Nvidia's Future GTX 580 Graphics Card Gets Pictured (Rumours)

[Arkadrel]

368mm^2 Nvidia chip vs 255mm^2 Amd (460 vs 6850) (44% bigger chip for same price range)
529mm^2 Nvidia chip vs 255mm^2 Amd (470 vs 6870) (107% bigger chip for same price range)

+ read somewhere that GDDR3 costs more to manufacture than the GDDR5.

= not how to make profits.

You might be right RussianSensation, maybe its for wall streets benefit.

 

[Genx87]

We already went over this about a month ago that a larger chip doesnt always translate into lower margins.

 

[Scali]

We already went over this about a month ago that a larger chip doesnt always translate into lower margins.

A month ago? I think this argument has been going on ever since the GeForce 8000/9000 vs Radeon 3000 series. Guess what? nVidia is still around, and still doing okay, despite the recent crisis and the bumpgate-affair.

 

[SickBeast]

A month ago? I think this argument has been going on ever since the GeForce 8000/9000 vs Radeon 3000 series. Guess what? nVidia is still around, and still doing okay, despite the recent crisis and the bumpgate-affair.

NV apparently is about to release some terrible financial information, which is why people are speculating that they are paper launching the GTX580.

A larger die costs more to manufacture, particularly when you look at the magnitude of the discrepancy between current NV and AMD GPUs. You can say that NV gets a volume discount if you want, but really, they're not going to get enough of a discount to negate their mammoth die sizes.

 

[Arkadrel]

@Genx87
If your useing more expensive ram, useing more die size/same price pr waffer as amd for the same price range in product, and in the case of the 470 and 480 useing more MB of ram than your competitors. Along with the fact that bigger dies = worst yeilds than competitors (more chips thrown out the window).

to me all that sums up as = not good for profits.

 

[busydude]

+ read somewhere that GDDR3 costs more to manufacture than the GDDR5.

I think both use GDDR5.

 

[RussianSensation]

A month ago? I think this argument has been going on ever since the GeForce 8000/9000 vs Radeon 3000 series. Guess what? nVidia is still around, and still doing okay, despite the recent crisis and the bumpgate-affair.

Correct. The difference is G80/G92/GT200 sold very well. The higher sales volume of the more expensive to manufacture chips most likely compensated. With GF100/104, this isn't the case. NV was 6 months late to the game. As a result, they have had much less time to recoup the higher manufacturing costs. GTX460/470/480 haven't exactly sold by the 10s of millions. Although, their Q3 desktop results do look promising.

 

[Scali]

Correct. The difference is G80/G92/GT200 sold very well. The higher sales volume of the more expensive to manufacture chips most likely compensated. With GF100/104, this isn't the case. NV was 6 months late to the game. As a result, they have had much less time to recoup the higher manufacturing costs. GTX460/470/480 haven't exactly sold by the 10s of millions. Although, their Q3 desktop results do look promising.

Yea well, then you're ignoring the fact that nVidia sells a lot more than just the 400-series.
The DX11 market is still a pretty small slice of the total sales of graphics chips... and then there's things like Ion and Tegra.

 

[Genx87]

@Genx87
If your useing more expensive ram, useing more die size/same price pr waffer as amd for the same price range in product, and in the case of the 470 and 480 useing more MB of ram than your competitors. Along with the fact that bigger dies = worst yeilds than competitors (more chips thrown out the window).

to me all that sums up as = not good for profits.

You are assuming equal yields. If Nvidia gets 80% of their wafer using a large die that disabled non-functional units for harvesting vs AMDs 50% using a smaller die that cant harvest as well. Which one do you think will have a better cost of goods sold?

There is more to the equation than simply die sizes.

 

[Genx87]

Correct. The difference is G80/G92/GT200 sold very well. The higher sales volume of the more expensive to manufacture chips most likely compensated. With GF100/104, this isn't the case. NV was 6 months late to the game. As a result, they have had much less time to recoup the higher manufacturing costs. GTX460/470/480 haven't exactly sold by the 10s of millions. Although, their Q3 desktop results do look promising.

