AMD Polaris 10 Samples work at 1.27 GHz

[swilli89]

Not sure if this thread was purposefully derailed but anyways..

Polaris 10 may very well have massive overclocking potentially if this is a reference clock without any boost. It's starting to seem like AMD will have a stronger chip per mm2 this round based on the rumors we're seeing. However as 1080 perf could more or less be estimated based on the fact it's basically an overclocked maxwell, GCN 4.0 may bring a massive shift in their performance per SP, which makes it all but impossible to assume what it's performance would be like.

While GP104 is 50% larger than Polaris 10, does anyone really think it will end up 50% faster? The people estimating P10 @ 290x performance are nuts to believe that. Vastly improved resources clocked 30% higher will make for a very interesting case.

On to mobile, AMD should have about 80% of laptop GPU shipments if the market was perfect and unbiased by marketing/backroom deals as nvidia has zero alternative to p11 and P10 which will serve pretty much the entire dGPU laptop market's needs.

 

[Mopetar]

Do we even have die sizes for Vega?

I recall several estimates of ~460 mm^2 for big Vega, but I don't remember if there was a solid source for that. I think most were arriving at that figure by using the Polaris die size estimate (~230 mm^2) and doubling it for a theoretical 5120 SP chip.

 

[Arachnotronic]

I recall several estimates of ~460 mm^2 for big Vega, but I don't remember if there was a solid source for that. I think most were arriving at that figure by using the Polaris die size estimate (~230 mm^2) and doubling it for a theoretical 5120 SP chip.

Right, so there's literally nothing known about Vega's die size. Just guesswork.

 

[Arachnotronic]

On to mobile, AMD should have about 80% of laptop GPU shipments if the market was perfect and unbiased by marketing/backroom deals as nvidia has zero alternative to p11 and P10 which will serve pretty much the entire dGPU laptop market's needs.

Proof? Seriously, what basis do you have for this statement?

 

[Bacon1]

there is wrong and there is missinformation

https://www.youtube.com/watch?v=UKkFqG77-x4
glorious as they say "dx12 example of dynamic load"
while the fraps is running
D:

Ahahahha thats awesome. Nvidia does DX12 so well they've even managed to support FRAPS

On topic, can't wait for reviews!

 

[crisium]

While GP104 is 50% larger than Polaris 10, does anyone really think it will end up 50% faster? The people estimating P10 @ 290x performance are nuts to believe that. Vastly improved resources clocked 30% higher will make for a very interesting case.

GTX 1080 is averaging 50% over Fury Air. People are guessing Polaris 10 to be anywhere from 390 to faster than Fury X based on their level of optimism, so matching Fury Air would be middle of the road with the expectations.

 

[Concillian]

It is not an apples to apples comparison between Polaris clock speeds and Pascal clock speeds. Wait for review benchmarks, that is all that matters.

But it should be ~reasonably apples to apples comparison to 28nm GCN per shader per clock performance and polaris GCN per shader per clock performance. Nobody is expecting massive IPC improvements from the architecture. So it's just a matter of clock x shader and comparing with existing AMD cards.

Obviously Pascal is very different in this aspect compared with previous gen nVidia cards.

2304 shaders at 1.27 GHz is within 5% of the shader power of a 390x if "IPC" remains equal. Probably there will be some minor architecture improvement, but if this info is true, we have at least a ballpark of where stock performance will be.

AMD was pretty conservative with clock speeds on 28nm launch.

 

[airfathaaaaa]

Right, so there's literally nothing known about Vega's die size. Just guesswork.

well amd stated up to 18 billion transistors for big vega so when p10 specs are here will can see

 

[Armsdealer]

P10 = 2048 SPs @ 1340mhz or so, 224GB/sec bandwidth 150W TDP

I think this makes sense, but it doesn't match with the 36 cu / 2304 sp listing on sisoft. If it's 32 cu, p10 uncut is only somewhat faster than 390x, and not a 1070 competitor. I wouldn't be surprised to see an x2 entrant in this scenario.

