Features - Roy Taylor: AMD Radeon GPUs remain unsurpassed

[crossy]

Interesting discussion (mainly) to date - especially the ones around Mantle v's DX11/12 and the current NVidia/AMD low level tech comparisons. And I'll second the "tick/tock" theory of AMD/NVidia supremacy - I remember at least two occasions where what I had (which was always NVidia up until last year) was "second best" compared to the latest hotness from the "Red Team".

But I get the distinct impression that the consensus answer to "AMD Radeon GPU's remain unsurpassed" is "no they don't, get your act in gear guys". Hopefully 2015's launches will bring something new and impressive from AMD since an effective NVidia monopoly is bad news for everyone. That said, folks I know with "budget" machines powered by AMD APU's seem to be pretty happy - especially as gaming seems to be better than on the Intel equivalent. (Slight digression there)

Pfffffffffffffffft. In other news, AMD learns trolling.

Don't agree that the article was trolling. What were you expecting "we're quite concerned by the current 970/980's but we hope that we'll have a faster/quieter/cheaper product next year"?

[CAT-THE-FIFTH]

Interesting discussion (mainly) to date - especially the ones around Mantle v's DX11/12 and the current NVidia/AMD low level tech comparisons. And I'll second the "tick/tock" theory of AMD/NVidia supremacy - I remember at least two occasions where what I had (which was always NVidia up until last year) was "second best" compared to the latest hotness from the "Red Team".

Yes,you get it like I do crossy and its been the case from at least 2003 onwards. When I was getting a new card last year,I got a Kepler based one,but all the cards were by that time were so close on feature set and performance it eventually came down to retailer pricing(and a game I wanted).

People have short memories - plenty of times Nvidia have been three to six months behind on moving their whole range to a new node - for example it took Nvidia nearly six months to release he GTX660 and GTX660TI,and so were using less efficient GTX570,GTX560TI 448 and GTX560TI cards to compete against the HD7870 and other AMD cards.

Yet,I did not see any of the Nvdia is "doomed" remarks on this forum,or AMD is worth the premium,etc. In fact when the HD7950 was released for a similar price as the GTX580 with double the VRAM,multi-monitor abilities out of the box,better overclocking and lower power consumption,people on this very forum were moaning that AMD was overpricing the cards yet did not say anything about Nvidia not dropping the price of their old tech.

OTH,when the GTX970 is released with similar performance to the R9 290/GTX780 cards(reviews used slower reference cards to compare performance which are harder to get anyway),lower power consumption,etc,it suddenly one side is doomed and its worth paying the extra,etc due to XYZ reasons.

GPU E-PEEN wars.

[spl]

I didn't say the article was trolling. At the very least it should be taken that I'm saying Roy's responses are trolling. I'm actually glad Hexus posted this and would enjoy reading more interviews with important figures from these companies (though preferably from more honest / less PR-ish types). What I was referring to specifically though was that 'finally some good products' bit that just makes the guy look very unprofessional (though Apple have really lowered the bar in that regard anyway across the whole tech industry).

[DanceswithUnix]

The second generation Fermi parts had better power containment features and the TSMC 40NM process was better yielding by then.

At the time Nvidia were the only people really complaining about yields, so I suspect if they had stuck with the 480 then yields wouldn't have improved anywhere near enough for them. That involved making a new part where they got the design rules right for the process. AMD had already gone through that pain on a lower end part that no-one really cared about so it didn't generate any drama or press.

I see where you are going with the thing about turning the limiter off, but my point was that the 580 running Crysis (not something stupid like furmark) gave better performance than the 480 in about 40W less. That is quite a lot when the silicon has the same number of shaders in the same layout, just not as many disabled because they leak so much power they are better off disabled to feed that power budget into the remaining units.

So I will stick with what I said, Fermi as an architecture wasn't really a problem, the 480 was broken, and you can't really use it as a comparison for anything meaningful.

How about these childish cartoons by Nvidia PR in 2009 taking swipes at Intel??

