Officially 10 Years since the ATI AMD merger. Was it a good idea?

[bfar]

Not sure if we are equipped with the knowledge to say it was badly implemented. Bit like if Brexit goes wrong, it'll be so easy to blame those that walked us through the pain of it (conservatives,Mrs May, the negotiators) without really knowing, than it will be to blame ourselves for taking us down the path . Slightly off-topic

It's been claimed by former employees that AMD struggled to integrate the ATI staff, and that employees continued to see themselves as either team AMD or team ATI for too long after the merger was completed. The merger happened 10 years ago, so we've plenty information to make a judgement call.

 

[Rroff]

Salvaging the salvaged because it does not make the salvaged grade?

The 1070 is already a salvaged 1080, IE the 1070 is a chip not making the grade for the 1080, the 1070 is a salvaged chip.

Its an easy argument to make, the fact is the GTX 770 was not made from the salvaged 550mm^2 780TI, that would be the 780.

The 432mm^2 290X was not the 280X, salvaged that was the 290P.

What you are saying was never actually done by anyone because they are too big and too expensive to make into GPU's like that.

Um that is basically how its been done since GPUs existed - look at some variants of the GTX560ti, GTX465, etc. in some older generations almost the entire mid and high end were trickle down salvaged parts.

 

[humbug]

None of them are from the biggest tho Roff ^^^^

Actually AMD only started really having issues,when Nvidia managed to cut down die sizes and costs with Maxwell. Nvidia managed to sell smaller dies with less RAM chips for more money and AMD was selling larger dies with more RAM chips for less money.

ATI and indeed some of the earlier AMD cards,did well since ATI and AMD managed to produce small reasonably well performing chips which meant they could maintain margins even at lower prices.

Its the same with CPUs - they are stuck selling much larger chips for a fraction of what Intel would charge.

This is why I hope any £300 to £400 Vega chip is NOT using HBM2,since the cost of packaging,etc is probably quite high until it becomes a common way of doing things for graphics cards.

If Vega ends up being some 400MM2 chip with HBM2 they are only charging £400 for,then AMD is not going to increase margins. Nvidia has massive margins which have doubled since the Fermi times.

Right.

They are actually doing that to themselves now, the GTX 1060 has less shaders than the GTX 970 (1664 vs 1280)

The difference is typically with GPU Boost 2 a reference GTX 970 runs at about <1250Mhz.
A reference GTX 1060 with GPU Boost 3 about 1900Mhz.

So the result is out of the box the 1060 is more like a GTX 980 in performance than a GTX 970, and yet because the 970 has more overclocking head room they are actually roughly about the same performance.

As for the GTX 980......

 

[FuryXGuy]

The 1070 is already a cut down 1080, just arbitrarily saying Nvidia can cut it down even more to decimate the RX 480 shows your not thinking about what you are saying.

The 1080 and there fore the 1070 is 40% larger than the RX480, so no, Nvidia like everyone else cannot take large dies and turn them into much lower performance products for half the money.

As for AMD selling £500 GPU's they cost a lot of money to make, they cost a lot of wafer space, they cost a lot of PCB components, they cost a lot of expensive memory chips.

GPU's like that are only viable if you can actually sell them at £500 and in high quantities.

Which AMD have never been able to do, not with the 6970, maybe with the 7970 eventually, not with the 290X, not with the Fury-X.

Its actually AMD's lower and mid range cards that paid for the losses and write off's of AMD's enthusiast range.

The RX 480 has been a fantastic card for AMD. Vega at 1080 performance sub £400 will also be, no need for a card bigger than that to compete with what will be a 1080TI as AMD cannot sell them.

If any reputation is going to be made of it its that Nvidia love to milk their user base charging £700 for the 1080TI.

Sure you are not going to sell as much in the enthusiasts range. But in the amount you sell the profit will be much larger. And the mid-range has always fueled most of the revenues and profits. This is no secret. 85% of the market buys cards in the $200 - $300 range.

My point was not focusing on the high end leaves them vulnerable to being faced with a cut down version of higher end cards which allows the said company to do what ever they want at will, even going as far as taking losses to hurt the competition.