The key question is whether it's 2048 or 2560 sp.

 

[Bacon1]

But it should be ~reasonably apples to apples comparison to 28nm GCN per shader per clock performance and polaris GCN per shader per clock performance. Nobody is expecting massive IPC improvements from the architecture. So it's just a matter of clock x shader and comparing with existing AMD cards.

Obviously Pascal is very different in this aspect compared with previous gen nVidia cards.

2304 shaders at 1.27 GHz is within 5% of the shader power of a 390x if "IPC" remains equal. Probably there will be some minor architecture improvement, but if this info is true, we have at least a ballpark of where stock performance will be.

AMD was pretty conservative with clock speeds on 28nm launch.

AMD has talked about a lot of hardware changes in GCN 4.

 

[MrTeal]

Off-topic to this thread but 1080 is not 15-20% faster than 980 SLI.
http://www.gamespot.com/articles/nvidia-geforce-gtx-1080-review/1100-6439863/

Paul from Newegg tested it too and 980 OC SLI outperformed 1080 @ 1.885mhz by even greater amounts.
https://www.youtube.com/watch?v=kYdzZzSzF2I

Of course 1080 is still a better buy than 980 SLI due to consistent performance of 1 card but this also means 1070 SLI will be a better buy than 1080.

Well yeah, when SLI works it the 980SLI should be faster than a single GTX1080. SLI working isn't a given though; look at TPU's game suite for that.
Game: 970 SLI scaling @ 4k
Anno 2205: -5%
AC Syndicate: 64%
BF3: 78%
BF4: 79%
Arkham Knight: -3%
Black Ops 3: 73%
Crysis 3: 83%
FO4: 57%
FC Primal: 68%
GTA V: 71%
Hitman: 82%
Just Cause 3: -7%
Rainbow 6 Seige: 17%
RotTR: -18%
TW3: 62.2%
WoW: 82%

So, how do you compare the speed of a single card to an SLI setup? Do you just include the typical games where you see 60-80% scaling? Should you include Rainbow 5 Seige where there is scaling but it sucks? Should you also include the four games with negative scaling?

I love my CF 290s, but I really think if you just include games where scaling works it paints much too rosey a pictures of multi-GPU setups. It's great when it works, but it is pretty limited in many way especially if you want to play games on launch. This applies just as much to AMD once Vega launches. Single Vega might be slower than Fury CF when CF works, but it's tough to say the CF setup is faster if it doesn't work a 1/4 of the time.

 

[RussianSensation]

GTX 1080 is averaging 50% over Fury Air. People are guessing Polaris 10 to be anywhere from 390 to faster than Fury X based on their level of optimism, so matching Fury Air would be middle of the road with the expectations.

1080 is not 50% faster than Fury X. A max overclocked 1080 is.

1080 is 26% and 33% faster (max power target) than Fury X @ 4K per Computerbase.

1080 is 33.3% faster than Fury X @ 4K per TPU.

1080 is 33.3% faster than Fury X @ 4K per Sweclockers.

1080 is 28% faster than Fury X @ 4K per TechSpot.

Nowhere close to 50% faster unless you meant 1080 max OC vs. a stock Fury X.

Right, so there's literally nothing known about Vega's die size. Just guesswork.

It would be odd to have a 315mm2 Vega with 700GB/sec+ HBM2 but Polaris 10 230mm2 with 256-bit bus and only 7Gbps 224GB/sec bandwidth. You can connect the dots on this one that there are either 2 Vega chips (1x HBM2 and 1x GDDR5X) or if Vega is a cut-down part, at least one of the Vega chips should have a die larger than GP104's.

Remember $400 R9 290 and $550 290 X vs. $500 780 and $700 780Ti?

AMD doesn't need to outright beat 1080 to shake up the market, especially so if most AIBs don't abide to NV's $379/$599 MSRP targets. I know you prefer NV and tend to underestimate AMD but last 2 out of 3 generations, AMD won.

