionary to the news in this OP about 4870 price cuts that didn't happen. Again, it'll be interesting to see if Nvidia raises prices again to match the 1GB 4870, although rumor has it the GTX 275 is going to move into that $230 price range.