This is what Saudi Arabia and OPEC countries were trying to do in the last two years when they increased output of oil production to ridiculous levels thus flooding the market with oil and thus reducing the price of oil in an attempt to put the US oil shale producers out of the market. Saudi Arabia can do that because they have large market share and have large cash reserves (kinda of like nVidia) that they can afford to take losses or less profit in order to put their competitors out of business (which has actually happened as many US Shale oil producers went out of market because they can longer be business with the low oil prices and others are on the brink of bankruptcy). After their competitors are out of business then Saudi Arabia and the large OPEC producers can cut production and reduce the number of barrels of oil they output in the market and get back to making more profits as they would have destroyed many of their competitors.

Having your competitor having superior technology puts you at risk to their will. Besides AMD was pretty competitive when they had competitive high end cards. They were making money in all those years the cards cards you stated came out in. I know plenty of people who brought 7970/7950 or R9 290/290X's. It also leaves people with some 'mind share' that AMD doesn't produce just cheap products. Kind of like people view BMW vs Audi. AMD needs to have top to bottom cards in every market segment in order to be less prone to risks of falling behind and to a certain extent have fat margins (which the high end cards helps in).

Let's put it this way if AMD hadn't released the R9 290X and if it wasn't that competitive then nVidia wouldn't have had to slash the price of the GTX 780 nearly by $150 to $500 after the R9 290X was released. It effectively put the original Titan useless as the R9 290X was $450 cheaper than the $1000 Titan and it was as good if not better than it.

Problem with the Fury X was that they priced it at the same price as the 980 Ti and it was behind in most benchmarks especially at 1080P. Only in 1440P did it match the 980 Ti, and that was after the driver updates several months later and in 4K did it beat the 980 Ti (but the problem was most people don't have 4K). So most people focused on the faster card at release which as the 980 Ti. Not to mention it had 2GB more, I know one person who brought a 980 Ti mention this.

This was completely the opposite of the R9 290X which beat the 780 consistently while costing $100 less. It was even matching and beating the original Titan. Same thing with the HD 7970 which was beating GTX 580 in nearly every single benchmark by double digit %.

If the Fury X was beating the 980 Ti by double digit % and costs $550 at launch like the R9 290X or the 7970 then AMD would have had a hit in their hands and many people would have brought the Fury X over the 980 Ti. Thus nVidia wouldn't be raking in the cash they have in the last year with respect to profit margins.

 

[FuryXGuy]

You are taking what he is saying a bit too literally - strategically having a top down product line gives you more options for instance just as an example you have a range of possibilities to use salvaged 1070 cores that are still functional but not making the grade for the 1070 including using them sacrificially to disadvantage the competition.

Exactly! Not to mention being on the cutting edge of technology which can be of use on any instance and can be deployed any time to harm the competition who would not have any answer for it.

What's to say that nVidia builds on their superior high end product in their next generation and comes up with something even better in the next gen targeted in for the mid-range or low end market. You get a lot of technical "know how" when you push technology and produce products for the high end.

"Know how" matters in the technology market. Just ask Blackberry, they got screwed by Apple's and Samsung;s superior phones which are technologically superior.

What was once a company that had the best tech in phones got owned by companies who made better and faster products than them with superior tech (also making poor business decisions didn't help).

 

[humbug]

The 290X was not that bad as it was a much smaller chip than the 780TI is was competing against. About 430mm^2 vs 550mm^2 and about the same performance, so AMD could afford to undercut the 780TI as it was more expensive for Nvidia to make.

Turn that on its head, could AMD afford to sell 550mm^2 at 80% the price of Nvidia's 430mm^2, and with 16 vs 12 Memory IC's?

Unless AMD can find a way to make chips that would be faster than the 1080TI and smaller the chances are AMD would lose money on them because Nvidia can afford to cut the price of their smaller chips to under what AMD can extract profits from.

AMD do not have the brand clout to sell £500 cards if Nvidia want to deny them sales at that level.

They do at the £150 to £250 range.

Yes Nvidia could sell GTX 1070 dies for £250 to kill the RX 480, but they would make a loss on every-single one they sold. AMD would still sell some while Nvidia are burning money and using £500 1080 dies for £250 chips to try and deny AMD sales there.