Maxwell = 980Ti > Fury X
Kepler gen 2 = 780Ti < 290X
Kepler gen 1 = 680/770 < 280X/7970Ghz

It would be foolish to write AMD off given how 1080 is ~28-33% faster than Fury X @ 4K. A stronger point to make is why wait for Vega that's out in 6-8 months when you can enjoy 1070/1080 for the next 6 months and then just sell those cards and get Big Pascal in 2017 for a 30-35% boost over 1080. That's a very fair and reasonable argument to make

So, how do you compare the speed of a single card to an SLI setup? ...It's great when it works, but it is pretty limited in many way especially if you want to play games on launch. This applies just as much to AMD once Vega launches. Single Vega might be slower than Fury CF when CF works, but it's tough to say the CF setup is faster if it doesn't work a 1/4 of the time.

Ya, you and I agree on this point.

"Of course 1080 is still a better buy than 980 SLI due to consistent performance of 1 card but this also means 1070 SLI will be a better buy than 1080."

Personally, I would pick GTX1070 OC or GTX1080 or GTX980Ti OC over R9 295X2 for games because the performance is more consistent.

I think this makes sense, but it doesn't match with the 36 cu / 2304 sp listing on sisoft. If it's 32 cu, p10 uncut is only somewhat faster than 390x, and not a 1070 competitor. I wouldn't be surprised to see an x2 entrant in this scenario.

The key question is whether it's 2048 or 2560 sp.

My view has been consistent than I never believed Polaris 10 would ~ 980Ti @ $299 and be 1070's competitor. I maintained that P10 would be ~390X/Fury non-X at 110-130W TDP. All of AMD's statements of bringing 290X level of performance below $349 never aligned with P10 being a 980Ti/Fury X replacement. Right now the official MSRP for 1070 is $449 until AIB cards show up and prove to us that there will be sub-$400 cards. That leaves AMD the entire sub-$350 space for Polaris 10. Then once GP106 launches, AMD can bundle AAA games and/or drop prices come Q3 2016.

Since more than 85% of the dGPU market purchases GPUs below $349 mark, AMD is going after more than 85% of consumers while NV is targeting less than 15%. Completely different launch strategies. No need to try and hype up Polaris 10 - a Pitcairn HD7850/7870 successor - to fit higher-end segments above it when it was not never designed to fit them. We should be cautious and assume P10 is only as fast as 390X+5% at much lower prices.

If AMD prices such a card at $269, what competition does NV have in the short-term? Nothing.

If 1070 = Titan X, and the cheapest 1070 is $379, that's 41% more expensive for ~25% higher performance. Worse deal.
If retail 1070 costs $429 on average, that would make it 59% more expensive for ~25% higher performance. An even worse deal.

But here is the catch, most people don't have 1440p monitors and don't have Core i7s @ 4.5Ghz+.

For 1920x1200 60Hz gaming and lower (what most PC gamers have), a 150W $269 card with ~390X level of performance is a WAY smarter buy than a $380-450 card that's 25% faster at 1440p. 1070/1080 only make sense for 1080p 120-144Hz, 1440p 60Hz and higher. That's something professional reviewers aren't talking about for now since it makes NV's high-end cards a complete waste of $ for the mainstream gamer who isn't gaming on an i7 4790K/i7 6700K max overclocked.

This Russian review highlights exactly what I am talking about by using modern titles with AA on higher end cards.
https://www.overclockers.ru/lab/76273_8/obzor-i-testirovanie-videokarty-nvidia-geforce-gtx-1080.html

1080p 60Hz gaming

1080p scores:

980 = 84.7 fps (overkill)
Fury X = 93.6 fps
980Ti = 105.9 fps
1080 = 136.4 fps
1080 OC = 142.5 fps

Even 980 level of performance for a 1080p 60Hz mainstream gamer is more than enough. Why spend more $ on a $400-700 card? Waste of $! Newsflash -- it's 2016 -- time to get a 1440p, 1600p, 4K monitor. At some point PC gamers have to realize the times move on and what used to be the standard in the past (1080p) is not the standard anymore. Today a $400-700 GPU paired with a $100-200 19-24" 1080p 60Hz monitor is just a horribly unbalanced PC. Time to realize that and get a monitor upgrade FIRST.