Wasn't 2009 the year that Intel stated that the FSB license they had sold to NVidia didn't cover their latest chips and hence crippling Nvidia's chipset business? So I think Intel started it, however childish NVidia may have come across in how they dealt with the fallout.

[CAT-THE-FIFTH]

At the time Nvidia were the only people really complaining about yields, so I suspect if they had stuck with the 480 then yields wouldn't have improved anywhere near enough for them. That involved making a new part where they got the design rules right for the process. AMD had already gone through that pain on a lower end part that no-one really cared about so it didn't generate any drama or press.

I see where you are going with the thing about turning the limiter off, but my point was that the 580 running Crysis (not something stupid like furmark) gave better performance than the 480 in about 40W less. That is quite a lot when the silicon has the same number of shaders in the same layout, just not as many disabled because they leak so much power they are better off disabled to feed that power budget into the remaining units.

So I will stick with what I said, Fermi as an architecture wasn't really a problem, the 480 was broken, and you can't really use it as a comparison for anything meaningful.

Yes,but you are forgetting the GF100 and GF110 were HUGE parts - 50% larger than what AMD was doing at the time,ie,nearly 550MM2 against under 370MM2. Yields were a problem - with Fermi Nvidia was paying per completed die,not per wafer(like they did with Kepler),so TSMC was supposedly taking some of the hit as Nvidia was a large customer.




It was a 30W difference with Crysis2.

However.lets look at the midrange.



Look at the HD6950 2GB and GTX560TI 1GB. The former had double the VRAM and still consumed less power.

Yet,once they got to the next generation they refined things.

Look at the HD4000 series against the GTX200 series - Nvidia had a performance/watt advantage which went away with the next generation and it then happened with the HD7000 for AMD and so on.

Its all very cyclic in nature. Also,people trying to say look what happened on the CPU side with AMD also don't seem to understand that Intel is ahead on process node tech too,which helps them a lot.

Even though I am not directing it at you,people peddling all this gloom and doom with AMD now,seem to ignore what Nvidia managed to do with Fermi,or what has happened before.

Like clockwork you see this doom and gloom on tech forums at nearly every major GPU launch too. Its almost Apple level hysteria IMHO.

Wasn't 2009 the year that Intel stated that the FSB license they had sold to NVidia didn't cover their latest chips and hence crippling Nvidia's chipset business? So I think Intel started it, however childish NVidia may have come across in how they dealt with the fallout.

Yep - but it shows you that Nvidia PR will put down the competition when it wants to. Intel AFAIK,did not really say anything in the media. In fact the outburst by one of their senior people when AMD bought SeaMicro was unheard off for Intel.

Look at how the Nvidia CEO himself put down an AMD product himself.

Or Nvidia comments about mocking AMD console wins.

Nvidia mocking the HD7970 on FB and so on.

So all the crocodile tears about some AMD PR chap doing something similar is funny.

But,I could OFC also do a list of all the times Nvidia and AMD PR have put down the competition too.

Even though this is not directed at you,I do think people live in a bubble at times.

[DanceswithUnix]

So,people peddling all this gloom and doom with AMD now,seem to ignore what Nvidia managed to do with Fermi.

Well I for one am not peddling doom and gloom, I just wanted to tidy up one point: That 480 was a broken outlier and shouldn't be used as a data point for anything.

Anandtech put the power difference at 36W, and yes I was sloppy and didn't bother to even do the subtraction and just went with "that looks about 40W" because precision wasn't central to my point, those lower power figures are for noticeably higher performance so it isn't like for like, the 580 is just plain better, and the 580 is Fermi.

[CAT-THE-FIFTH]

Well I for one am not peddling doom and gloom, I just wanted to tidy up one point: That 480 was a broken outlier and shouldn't be used as a data point for anything.

Anandtech put the power difference at 36W, and yes I was sloppy and didn't bother to even do the subtraction and just went with "that looks about 40W" because precision wasn't central to my point, those lower power figures are for noticeably higher performance so it isn't like for like, the 580 is just plain better, and the 580 is Fermi.