 

[CAT-THE-FIFTH]

None of them are from the biggest tho Roff ^^^^



Right.

They are actually doing that to themselves now, the GTX 1060 has less shaders than the GTX 970 (1664 vs 1280)

The difference is typically with GPU Boost 2 a reference GTX 970 runs at about <1250Mhz.
A reference GTX 1060 with GPU Boost 3 about 1900Mhz.

So the result is out of the box the 1060 is more like a GTX 980 in performance than a GTX 970, and yet because the 970 has more overclocking head room they are actually roughly about the same performance.

As for the GTX 980......

But you can see the issues as time progressed. Up until GCN,AMD tended to have smaller die sizes for the HD4000,HD5000 and HD6000 series.

At the high end,the Tahiti chips were larger than the GK104 chips. However,for the rest of the range,it was more comparable to Kepler.

The problem is AMD tried to hard to get Nvidia higher end sales since Hawaii,and basically released sub-par midrange GPUs.

This is the problem - they just rebadged older and older chips,and the few updates they made didn't improve performance or power consumption that much and didn't lower costs.

If they kept their eye on the midrange and released more polished cards,they would not have lost marketshare as quickly as they did.

They also did the same with mobile and lost massive amounts of sales.

Going after the high-end at all costs didn't help them IMHO.

The old ATI and AMD generally made sure their mid-range stack could hold their own with Nvidia in all metrics,bar the HD2000 series disaster.

But you saw what they did with the HD3000 series - they dropped costs and reduced power consumption and were still able to salvage that disaster.

Even the HD2000 and HD3000 series disasters,never dropped AMD sales marketshare and perception as bad as during 2014 and 2015.

Then you had 600MM2 Fiji GPU based cards with HBM selling for the same as a GTX980 using a 398MM2 GPU and common GDDR5.

Its the same with their CPUs - they are selling huge chips for little money.

Nvidia margins have double since Fermi with Maxwell and Pascal.

Polaris 10 and 11 is a step in the right direction and they need to start thinking less about innovative tech and more about cost effective tech.

 

[FuryXGuy]

It's been claimed by former employees that AMD struggled to integrate the ATI staff, and that employees continued to see themselves as either team AMD or team ATI for too long after the merger was completed. The merger happened 10 years ago, so we've plenty information to make a judgement call.

This is what a former AMD GPU architect and one of the lead architechts of the 4800 series and the 5800 series had to say about ATI employees after being brought on by AMD, he was also the inventor of AMD's Eyefinity technology. Basically, they were competing head to head with nVidia in therms of competitivenes, and then CEO Rory Read came in 2011 and he laid off many graphics talent including himself. After than a lot of AMD top rated graphics talent left 2012+, which eventually hurt AMD. Is it a coincidence that AMD's graphics market share dropped to 18% in Mid 2015 after all these people left? I think not!

This is what he said:
""Many people at AMD are looking to leave. They talk to their friends, especially those who have told people privately, "I'm' leaving", or friends who have just left. They ask, "hey, is there any room for me where you are going?"

AMD's losses of top-rate graphics talent is appalling. In order of losses, AMD lost Rick Bergman, me, Eric Demers, Clay Taylor, Bob Feldstein, Mark Leather, Fritz Kruger, and too many others to name. They've lost a substantial part of the Orlando design team to Apple (about a dozen people I hear). In our business we all know the difference between success and failure is a few percent. Lose key leadership and you've probably lost the critical few percent. Make a graphics chip a bit too power hungry, a bit too expensive, a couple of features substandard, and even more importantly miss market cycles and you start the downward spiral.

It's a real shame. ATI in its latter days and as part of AMD through 2011 was contending against nVidia head-to-head. It is very hard to see how that will happen in the future, and even worse to see the destruction and disintegration of a a world class team such as that.""

Source:

Link


Hopefully with the creation of the RTG and AMD making Raja who came back from Apple and who worked on the original 9700 Pro will help turn things around.