1440p scores:

980 = 57.6 fps
Fury X = 69.7 fps
980Ti = 73.4 fps
1080 = 95.1 fps
1080 OC = 101.1 fps

We have to be very careful looking at just average charts such as Sweclockers, Computerbase, TPU. Once in a while we have to seriously step back and actually look at the level of perfomrance. Ya, so one card is way faster than another card but if the slower card is getting 85 fps and the faster card is getting 140 fps, on a 60Hz 1080p monitor, there is no difference without going crazy with VSR/DSR.

What NV doesn't want anyone to know is that 1070/1080 are NOT 1080p 60Hz cards because it would crater their sales since only a small fraction of enthusiasts have 1440p-4K screens and multi-monitors.

This is where AMD strikes with Polaris 10, straight into the heart of mainstream. Why overpay for a faster card for 1080p 60Hz gaming when you won't feel the benefits?

 

[Arachnotronic]

I love my CF 290s, but I really think if you just include games where scaling works it paints much too rosey a pictures of multi-GPU setups. It's great when it works, but it is pretty limited in many way especially if you want to play games on launch.

Truth! Having a GPU sitting there unused when you've paid $$ for it sucks. Period.

 

[crisium]

1080 is not 50% faster than Fury X. A max overclocked 1080 is.

Fury Air, I said. And it is 50%+ faster at 1080p and very close to that at 1440p (47%-49% from your 2 sources), which in the context of Polaris 10 I would imagine most would play games here. If Polaris 10 performance is here, based on speculative die sizes, I would be happy.

 

[Bacon1]

@RS, he did say Fury Air not Fury X, and that is probably @ 1080p not 4k.

Why you'd pay $700 for 1080p... I'd never know.

 

[Armsdealer]

My view has been consistent than I never believed Polaris 10 would ~ 980Ti @ $299 and be 1070's competitor. I maintained that P10 would be ~390X/Fury non-X at 110-130W TDP. All of AMD's statements of bringing 290X level of performance below $349 never aligned with P10 being a 980Ti/Fury X replacement. Right now the official MSRP for 1070 is $449 until AIB cards show up and prove to use there will be sub-$400 cards. That leaves AMD the entire sub-$350 space for Polaris 10. Then once GP106 launches, AMD can bundle AAA games and/or drop prices come Q3 2016.

It's a perfectly logical and rational view... assuming AMD were in a profit-making position and not *desperate* to regain share. A 25% share loss in two years in any industry is nothing short of an epic catastrophe and should result in very desperate measures. I still think you're probably right, but I think there's a higher probability than you that they actually show up with a 36 (near 1070) or 40 cu ( above 1070 ) part at the $300 point. Keep in mind also that for your view to be correct, it means SiSoft is misrepresenting the number of compute units.

Also, keep in mind that AMD has very much indicated they expect some aib to include g5x, which wouldn't make sense on a 32cu part.

 

[Orvogg]

While GP104 is 50% larger than Polaris 10, does anyone really think it will end up 50% faster? The people estimating P10 @ 290x performance are nuts to believe that. Vastly improved resources clocked 30% higher will make for a very interesting case.

When applying logic to the relationship between Hawaii and Polaris it is so hard not to become a believer, but I don't think anyone is better off if people were to start drumming up expectations.

I think you are right, though. If "all" AMD wanted was to achieve 290x performance at a far better price, they would have just shrunk Hawaii, kept the frequence and had a bit of overclocking headroom. But they didn't. They added transistors and they improved on the architecture...

BTW, AMD never said which Polaris chip would be at Hawaii-performance. They only stated that was one of the performance goals when designing the architecture.

"The reason Polaris is a big deal," continued Taylor, "is because I believe we will be able to grow that TAM significantly. [...] We're going on the record right now to say Polaris will expand the TAM. Full stop."

 

[swilli89]

Proof? Seriously, what basis do you have for this statement?