But it also does even change the fact that throughout the whole range(maybe not so much with the HD6970),AMD was ahead in performance/watt for the entire HD5000 and HD6000 series and especially at the midrange too. Even go back to the Anandtech HD6850 review - the HD6850 at nearly 1GHZ was lucky to match a stock clocked GTX460 in power consumption,and most GTX460 cards had largish pre-overclocks which made power consumption worse. Fermi lost on both performance/watt and and ESPECIALLY performance/mm2 metrics as a gaming GPU(one of the effects of having a better GPGPU capable one at the time),so ultimately they still refined it enough in the end with Kepler and Maxwell.

If anything the jump from VLIW4/VLIW5 to GCN,was HUGE in terms of the changes. Nvidia did the changes over the GT200 and both Fermi generations. AMD caught up with Nvidia in general compute performance in one generation,and interestingly did not pay as big as die area penalty either - in fact AMD still seems to have very dense dies at the expense of some extra power consumption over equivalent Nvidia GPUs.

This is the thing - the GK104 and Tahiti comparisons were comparing a compute stripped part and a part with massively greater DP performance,so ultimately you could not compare uarchs at all.

But look at something like the HD7850 and HD7870 - the AMD midrange parts were very competitive in performance and power consumption with my GTX660 for instance or the GK104. Nvidia just stuck with a larger chip.

Its a fallacy GCN was worse than Kepler in power consumption in every part - it was only the top end parts with greater DP performance which showed this and even then they had far greater transistor density and in the case of Hawaii additional logic for audio,etc which took up more power. Nvidia was again using larger die area.

This is also the other thing - since 2011(in three years) AMD has caught up or exceeded Nvidia with the fastest single GPU and dual GPU solutions on 5 instances.

Yet from late 2006 to late 2010,ie,4 years,AMD surpassed Nvidia fewer times overall and yet at that time it was power consumption be damned - Nvidia has the performance GPUs,its what makes them better and power consumption is not so important. Yet,now it appears things have changed.

Funnily enough even the R9 290X and GTX780TI consume less power than a GTX570 or even the HD6970 IIRC.

The funny thing is if AMD were to release a 20NM GPU in January which exceeded the performance/watt of the GM204,was slightly faster and was cheaper,I expect there will be no "Nvidia is doomed since they XYZ months behind AMD in process node jumps" at all. I expect more "AMD is doomed since they need a new process node and took three to four months and are always behind Nvidia" and if Nvidia releases something 4 months later which consumes a bit less power and is slightly faster at the same price it will be "I told you so" I suspect.

You know why I say this?? Its happened before!

Even when AMD/ATI was ahead on all accounts with the HD5000 series against the G92/GT200,look how many waited for six to nine months for Fermi??

Yet,if that happened with AMD...the sky would be falling.

[DanceswithUnix]

Yep - but it shows you that Nvidia PR will put down the competition when it wants to. Intel AFAIK,did not really say anything in the media.

Intel wouldn't say anything to the media, they were probably laughing too hard at what they had pulled off. They had traded use of Nvidia's graphics patents for use in their integrated graphics in exchange for letting Nvidia make chipsets, and then told Nvidia they could only make chipsets for socket 775 and nothing newer. They were boldly telling the world how good their Larrabee graphics was going to be (which no doubt also required Nvidia patents), while starting to block Nvidia from even making Ion addons for Atom CPUs.

[CAT-THE-FIFTH]

Didn't some senior AMD and Nvidia PR people have a slinging match in one of the threads on Hexus a few years ago?? It was an amusing read.

[spl]

Didn't some senior AMD and Nvidia PR people have a slinging match in one of the threads on Hexus a few years ago?? It was an amusing read.

That is something I would love to read!

[crossy]

I didn't say the article was trolling. At the very least it should be taken that I'm saying Roy's responses are trolling. I'm actually glad Hexus posted this and would enjoy reading more interviews with important figures from these companies (though preferably from more honest / less PR-ish types). What I was referring to specifically though was that 'finally some good products' bit that just makes the guy look very unprofessional (though Apple have really lowered the bar in that regard anyway across the whole tech industry).