 

[CAT-THE-FIFTH]

This is what a former AMD GPU architect and one of the lead architechts of the 4800 series and the 5800 series had to say about ATI employees after being brought on by AMD, he was also the inventor of AMD's Eyefinity technology. Basically, they were competing head to head with nVidia in therms of competitivenes, and then CEO Rory Read came in 2011 and he laid off many graphics talent including himself. After than a lot of AMD top rated graphics talent left 2012+, which eventually hurt AMD. Is it a coincidence that AMD's graphics market share dropped to 18% in Mid 2015 after all these people left? I think not!

This is what he said:
""Many people at AMD are looking to leave. They talk to their friends, especially those who have told people privately, "I'm' leaving", or friends who have just left. They ask, "hey, is there any room for me where you are going?"

AMD's losses of top-rate graphics talent is appalling. In order of losses, AMD lost Rick Bergman, me, Eric Demers, Clay Taylor, Bob Feldstein, Mark Leather, Fritz Kruger, and too many others to name. They've lost a substantial part of the Orlando design team to Apple (about a dozen people I hear). In our business we all know the difference between success and failure is a few percent. Lose key leadership and you've probably lost the critical few percent. Make a graphics chip a bit too power hungry, a bit too expensive, a couple of features substandard, and even more importantly miss market cycles and you start the downward spiral.

It's a real shame. ATI in its latter days and as part of AMD through 2011 was contending against nVidia head-to-head. It is very hard to see how that will happen in the future, and even worse to see the destruction and disintegration of a a world class team such as that.""

Source:

Link


Hopefully with the creation of the RTG and AMD making Raja who came back from Apple and who worked on the original 9700 Pro will help turn things around.

Which is ironic since Nvidia did an HD5000 series with Maxwell,and ended up selling smaller and cheaper to make graphics cards at the same time AMD increased the costs of their own cards and drop the price,and the performance/watt kind of went a bit south,a bit like Fermi.

The "small die" approach helped ATI and AMD a lot. The moment they started making monster chips again,it all went a bit south since they ignored the midrange and laptops which are much bigger markets.

Remember,the last time ATI/AMD made a monster chip - it was the 2900XT.

I really am concerned AMD will release a GTX1070/GTX1080 level card in a few months which will favour technology over cost effectiveness meaning Nvidia can easily maintain margins even with a price-drop.

 

[Rroff]

None of them are from the biggest tho Roff ^^^^

GeForce GTX 580 520mm2 - GTX560ti ~332 and 520mm2 variants.

GeForce GTX 480 529mm2 - GTX465 529mm2

If you go back beyond that some generations i.e. G70 cores were almost all made at the same size through much of the line up.

Kepler is a different story as things changed a bit there with nVidia trickling out tech and some parts are salvages of salvages from the professional side.

I'm not so familiar with the AMD side but I suspect at times where they've been in that position they've done similar.

 

[CAT-THE-FIFTH]

GeForce GTX 580 520mm2 - GTX560ti ~332 and 520mm2 variants.

GeForce GTX 480 529mm2 - GTX465 529mm2

If you go back beyond that some generations i.e. G70 cores were almost all made at the same size through much of the line up.

Kepler is a different story as things changed a bit there with nVidia trickling out tech and some parts are salvages of salvages from the professional side.

I'm not so familiar with the AMD side but I suspect at times where they've been in that position they've done similar.

Yeah,but Nvidia learned from ATI with the Fermi debacle.

Look at how from Fermi to Kepler to Maxwell to Pascal,they have been trying to reduce production costs whilst increasing ASP. This is why margins are double that of the Fermi days.

AMD cannot keep selling 600MM2 chips with complex packaging for much less than their competitors using tried and tested technology.

OTH,AMD hadn't learnt from the issues Nvidia had with Fermi.

The Polaris series is a step back to a smaller set of dies for AMD. Lets hope they learn this for Vega,so they can fight Nvidia whilst able to make a decent amount of money.

If is another massive chip with complex packaging and barely beats a GTX1080 whilst selling for less,its not going to do much for their bottom line.

Nvidia sell graphics cards with a chip smaller than the one in the GTX970 for well over £500 is only going to increasing margins.

