AMD To Commence Mass Shipping of Hawaii GPU in Early October

[JediFragger]

I'm expecting 25% on average over the 7970. So peeps, get clocking the balls off those 7970s and you won't be too far off

 

[Kaapstad]

Yes it did with the Fermi. Nvidia paid for a fixed number of chips and per chip,not per wafer. The usual agreement with AMD and Nvidia is that they pay for whole wafers not the individual chips from the wafer. They have to take the hit for poor yields.

However,Nvidia managed to change this for Fermi by blaming their yield issues on TSMC,who gave into them for some reason.

It meant that TSMC ended up taking most of the cost of the poor Fermi yields not Nvidia. It also meant Nvidia could get away with making a huge, poor yielding GPU on the TSMC 40NM process,without the risks of having to pay for a massive amount of wafers themselves.

However,with the 28NM process they had to go back to paying per wafer,and the wafers were more expensive than 40NM ones,so hence you can see why the GK110 based cards cost so much when compared to the GF100 and GF110 based cards,and they were introduced quite late compared to the midrange chips. Nvidia has to pay for any poor yields this time around.

You are going all round the houses to say what I said in the first place.

bigger chips equal lower yields equals more expensive chips.

 

[CAT-THE-FIFTH]

why expensive ?

the GTX470 had a die size of over 500mm2

Bigger chips, lower yields

Nvidia only paid for functioning GPUs with the GTX400 series,not whole wafers.

It does not matter how NVidia pay for them, bigger chips equal lower yields equals more expensive chips.

Yes it did with the Fermi. Nvidia paid for a fixed number of chips and per chip,not per wafer. The usual agreement with AMD and Nvidia is that they pay for whole wafers not the individual chips from the wafer. They have to take the hit for poor yields.

However,Nvidia managed to change this for Fermi by blaming their yield issues on TSMC,who gave into them for some reason.

It meant that TSMC ended up taking most of the cost of the poor Fermi yields not Nvidia. It also meant Nvidia could get away with making a huge, poor yielding GPU on the TSMC 40NM process,without the risks of having to pay for a massive amount of wafers themselves.

However,with the 28NM process they had to go back to paying per wafer,and the wafers were more expensive than 40NM ones,so hence you can see why the GK110 based cards cost so much when compared to the GF100 and GF110 based cards,and they were introduced quite late compared to the midrange chips. Nvidia has to pay for any poor yields this time around.

You are going all round the houses to say what I said in the first place.

The problem is what you are saying is not always true as indicated by the example of Fermi,which Buckster was talking about.

Bigger chips don't necessarily cost more based on yields,as it dependent on what agreement there is between the company making the chip and the company buying it.

You either buy the finished chips or you buy the whole wafers and finish the chips yourself.

Yields were not a finanical issue for Nvidia with Fermi.It only makes a massive impact if you are paying per wafer which Nvidia was not at the time.

It makes an impact now since both AMD and Nvidia buy the whole wafers themselves.

If you pay per wafer and have poor yields,you have to buy more wafers to get your targeted number of chips. Nvidia didn't need to do that since they paid a fixed price per chip,which meant the GF100 and GF110 could have had eff all yields,but Nvidia got the number of chips they wanted without having to pay the price for their poorer yielding larger chips. TSMC took the financial hit not Nvidia.

This is why Nvidia could sell the GTX460 at prices lower than an HD5850 and still make decent profits. The GF104 and the Cypress GPU were the same size.

The GF114 in the GTX560 were much larger than the Barts GPU in the HD6870 and yet was closely priced. The GTX460 was closely priced to the HD6850 which used a smaller GPU.

The vast majority of the Fermi chips were larger than their comparable AMD equivalents, and required more expensive PCBs and coolers too due to higher power and cooling requirements. Nvidia still made decent profits.

 

[Kaapstad]

The problem is what you are saying is not always true as indicated by the example of Fermi,which Buckster was talking about.

Bigger chips don't necessarily cost more,as it dependent on what agreement there is between the company making the chip and the company buying it.

You either buy the finished chips or you buy the whole wafers and finish the chips yourself.

Yields were not a finanical issue for Nvidia with Fermi.It only makes a massive impact if you are paying per wafer which Nvidia was not at the time.