The lack of any information on their plans for mobile maybe? We have zero information about GP106/GP108 and release schedules/design wins.

AMD has drilled it in our heads over and over that P11 is laptop/mobile focused, we've seen an incredible improvement of performance per watt compared to the GTX 950 demonstrated 6 months ago, and we all know about the rumored design wins Polaris has already received, especially the Apple deal. All this being said while keeping in mind nVidia's best mobile chip right now, GM206 is undoubtedly the market volume leader and this is squarely where AMD is bringing the fight to nVidia.

I'd say that to show that nVidia chose to win and will most likely win in reality at the very top end, and AMD chose to win and most likely will win in mobile and midrange GPUs up to ~$300.

Actually, my statement is about the only thing we DO know decisively at this point. Laptop graphics is utterly dominated by power profiles unless you are talking about the desktop replacement market. Desktop graphics is an entirely different story but I don't know why you were shocked about my statement?

The ONLY concrete Pascal mobile news we have? A random Asus desktop replacement laptop that requires LIQUID COOLING to run. Talk about niche.
http://wccftech.com/nvidia-gtx-1080-m-pascal-mobility-flagship-inbound/

Polaris will dominate Q3 and possibly Q4 design wins in mobile IF like i said, the market reacts to pure performance and not marketing/backroom deals.

I say all of this after a full generation of me recommending my friends ONLY buy GTX950 or GTX960 laptops, they were the only choice if you wanted to have an actual laptop. This generation will be Polaris until we see something from nVidia that tells us otherwise.

 

[Qwertilot]

Can we be sane here?

It is a vanishingly unlikely proposition that NV won't get a die shrunk mobile GPU range out for the big Autumn/xmas schedule. Profit wise its a big priority for them and they're executing basically brilliantly in these sorts of terms nowadays.

Its considerably more likely that AMD might get a few months to themselves in the mainstream dGPU market than that, although they do have to launch soon to make sure of that.

 

[Glo.]

The lack of any information on their plans for mobile maybe? We have zero information about GP106/GP108 and release schedules/design wins.

AMD has drilled it in our heads over and over that P11 is laptop/mobile focused, we've seen an incredible improvement of performance per watt compared to the GTX 950 demonstrated 6 months ago, and we all know about the rumored design wins Polaris has already received, especially the Apple deal. All this being said while keeping in mind nVidia's best mobile chip right now, GM206 is undoubtedly the market volume leader and this is squarely where AMD is bringing the fight to nVidia.

I'd say that to show that nVidia chose to win and will most likely win in reality at the very top end, and AMD chose to win and most likely will win in mobile and midrange GPUs up to ~$300.

Actually, my statement is about the only thing we DO know decisively at this point. Laptop graphics is utterly dominated by power profiles unless you are talking about the desktop replacement market. Desktop graphics is an entirely different story but I don't know why you were shocked about my statement?

The ONLY concrete Pascal mobile news we have? A random Asus desktop replacement laptop that requires LIQUID COOLING to run. Talk about niche.
http://wccftech.com/nvidia-gtx-1080-m-pascal-mobility-flagship-inbound/

Polaris will dominate Q3 and possibly Q4 design wins in mobile IF like i said, the market reacts to pure performance and not marketing/backroom deals.

I say all of this after a full generation of me recommending my friends ONLY buy GTX950 or GTX960 laptops, they were the only choice if you wanted to have an actual laptop. This generation will be Polaris until we see something from nVidia that tells us otherwise.

IF AMD's Mainstream chip will be able to tie with Nvidia's High-End chip - ONLY then there is a reason to clap and be astonished.

Don't get your hopes too high, without understanding whole context.

 

[swilli89]

When applying logic to the relationship between Hawaii and Polaris it is so hard not to become a believer, but I don't think anyone is better off if people were to start drumming up expectations.

I think you are right, though. If "all" AMD wanted was to achieve 290x performance at a far better price, they would have just shrunk Hawaii, kept the frequence and had a bit of overclocking headroom. But they didn't. They added transistors and they improved on the architecture...