To put it context the GTX1080 has a similar sized chip to the one in the GTX460!!

 

[Rroff]

Yeah,but Nvidia learned from ATI with the Fermi debacle.

Look at how from Fermi to Kepler to Maxwell to Pascal,they have been trying to reduce production costs whilst increasing ASP. This is why margins are double that of the Fermi days.

AMD cannot keep selling 600MM2 chips with complex packaging for much less than their competitors using tried and tested technology.

OTH,AMD hadn't learnt from the issues Nvidia had with Fermi.

The Polaris series is a step back to a smaller set of dies for AMD. Lets hope they learn this for Vega,so they can fight Nvidia whilst able to make a decent amount of money.

Its a balancing act for sure.

 

[FuryXGuy]

The 290X was not that bad as it was a much smaller chip than the 780TI is was competing against. About 430mm^2 vs 550mm^2 and about the same performance, so AMD could afford to undercut the 780TI as it was more expensive for Nvidia to make.

Turn that on its head, could AMD afford to sell 550mm^2 at 80% the price of Nvidia's 430mm^2, and with 16 vs 12 Memory IC's?

Unless AMD can find a way to make chips that would be faster than the 1080TI and smaller the chances are AMD would lose money on them because Nvidia can afford to cut the price of their smaller chips to under what AMD can extract profits from.

AMD do not have the brand clout to sell £500 cards if Nvidia want to deny them sales at that level.

They do at the £150 to £250 range.

Yes Nvidia could sell GTX 1070 dies for £250 to kill the RX 480, but they would make a loss on every-single one they sold. AMD would still sell some while Nvidia are burning money and using £500 1080 dies for £250 chips to try and deny AMD sales there.

Yes die size matters. But technology leadership also matters. You could make up for not selling much high end line ups by having a strong mid range products. By totally ignoring the high end would be stupid. They may not have the brand clout as nVidia but I do know that may people chose the R9 290X over the GTX 780 simply because it was faster. If it is faster and better people will buy it despite "brand clout" AND still be profitable. This has been proven time and time again with the 9700 Pro and the X800XT PE series, the 5870 and the 7970 series. All years where ATI and AMD's graphics division made a profit. A lot of it depends on having a smart architecture and having the right balance in die size and performance/costs. It can be a bit tricky.

Not to mention you get a lot of technical know how. Also, it can be test bed for implementing new tech and gaining know how which can be trickled down to the next gen products at lower prices.

This is what Tesla did when they produced the Roadster. The used many of the technology the learned producing the Roadster and it was implemented in the Model S, which was cheaper. Even though they didn't sell much Roadster but they used the tech they learned in making the Roadster into the Model S which sold significantly more.

 

[humbug]

But you can see the issues as time progressed. Up until GCN,AMD tended to have smaller die sizes for the HD4000,HD5000 and HD6000 series.

At the high end,the Tahiti chips were larger than the GK104 chips. However,for the rest of the range,it was more comparable to Kepler.

The problem is AMD tried to hard to get Nvidia higher end sales since Hawaii,and basically released sub-par midrange GPUs.

This is the problem - they just rebadged older and older chips,and the few updates they made didn't improve performance or power consumption that much and didn't lower costs.

If they kept their eye on the midrange and released more polished cards,they would not have lost marketshare as quickly as they did.

They also did the same with mobile and lost massive amounts of sales.

Going after the high-end at all costs didn't help them IMHO.

The old ATI and AMD generally made sure their mid-range stack could hold their own with Nvidia in all metrics,bar the HD2000 series disaster.

But you saw what they did with the HD3000 series - they dropped costs and reduced power consumption and were still able to salvage that disaster.

Even the HD2000 and HD3000 series disasters,never dropped AMD sales marketshare and perception as bad as during 2014 and 2015.

Then you had 600MM2 Fiji GPU based cards with HBM selling for the same as a GTX980 using a 398MM2 GPU and common GDDR5.

Its the same with their CPUs - they are selling huge chips for little money.

Nvidia margins have double since Fermi with Maxwell and Pascal.

Polaris 10 and 11 is a step in the right direction and they need to start thinking less about innovative tech and more about cost effective tech.