Darn it - I'm going to have to agree with most of what you're saying above. And I'll second an interest in the "voice of the industry" type articles, if (and it's a big "if"), they can be arranged.

Where I'm going to disagree is with the "trolling" accusation. RT is the PR person for AMD, so you expect him to "big up" his company at the expense of the nearest competition. You see it with Intel/AMD, IBM/HP, Apple/everyone-else (oops, just sunk to trollage myself) so that's pretty normal PR/marketing 101. Heck, I'd have even understood if he'd come out with "anyone who buys into NVidia's snake oil is a fool" type comments. Comments were relevant, even if I'm going to politely disagree with some of what was said. Bombast level - for an American - was also pretty much what I'd expect, although I've seen a lot worse out there.

Downside of the Voice of the Industry articles is that you're always going to get effectively an advert for whichever company is being interviewed, and good luck thinking that we'd get some "techies" - they don't get "let out", instead it's the marketing dudes (and dudettes) that do that kind of thing. Like this article though, they're potentially a great starting point for an informed and informative discussion.

[CAT-THE-FIFTH]

That is something I would love to read!

I think it was Richard Huddy and someone else. It was a few years ago during 2009 or 2010 IIRC.

[spl]

Darn it - I'm going to have to agree with most of what you're saying above. And I'll second an interest in the "voice of the industry" type articles, if (and it's a big "if"), they can be arranged.

Where I'm going to disagree is with the "trolling" accusation. RT is the PR person for AMD, so you expect him to "big up" his company at the expense of the nearest competition. You see it with Intel/AMD, IBM/HP, Apple/everyone-else (oops, just sunk to trollage myself) so that's pretty normal PR/marketing 101. Heck, I'd have even understood if he'd come out with "anyone who buys into NVidia's snake oil is a fool" type comments. Comments were relevant, even if I'm going to politely disagree with some of what was said. Bombast level - for an American - was also pretty much what I'd expect, although I've seen a lot worse out there.

Downside of the Voice of the Industry articles is that you're always going to get effectively an advert for whichever company is being interviewed, and good luck thinking that we'd get some "techies" - they don't get "let out", instead it's the marketing dudes (and dudettes) that do that kind of thing. Like this article though, they're potentially a great starting point for an informed and informative discussion.

Yeah, as I said, Apple have really taken things down a notch in that regard starting with their "I'm a Mac" advertisements (maybe even earlier?). So now everyone is doing it and it's just ugly. I'd argue that the fact it's widespread doesn't mean it no longer qualifies as trolling.

It's a real shame these companies don't have the decency to sell their products on their own merits rather than bashing the competition. Surely it's counter-productive anyway - you'd think that kind of behaviour would put a lot of people off them. Or at least it might do if everyone else wasn't doing the same. On the flip side, it does increase the respect I have for the few companies that don't engage in that kind of behaviour. And I know, you can be polite on your public face while running a business in a dishonourable way, but realistically, maybe a polite public face is the best we can hope for :-)

You're absolutely right about the marketing crap that comes out of these interviews. I was thinking that as I wrote it (hence the 'less PR-ish' addition) but yeah, it's probably not ever going to happen. Hexus could set some 'no BS' ground rules, but assuming they'd even still then get the interview, it would just be the same stuff dressed up in different words. Maybe a better requirement would be 'tell us something we don't already know', like maybe some release plans for example! Some kind of hard facts anyway that aren't already known.

[edzieba]

compute performance [...] People like you did the same with Tahiti vs the GK104 too.

Because we're talking about playing games, not low-end simulation (high-end simulation would be on cards that can use ECC, or dedicated supercomputer cores like Cray's custom vector ASICs).

Lets look at the gaming power consumption figures in more detail though.