BTW, AMD never said which Polaris chip would be at Hawaii-performance. They only stated that was one of the performance goals when designing the architecture.

Yeah I think you are the right track. Hawaii was AMD's last well designed chip. It was designed for 28nm from the get-go. Its architecture was developed before AMD even had a foothold in consoles, much less the 100% dominance they enjoy today.

I can't imagine that if you take a Hawaii that has 2816 Stream Processors, and 64 ROPs, compare it to the soon to be released top end Polaris 10 @ 2304-2560 Stream Processors and the same 64 ROPs at a much higher frequency all with entirely new GCN 4.0 architecture that it will merely equal it in performance.

From the technical analysis I've seen, AMD's new architecture is an even more radical redesign that Kepler to Maxwell. Think about this, the 780ti had 2880 SPs compared to the 980's 2048 SPs yet the 980 is 30-60% faster and obviously can be much higher considering games where nVidia has neglected to update Kepler drivers (that's an entirely different point of course ())

So nVidia was able to shrink their die size, improve power efficiency, and almost double per SP performance all on the same node. If anyone doubts AMD can't do this as well they're kidding themselves. I've been constantly saying here for a few months that top end Polaris 10 will equal a 980ti/Fury X and beat them in some cases and I remain with that viewpoint. My guess is Polaris 10 will beat the 1070 in outright performance, be more affordable, all while using less power. Time will tell of course.

Can we be sane here?

It is a vanishingly unlikely proposition that NV won't get a die shrunk mobile GPU range out for the big Autumn/xmas schedule. Profit wise its a big priority for them and they're executing basically brilliantly in these sorts of terms nowadays.

Its considerably more likely that AMD might get a few months to themselves in the mainstream dGPU market than that, although they do have to launch soon to make sure of that.

Did you miss where I said "Polaris will dominate Q3 and possibly Q4 design wins in mobile". Of course I know nVidia will be bringing a smaller Pascal derived chip out, but we haven't heard a peep about it so I believe it might still be a ways off.

 

[Arachnotronic]

The lack of any information on their plans for mobile maybe? We have zero information about GP106/GP108 and release schedules/design wins.

Remember how on this forum when everyone was jumping up and down about how NVIDIA must be behind AMD in terms of FinFET because AMD was making loud announcements and NVIDIA wasn't? Remember how I told you that NVIDIA was keeping its mouth shut because it didn't want to obsolete all of the cards its partners were trying to sell in the channel?

Well, guess what, same thing applies here. In fact, it's even worse because once a whole system is obsolete, the system builder has to take a big loss on a complete computer rather than just a PCB with a GPU and some memory on it. So, of course NVIDIA isn't going to tell you when GP106/GP107/GP108 are coming because that would screw its partners over big time.

AMD has drilled it in our heads over and over that P11 is laptop/mobile focused, we've seen an incredible improvement of performance per watt compared to the GTX 950 demonstrated 6 months ago, and we all know about the rumored design wins Polaris has already received, especially the Apple deal. All this being said while keeping in mind nVidia's best mobile chip right now, GM206 is undoubtedly the market volume leader and this is squarely where AMD is bringing the fight to nVidia.

So what you're telling me is that NVIDIA correctly identified the market for power efficient notebook-oriented GPUs, made a ton of money selling such GPUs and gaining share against its rival, but this same company -- the one that was smart enough to do the above -- is just going to cede the entire market to AMD?

I'm sorry, but that's just ridiculous.

I'd say that to show that nVidia chose to win and will most likely win in reality at the very top end, and AMD chose to win and most likely will win in mobile and midrange GPUs up to ~$300.

Why can't NVIDIA dedicate its resources to trying to win in both? Do you really think that if NVIDIA could spin GTX 1080 and Tesla P100, it will have issues bringing out easier-to-build derivatives for lower-cost desktop GPUS and notebooks?

Polaris will dominate Q3 and possibly Q4 design wins in mobile IF like i said, the market reacts to pure performance and not marketing/backroom deals.