Yes exactly.

The 290X was much smaller than the 780TI..... so much harder for Nvidia to use it to deny AMD sales or profit at the highest end at the time.

The Fury-X vs Maxwell.... absolute disaster, i wonder how much AMD lost on Fiji, i bet its a lot and its not because of the performance difference between it and the 980TI, which was not that much, the production costs Fury-X vs 980TI would have been huge.

The problem remains today, Polaris vs Pascal.
In this the mid range AMD have a lot more control, they can sell a slightly larger die for about the same or even slightly more money, and do that in the numbers they need.

The argument that Nvidia can just use the 1080 to kill the RX 480 is not a valid one, that's Nvidia setting their own coffers alight to hurt AMD, they would burn money on everyone one they sold.

The 1060 is what Nvidia can and have done to compete with the RX 480.

At the enthusiast end Vega needs to be a lot smaller and faster than Polaris to have a hope in competing profitably against a 1080TI.

If its not that then don't, AMD don't have any money left to burn.

 

[CAT-THE-FIFTH]

Its a balancing act for sure.

I managed to add some more musings to my post BTW!!

But Nvidia selling a card with a GF104 sized die(GTX460) for over £500+ is only going to push margins one way.

Yes exactly.

The 290X was much smaller than the 780TI..... so much harder for Nvidia to use it to deny AMD sales or profit at the highest end at the time.

The Fury-X vs Maxwell.... absolute disaster, i wonder how much AMD lost on Fiji, i bet its a lot and its not because of the performance difference between it and the 980TI, which was not that much, the production costs Fury-X vs 980TI would have been huge.

The problem remains today, Polaris vs Pascal.
In this the mid range AMD have a lot more control, they can sell a slightly larger die for about the same or even slightly more money, and do that in the numbers they need.

The argument that Nvidia can just use the 1080 to kill the RX 480 is not a valid one, that's Nvidia setting their own coffers alight to hurt AMD, they would burn money on everyone one they sold.

The 1060 is what Nvidia can and have done to compete with the RX 480.

At the enthusiast end Vega needs to be a lot smaller and faster than Polaris to have a hope in competing profitably against a 1080TI.

If its not that then don't, AMD don't have any money left to burn.

Agreed - they need to make sure they watch production costs. A graphics card with a GF104 sized die(GTX460) is selling is now selling for £500+ in the form of the GTX1080.

 

[Rroff]

The argument that Nvidia can just use the 1080 to kill the RX 480 is not a valid one, that's Nvidia setting their own coffers alight to hurt AMD.

The 1060 is what Nvidia can and have done to compete with the RX 480.

At the enthusiast end Vega needs to be a lot smaller and faster than Polaris to have a hope in competing profitably against a 1080TI.

nVidia can very well tank that a lot longer than AMD can though - but again you are taking what people are saying extremely literally.

The problem for AMD is that the average consumer is nothing like the average poster in this part of the forum - image carries a lot of weight in the mainstream consumer space and AMD can't push a mid-range solution forever without competing in the top end and necessarily do well. Association is very powerful in sales and will direct a lot of people's purchasing.

 

[CAT-THE-FIFTH]

nVidia can very well tank that a lot longer than AMD can though - but again you are taking what people are saying extremely literally.

The problem for AMD is that the average consumer is nothing like the average poster in this part of the forum - image carries a lot of weight in the mainstream consumer space and AMD can't push a mid-range solution forever without competing in the top end and necessarily do well. Association is very powerful in sales and will direct a lot of people's purchasing.

The thing is the HD4850 and HD4870 didn't beat the Nvidia top end cards,but were much smaller chips ATI/AMD could use to get close but charge much less for.

The same goes with the HD5000 series - ATI/AMD didn't go for massive chips,so were able to push the entire range out quicker and so on.

The "small die" strategy worked well.

Unless AMD can launch halo cards which are faster than Nvidia and before Nvidia,they have more chance to denegrate the whole range below it.

AMD,cutting costs on the R9 290 series cooler,made it look hotter than it was and that stuck with AMD cards ever since.