Yes, lets. Remember, PSUs are designed to handle average loads (unless you buy a real cheap and nasty one that wouldn't meet the ATX spec). Momentary peaks are what all those capacitors in a PSU are for:

If we compare cards of similar performance, you can see the efficiency jump between Kepler and Maxwell. 780 to 970 = 37% increase. If Tonga Pro (AMDs latest iteration and most efficient mid-high end chip) used the same improvements, it would still just be tailing the 970 at lower performance. Even if we assume a hypothetical super-Tonga Pro for high-end cards would be just as efficient at higher clocks to match performance, this is just an improvement up to parity. And Tonga Pro is ALREADY a power-optimised Tahiti.

this slide you on purpose keep ignoring:

http://images.anandtech.com/doci/797...8-AM_575px.jpg

It's a roadmap. There's no guarantee what appears on the roadmap will actually materialise at the time and in the form estimated. See: the sudden loss of unified memory from Maxwell, the appearance of Pascal ad the pushback of Volta, shortly before Maxwell's release, after being on the roadmap for quite some time (over a year and a half!). Publicly traded companies have an incentive to keep their roadmaps as optimistic as possible right up until product release.

Furthermore, that's a roadmap for AMD's CPUs, not GPUs. AMD have not released a recently updated GPU roadmap, and the last leaked roadmap had Tonga arriving with HSA 2H 2014 as part of an across-the-board refresh, which clearly has not occurred. AMD's last official roadmap update focussed on their APU line, with only a side mention of HBM integration in terms of cost reduction during Q&A.

Maxwell is very peaky in power consumption but any body who had bothered to keep up to date would have known this for ages.

And nobody is disputing this. But as you can see from the TH power graphs, the peaks are at the ms timescale. This is well within what a correctly designed (i.e. 50% rated output transient, 2.5A/microsecond according to the ATX specifications) PSU is designed to handle. It's a non-issue.

Plus trying to sound cool by saying AMD needs a new uarch all the time is funny.

No, it's pragmatic. AMD can roll out theoretical improved Tonga Pro derivatives, but that's going to be an exercise in catching up, and from several months head start. Then there's the more wide-reaching changes line the integration of HBM which will require major updating.


We all want AMD to pull off a cheap, high performance, low-power release. Doing so is good for everyone. It's just not likely without AMD investing a LOT into testing a new process node while modifying their architecture further (or moving to a new one), and AMD is not cash-flush at the moment. A low-yeild super-part will not net them the profits they need.


It's likely we'll see the node-move first appear in volume chips for the PS4 and XB1. With an identical chip design the node-move can be done more safely, and the increased yeilds and lower per-chip costs will relieve the pressure on AMD to drop supply costs (see: the dropping component prices thoughout the life of the 360 and PS3). AMDs low-mid range chips are still competitive, so it's possible we may see them refreshed at 20nm too. But I don't think

[CAT-THE-FIFTH]

Because we're talking about playing games, not low-end simulation (high-end simulation would be on cards that can use ECC, or dedicated supercomputer cores like Cray's custom vector ASICs).

Because you fail to realise the GK110 and Hawaii GPUs were designed for mixed compute and gaming workloads?? There are extra transistors and things like high memory bandwith controllers which are needed for such thins - again you are stick in a microcosm. They are built for DP compute performance too.

Throwing around terms to appear to sound clever is not going to help you at all.

Yes, lets. Remember, PSUs are designed to handle average loads (unless you buy a real cheap and nasty one that wouldn't meet the ATX spec). Momentary peaks are what all those capacitors in a PSU are for:

If we compare cards of similar performance, you can see the efficiency jump between Kepler and Maxwell. 780 to 970 = 37% increase. If Tonga Pro (AMDs latest iteration and most efficient mid-high end chip) used the same improvements, it would still just be tailing the 970 at lower performance. Even if we assume a hypothetical super-Tonga Pro for high-end cards would be just as efficient at higher clocks to match performance, this is just an improvement up to parity. And Tonga Pro is ALREADY a power-optimised Tahiti.

You mean like the fact that Tonga is binned desktop part - its not even fully enabled. The M295X OTH is a fully enabled chip(well not for the memory controller) and is already in the new iMac:

http://www.barefeats.com/imac5k3.html

Looks like the Tonga parts for desktop are salvage parts - probably like the HD7870XT/LE.