Do you have any links to show it isnt selling well? The 460 has a nice % on Steam as a single card.

I honestly suspect they recoup those costs with their Tesla and Quadro lines.

 

[SlowSpyder]

Lets wait until the products are out in the wild. This thing looks like it is about 40-45% faster than the 5870. How much faster are we expecting the top end card from AMD to be compared to the 5870? The mid-series cards were underwhelming imo. Basically a side grade from the 5850. Considering this is on the same process as the 480 and of the same arch. This is a pretty good step up in performance from the 480 if the performance holds true.

How is a mid range product that very nearly almost equals the former highest end single GPU, while doing it with about 33% less silicon and on the same 40nm process, has more tessellation performance, uses less power than the 5870 (which already had decent power use) underwhelming?

Yet a new high end part from Nvidia that appears to be little more than a tweaked GF100 with another spin, that might be 20% faster and we presume less power, is a 'pretty good step up' because it is on the same process?

 

[Ares1214]

You are assuming equal yields. If Nvidia gets 80% of their wafer using a large die that disabled non-functional units for harvesting vs AMDs 50% using a smaller die that cant harvest as well. Which one do you think will have a better cost of goods sold?

There is more to the equation than simply die sizes.

While we cant know for sure, larger die generally=lower yields. But its always worse to have a far bigger die, we just dont know how much of an effect its having, if any.

 

[AnandThenMan]

Even if yields are 100%, a larger die means higher cost, you get fewer parts per wafer. And you almost always get higher yields with a smaller die, unless your design has some serious flaws.

 

[Paratus]

You are assuming equal yields. If Nvidia gets 80% of their wafer using a large die that disabled non-functional units for harvesting vs AMDs 50% using a smaller die that cant harvest as well. Which one do you think will have a better cost of goods sold?

There is more to the equation than simply die sizes.

Yup. People don't understand that since NV has like 2X the marketshare of AMD they get a Much better deal from TSMC.

Example: Say AMD's small die strategy means they can get 150 dies from a wafer and TSMC charges AMD $5000.

Now since NV's dies are larger and since they own like almost all the market they need a ton more wafers than AMD. So if NV can only get 100 dies per waffer then I'm sure TSMC only charges them $2000 per wafer

In fact I bet the larger the die NV has the cheaper it is for NV because nothing is more important to TSMC than NV's profits!

I'm sure it works this way for the memory manufactures and board partners as well!

 

[AnandThenMan]

Yup. People don't understand that since NV has like 2X the marketshare of AMD they get a Much better deal from TSMC.

I've seen this repeated many times. Have anything concrete to back that up besides your guesses?

 

[Paratus]

I've seen this repeated many times. Have anything concrete to back that up besides your guesses?

I can't tell you where I "pulled it from". It's a ssecret.

 

[Arkadrel]

@Paratus

I doubt TSMC shows that much favor to nvidia... I think they get same price for waffers.

I dont see that happending, TSMC would want to keep both happy and probably sell both the same waffers at same price (I dont think they care about who buys them, just that they make good profits, selling to nvidia for under half price would suck for their own profits). Something like you said could "scare" one or the other away, and twist competiton and you might see like FCC step in ect (wouldnt they?).

also what twice market share? Amd is selling more current new cards total than nvidia.

 

[Paratus]

@Paratus

I doubt TSMC shows that much favor to nvidia... I think they get same price for waffers.

I dont see that happending, TSMC would want to keep both happy and probably sell both the same waffers at same price (I dont think they care about who buys them, just that they make good profits, selling to nvidia for under half price would suck for their own profits). Something like you said could "scare" one or the other away, and twist competiton and you might see like FCC step in ect (wouldnt they?).

also what twice market share? Amd is selling more current new cards total than nvidia.

[Taps Akadrels Sarcasm Meter]

This thing on?

 

[Arkadrel]

lol apperntly not <.< didnt get you where being sarcastic. ;_; (its late here, need more coffee)

forgiv mez Paratus? m(_ _)m

Paratus about that market share thingy:
http://www.theinquirer.net/inquirer/news/1866086/amd-dominates-discrete-graphics-market

"..the PC graphics market share update, tells us that AMD is taking large strides in the market and added some 5.6 percentage points to its market share in the last quarter alone. This, Mercury added, has taken it up to a 61.9 per cent share of the market."