If AMD want to compete at the high end,either blow Nvidia away in performance or make cheaper cards which can blow them away in price with 90% of the performance.

If they can't just posing won't help - nobody will buy those high end cards.

They will just cause an "anti-halo effect".

AMD lost sight of both mobile and the midrange for a few years,and if you look at Polaris 10 and 11 its a return to smaller and cheaper to make GPUs.

 

[humbug]

nVidia can very well tank that a lot longer than AMD can though - but again you are taking what people are saying extremely literally.

The problem for AMD is that the average consumer is nothing like the average poster in this part of the forum - image carries a lot of weight in the mainstream consumer space and AMD can't push a mid-range solution forever without competing in the top end and necessarily do well. Association is very powerful in sales and will direct a lot of people's purchasing.

Then Nvidia would be burning their own money indefinitely, whats more this 1060TI? would be very popular which means they wouldn't have any actual 1080 at £500 or 170 at £350 to sell while they behave in this way, indefinitely?

in this range AMD do not need to drop their prices, they may lose 70% of their RX 480 retail sales but those chips will still go in the new XBox and PS and they would still have CPU's brining in revenues ontop of the Polaris chips sold to Sony and Microsoft.

In other words Nvidia will hurt themselves unsuccessfully trying to hurt AMD.

 

[FuryXGuy]

Which is ironic since Nvidia did an HD5000 series with Maxwell,and ended up selling smaller and cheaper to make graphics cards at the same time AMD increased the costs of their own cards and drop the price,and the performance/watt kind of went a bit south,a bit like Fermi.

The "small die" approach helped ATI and AMD a lot. The moment they started making monster chips again,it all went a bit south since they ignored the midrange and laptops which are much bigger markets.

Remember,the last time ATI/AMD made a monster chip - it was the 2900XT.

I really am concerned AMD will release a GTX1070/GTX1080 level card in a few months which will favour technology over cost effectiveness meaning Nvidia can easily maintain margins even with a price-drop.

This is true. Good analogy with Maxwell vs the Hawaii. This is why I stated you have to smart about die size. Interestingly Carrel Killbrew, the guy I quoted above who was AMD/ATI GPU architect was actually one of the proponents of the "small die strategy". Having a small die doesn't necessarily mean giving up technology leadership. I call it the "smart die" strategy. R9 290/290X was going up against GTX 780/780 Ti, it was not intended to go against Maxwell. Fury was the one which was supposed to counter Maxwell, unfortunately, I just think that the Maxwell architecture and design was just better than Fijii. And it was late to market. it should have been release in 2014 or late 2014 early 2015. That's one of the reason which hurt them.
As a matter of fact, 2014 was the only years over the past 16+ years where AMD/ATI didn't release a top to bottom series of GPU's or even refreshed parts.

Anandtech had a very good article about how Carrell Killbrew came up with the "small die" strategy with the RV770 and it effectively up ended nVidia in that generation cycle. As a matter of fact nVidia actually had losses in the years covering 2008/2009 thanks to HD 4800 series and the 5800 series:

Link

^^ This is probably my all time favorite article on Anandtech.

They also have one about the HD 5800 series.

 

[Rroff]

The thing is the HD4850 and HD4870 didn't beat the Nvidia top end cards,but were much smaller chips ATI/AMD could use to get close but charge much less for.

The same goes with the HD5000 series - ATI/AMD didn't go for massive chips,so were able to push the entire range out quicker and so on.

The "small die" strategy worked well.

Unless AMD can launch halo cards which are faster than Nvidia and before Nvidia,they have more chance to denegrate the whole range below it.

AMD,cutting costs on the R9 290 series cooler,made it look hotter than it was and that stuck with AMD cards ever since.

If AMD want to compete at the high end,either blow Nvidia away in performance or make cheaper cards which can blow them away in price with 90% of the performance.

If they can't just posing won't help - nobody will buy those high end cards.

They will just cause an "anti-halo effect".

AMD lost sight of both mobile and the midrange for a few years,and if you look at Polaris 10 and 11 its a return to smaller and cheaper to make GPUs.

They don't need to beat nVidia they just need a credible product that can compete with nVidia's better cards - the 4870 is a good example.