Its fitting into the same TDP and power budget as the GTX780M which has a 100W TDP.


Lets look at the gaming power consumption figures in more detail though since you want to conveniently ignore them to peddle more doom and gloom.





The same was noted with the GTX750TI(GK107) since TH uses very high speed measuring equipment - which again does not negate what I said before,or this slide you on purpose keep ignoring:

http://images.anandtech.com/doci/797...8-AM_575px.jpg



Maxwell is very peaky in power consumption but any body who had bothered to keep up to date would have known this for ages.

You on purpose seem to ignore or fail to understand what Maxwell does differently over Kepler. Its been described in various articles since teh beginning of this year - the peaky power consumption is consistent with how the effiency gains have happened. You on purpose feign ignorance.

FFS,during compute loads which are more constant the power consumption and TDP rises which is consistent with how the Maxwell voltage and power scaling mechanism work.

It's a roadmap. There's no guarantee what appears on the roadmap will actually materialise at the time and in the form estimated. See: the sudden loss of unified memory from Maxwell, the appearance of Pascal ad the pushback of Volta, shortly before Maxwell's release, after being on the roadmap for quite some time (over a year and a half!). Publicly traded companies have an incentive to keep their roadmaps as optimistic as possible right up until product release.

Furthermore, that's a roadmap for AMD's CPUs, not GPUs. AMD have not released a recently updated GPU roadmap, and the last leaked roadmap had Tonga arriving with HSA 2H 2014 as part of an across-the-board refresh, which clearly has not occurred. AMD's last official roadmap update focussed on their APU line, with only a side mention of HBM integration in terms of cost reduction during Q&A.

Oh please,you want to ignore that since you want to peddle all your stupid doom and gloom nonsense.

Its an APU roadmap which covers tech which is integrated on both parts of it. But OFC,don't add two and two together.

Funny how you come out of the woodwork just about now,and then you will dissapear when all your crap doom and gloom goes away.

You don't even understand what the relevance of teh roadmap let alone why its important.

I am not stuck in a microcosm like you - I update CPU and GPU roadmaps on here and OcUK. The tech in that roadmap is appearing in multiple products.

AMD makes both and maybe you cannot add two and two together,but AMD is still stuck on 28NM for CPUs for another year:

http://techreport.com/news/27169/rep...ng-in-december

I wonder how they are going to improve performance/watt on a stagnant node. Oh wait - lets look at that roadmap.

And nobody is disputing this. But as you can see from the TH power graphs, the peaks are at the ms timescale. This is well within what a correctly designed (i.e. 50% rated output transient, 2.5A/microsecond according to the ATX specifications) PSU is designed to handle. It's a non-issue. No, it's pragmatic. AMD can roll out theoretical improved Tonga Pro derivatives, but that's going to be an exercise in catching up, and from several months head start. Then there's the more wide-reaching changes line the integration of HBM which will require major updating.

No,because it does not fit your doom and gloom scenario. You can deflect all you want but that peaky power consumption is consistent with the reasons why Maxwell appears more efficient, and why under constant fixed GPU load the power consumption jumps up and unlike you I have had a far better track record on forums of actually being right about these things. I even predicted the problems with the new generation boost systems from AMD and Nvidia nearly three years ago.

No, it's pragmatic. AMD can roll out theoretical improved Tonga Pro derivatives, but that's going to be an exercise in catching up, and from several months head start. Then there's the more wide-reaching changes line the integration of HBM which will require major updating.

We all want AMD to pull off a cheap, high performance, low-power release. Doing so is good for everyone. It's just not likely without AMD investing a LOT into testing a new process node while modifying their architecture further (or moving to a new one), and AMD is not cash-flush at the moment. A low-yeild super-part will not net them the profits they need.


It's likely we'll see the node-move first appear in volume chips for the PS4 and XB1. With an identical chip design the node-move can be done more safely, and the increased yeilds and lower per-chip costs will relieve the pressure on AMD to drop supply costs (see: the dropping component prices thoughout the life of the 360 and PS3). AMDs low-mid range chips are still competitive, so it's possible we may see them refreshed at 20nm too. But I don't think

It is not pragmatic in any way. You are just sticking your head in the sand and going la!la!la!la! AMD is doomed!la!la!la!la!