I say all of this after a full generation of me recommending my friends ONLY buy GTX950 or GTX960 laptops, they were the only choice if you wanted to have an actual laptop. This generation will be Polaris until we see something from nVidia that tells us otherwise.

This is completely baseless and assumes that NVIDIA is run by morons. I don't think that's a good assumption to make.

 

[ElFenix]

All this being said while keeping in mind nVidia's best mobile chip right now, GM206 is undoubtedly the market volume leader and this is squarely where AMD is bringing the fight to nVidia.

huh? pretty sure nvidia uses GM204 and GM107 as its main mobile parts, as far as gaming goes.

if AMD is smart (a stretch, but bear with me), there's a polaris variant that's an exact thermal replacement for each of those.

that's a big chunk of your gaming market right there.

Remember how on this forum when everyone was jumping up and down about how NVIDIA must be behind AMD in terms of FinFET because AMD was making loud announcements and NVIDIA wasn't? Remember how I told you that NVIDIA was keeping its mouth shut because it didn't want to obsolete all of the cards its partners were trying to sell in the channel?

people get far too excited over both "news" and "no news." both teams do it.

 

[swilli89]

So what you're telling me is that NVIDIA correctly identified the market for power efficient notebook-oriented GPUs, made a ton of money selling such GPUs and gaining share against its rival, but this same company -- the one that was smart enough to do the above -- is just going to cede the entire market to AMD?

Sure a lot of what I said is based on assumptions. I do NOT think nVidia is going to cede the entire market to AMD. However, this is all based on a feeling I have that AMD is going to win the performance/watt game this round, at least until Volta/Vega. I think NV had to play catchup in a lot of compute-oriented methods and that cost them on the efficiency side. We can surely both agree that whomever wins the efficiency game wins the mobile market.

So what you're telling me is that NVIDIA correctly identified the market for power efficient notebook-oriented GPUs, made a ton of money selling such GPUs and gaining share against its rival, but this same company -- the one that was smart enough to do the above -- is just going to cede the entire market to AMD?

This is completely baseless and assumes that NVIDIA is run by morons. I don't think that's a good assumption to make.

Don't put those words in my mouth. I'm merely stating that by the looks of it, AMD should do great in Q3 mobile design wins evidenced by what we know and what we think we know. I think we also know how battery life obsessed Apple is. They've shown that they will not spare cost when it comes to performance/design/efficiency and they are putting Polaris in pretty much any product they make which from this link includes 100% of their notebooks. To assume that Apple chose an inferior product for its highest margin mobile products is very misguided and to quote a certain arachnid friend of mine "I don't think that's a good assumption to make."
http://appleinsider.com/articles/16/04/19/apple-to-adopt-amds-new-polaris-graphics-chips-in-upcoming-macs---report

huh? pretty sure nvidia uses GM204 and GM107 as its main mobile parts, as far as gaming goes.

if AMD is smart (a stretch, but bear with me), there's a polaris variant that's an exact thermal replacement for each of those.

that's a big chunk of your gaming market right there.

I double checked and GM206 underpins both the GTX950 and the GTX960, easily the highest volume dGPU laptop products. You are indeed correct about the top end, GM204 powering the GTX980M but this again will be countered by a Polaris 10 in mobile.

 

[sirmo]

Can we be sane here?

It is a vanishingly unlikely proposition that NV won't get a die shrunk mobile GPU range out for the big Autumn/xmas schedule. Profit wise its a big priority for them and they're executing basically brilliantly in these sorts of terms nowadays.

Its considerably more likely that AMD might get a few months to themselves in the mainstream dGPU market than that, although they do have to launch soon to make sure of that.

I think this time around AMD have a possible advantage in mobile. It seems to me they have taken the IPC route instead of the brute force high clock approach Nvidia have taken. And they also enjoy a noticeable process advantage 14nm to 16nm. We already know that Apple's iPhone SoCs are more efficient on the Samsung's 14nm process compared to TSMCs 16nm.

I expect AMD to nail the mobile market this time around. Considering also like you said that they will be first to market.