If you pay per wafer and have poor yields,you have to buy more wafers to get your targeted number of chips. Nvidia didn't need to do that since they paid a fixed price per chip,which meant the GF100 and GF110 could have had eff all yields,but Nvidia got the numbers of chips without having to pay the finanical price for their poorer yielding larger chips. TSMC took the financial hit not Nvidia.

This is why Nvidia could sell the GTX460 at prices lower than an HD5850 and still make decent profits. The GF104 and the Cypress GPU were the same size.

The GF114 in the GTX560 were much larger than the Barts GPU in the HD6870 and yet was closely priced. The GTX460 was closely priced to the HD6850 which used a smaller GPU.

The vast majority of the Fermi chips were larger than their comparable AMD equivalents, and required more expensive PCBs and coolers too due to higher power and cooling requirements.

Still going round the houses

Bigger chips, lower yields.

 

[CAT-THE-FIFTH]

why expensive ?

the GTX470 had a die size of over 500mm2

Bigger chips, lower yields

Nvidia only paid for functioning GPUs with the GTX400 series,not whole wafers.

It does not matter how NVidia pay for them, bigger chips equal lower yields equals more expensive chips.

Yes it did with the Fermi. Nvidia paid for a fixed number of chips and per chip,not per wafer. The usual agreement with AMD and Nvidia is that they pay for whole wafers not the individual chips from the wafer. They have to take the hit for poor yields.

However,Nvidia managed to change this for Fermi by blaming their yield issues on TSMC,who gave into them for some reason.

It meant that TSMC ended up taking most of the cost of the poor Fermi yields not Nvidia. It also meant Nvidia could get away with making a huge, poor yielding GPU on the TSMC 40NM process,without the risks of having to pay for a massive amount of wafers themselves.

However,with the 28NM process they had to go back to paying per wafer,and the wafers were more expensive than 40NM ones,so hence you can see why the GK110 based cards cost so much when compared to the GF100 and GF110 based cards,and they were introduced quite late compared to the midrange chips. Nvidia has to pay for any poor yields this time around.

You are going all round the houses to say what I said in the first place.

The problem is what you are saying is not always true as indicated by the example of Fermi,which Buckster was talking about.

Bigger chips don't necessarily cost more based on yields,as it dependent on what agreement there is between the company making the chip and the company buying it.

You either buy the finished chips or you buy the whole wafers and finish the chips yourself.

Yields were not a finanical issue for Nvidia with Fermi.It only makes a massive impact if you are paying per wafer which Nvidia was not at the time.

It makes an impact now since both AMD and Nvidia buy the whole wafers themselves.

If you pay per wafer and have poor yields,you have to buy more wafers to get your targeted number of chips. Nvidia didn't need to do that since they paid a fixed price per chip,which meant the GF100 and GF110 could have had eff all yields,but Nvidia got the number of chips they wanted without having to pay the price for their poorer yielding larger chips. TSMC took the financial hit not Nvidia.

This is why Nvidia could sell the GTX460 at prices lower than an HD5850 and still make decent profits. The GF104 and the Cypress GPU were the same size.

The GF114 in the GTX560 were much larger than the Barts GPU in the HD6870 and yet was closely priced. The GTX460 was closely priced to the HD6850 which used a smaller GPU.

The vast majority of the Fermi chips were larger than their comparable AMD equivalents, and required more expensive PCBs and coolers too due to higher power and cooling requirements. Nvidia still made decent profits.

Still going round the houses

Bigger chips, lower yields.

Still not bothering to actually read what I said. Thats your problem. Nvidia obviously did know the financial side well enough at the time,to get themselves out of a sticky situation and that is why they did relatively well even with the problems they had with Fermi.

 

[Kaapstad]

Still not bothering to actually read what I said. Thats your problem. Nvidia obviously did know the financial side well enough at the time,to get themselves out of a sticky situation and that is why they did relatively well even with the problems they had with Fermi.

You are still going round the houses

And I did read what you wrote

Bigger chips, lower yields

This has got nothing to do with an NVidia pricing deal

 

[ICDP]

I would be surprised.

AMD haven't been in the flagship camp for a long time. They always target volume value.

AMD had the flagship card for most of 2012 with the 7970 and 7970 GE.

 

[Kaapstad]

AMD had the flagship card for most of 2012 with the 7970 and 7970 GE.

+1

If you count dual GPU cards as well AMD have done well for most of the DirectX 11 period with the HD 5970, HD 6990 and HD 7990.