At 90% of new Nvidia card launches in the last 7 years,it has been AMD has been doomed trolling rubbish from people like you.

Agreed.

Looking at the last few years of GPU releases.

2002-2003 ATI is doomed due to the cancelation of the 8500XT and Nvidia having the TI4200 and TI4600

2003-2004 Nvidia is doomed since the R300 is first to DX9 and faster than the FX series.

2004-2005 ATI is doomed since the 6000 series Nvidia cards support DX9C and they don't. Nvidia is lower power consumption.

2005-2006 Shifts between ATI is doomed and Nvidia is doomed.

2006-2007 ATI is doomed as the G80 and G92 have better performance and lower power consumption than the HD2000 and HD3000 series

2008-2009 ATI is first doomed since they having nothing to compete with the GT200. Then AMD launches the R700,and now Nvidia is doomed as Nvidia has to price cut huge chips against AMD ones which are smaller

2009 - 2010 Nvidia is doomed since ATI/AMD is first to DX11,has lower power consumption and smaller dies than the GTX400 series. Although soon with the GTX460 series,AMD is doomed.

2010 -2011 AMD is partially doomed since they don't have the fastest card anymore,and Nvidia has better tessellation

Late 2011 to early 2012. AMD is doomed due to the HD7970 and even more doomed with the GTX680.

Late 2012 to middle 2013. AMD is doomed as they have nothing to compete with once the Geforce Titan and GTX780 are released

Late 2013. AMD is still doomed with the R9 290 and R9 290X releases,since Nvidia quickly launched the GTX780TI and the AMD cards have black screens,throttle and run too hot and explode all the time.

Late 2012 to 2013. Nvidia is doomed due to AMD winning console contracts.

Early 2014. AMD is doomed due to the GTX750TI

Late 2014. AMD is doomed due to the GM204.

Potential next doom point - Nvidia releases 20NM GM200/GM210 in small quantities at £1000 and even if AMD has the fastest card in the R9 390X at £500 before then is still doomed.

And so on.

When it comes to GPU releases people seem to have very short memories indeed. At every launch the standard stuff happens.Hilarious.

Its so predictable,so much so I predicted the response to this launch months before it came out. I believe I mentioned it here or on OcUK or some other forum.

Funnily enough I got quite good at predicting the response to Apple and Android launches for a few years too.

So like I said we can agree to disagree.

You can play the broken record for as long as you want,and hope desperately AMD will stay behind.

I have no interest in continuing this "discussion" with you.

Have the last "AMD is doomed" word but you are on my ignore list now.Chau.

[edzieba]

You mean like the fact that Tonga is binned desktop part - its not even fully enabled. The M295X OTH is a fully enabled chip(well not for the memory controller) and is already in the new iMac:

And is a mobile part. AMD, like Nvidia, do not just cherry-pick high performing desktop parts for their mobile parts. The design is modified both in the actual transistor arrangement, and often even the chemistry (to lower leakage current at the detriment of maximum switching speed and more expensive processes).

Maxwell is very peaky in power consumption but any body who had bothered to keep up to date would have known this for ages.

Again, it doesn't matter. Power draw peaks and dips, the PSU continues to provide power. This is normal behaviour.

I even predicted the problems with the new generation boost systems from AMD and Nvidia nearly three years ago.

It's a nonexistant 'problem'. Cutting power supply to unused chip sections is not 'cheating', it's good design. To do otherwise is merely to waste power. Finer grained power gating has been used for years in CPUs and GPUs, and efficiency gains are more often being made by more aggressive gating than by node improvement.



Throughout this entire thread, the one person who has been repeatedly mentioning "AMD is doomed" is you.
Everyone else has been recounting the cycle of one manufacturer having the upper hand, then the other. We can all hope this continues to be the case, as on the occasion that one manufacturer does consistently outperform another, prices go up and say up, and pressure to increase performance decreases.