 

[Idontcare]

I've seen this repeated many times. Have anything concrete to back that up besides your guesses?

I've worked with TSMC, as a foundry customer and co-developer for process technology, and yes they do give pricing discounts based on wafer volumes.

They also give pricing discounts based on other business relationship projects.

Pick your industry, its the same wherever you go. But I can confirm from first-hand experience it is the same TSMC.

edit: if you search for my posts from a year ago when Cypress release was imminent and then thereafter you'll find scores of posts discussing pricing, yields, etc.

 

[RussianSensation]

Do you have any links to show it isnt selling well? The 460 has a nice &#37; on Steam as a single card.

You are misreading what I am saying. I never said Fermi is not selling well right now. I said they have not sold a lot of chips in the grand scheme (i.e., 20-30 million 332/529mm2 chips) since NV was a no show for 7+ months basically. On the contrary, G80/92/GT200 chips sold for a long time (2+ years). Even if NV is selling on par with AMD now, they are almost 3 quarters behind in sales #s. Therefore, by the very nature of NV being late with no cards to show at any reasonable price points until July 2010, NV has not had a long time to recoup the costs of expensive large dies (at least not in the desktop consumer space). Also, steam #s mean nothing except their relation to Steam users.

Here are some glaring standouts from the "reliable" Steam survey in how it represents real world market trends.

"increased" market share for GTX260 every month from July to October 2010
"increased" market share for GTX275 every month from August to October 2010
"increased" market share for GTX285 every month from August to October 2010
"increased" market share for GTX295 every month from August to October 2010

Please, let's not bring "scientific" data of Steam into this.

Yea well, then you're ignoring the fact that nVidia sells a lot more than just the 400-series.

Yes, it might be good business to force your AIBs to sell 4 variants of GT200 chips with every single 1 Fermi chip, but use this logic when Fermi II comes out. How do you expect Fermi I large die size chips are going to fulfill the role of G80/92/GT200 style chips? The cost of selling the chip isn't just manufacturing cost, there are R&D costs as well. So Fermi itself may be profitable per chip, but unprofitable on the whole unless they sell them for a long time to make a profit vs. R&D costs.

Think of this as a car company investing $3B into a new platform. Every single car sold may indeed be profitable, but on the whole the product line will remain unprofitable until the point that profit margin per car x # of total cars sold > initial investment costs. This is exactly how Fermi is like (massive initial investment --> recouping costs over time). It just takes a lot longer to do this if your chip costs more to manufacture; and is a lot more difficult when your competitor forces you to drop prices all the time.

 

[AnandThenMan]

I've worked with TSMC, as a foundry customer and co-developer for process technology, and yes they do give pricing discounts based on wafer volumes.

Thanks for the info. But the statement was:

Yup. People don't understand that since NV has like 2X the marketshare of AMD they get a Much better deal from TSMC.

BTW, I recall JHH saying that Nvidia does pay for poor yields (he didn't reveal much granularity however).

 

[Paratus]

lol apperntly not <.< didnt get you where being sarcastic. ;_; (its late here, need more coffee)

forgiv mez Paratus? m(_ _)m

Paratus about that market share thingy:
"..the PC graphics market share... oublicized cash reserves. IMHO of course.

 

[Silverforce11]

IMO, If they are launching first, then they likely know their GPU won't be competitive. If they launch at the same time, then its the reverse. The most dominant graphics AIB supplier to SE Asia/Australia (AtomicPC) are Gigabyte & Asus. If the article is right, these guys not having stock or info on the gtx580 is telling.

 

[Teizo]

IMO, If they are launching first, then they likely know their GPU won't be competitive. If they launch at the same time, then its the reverse. The most dominant graphics AIB supplier to SE Asia/Australia (AtomicPC) are Gigabyte & Asus. If the article is right, these guys not having stock or info on the gtx580 is telling.

Or, it's just a paper launch next week..not a hard launch.