I think Vega 10 will be AMD's top GPU and will be positioned as a competitor to the 1080 Ti. I think Vega 11 will be the GPU that AMD uses to try to compete with the 1070/1080.
Honestly, I think we are all confused about your argument here. Apparently one set of rumors should be believed above the other set of rumors?
Edited.I think Vega 10 will be AMD's top GPU and will be positioned as a competitor to the 1080 Ti. I think Vega 11 will be the GPU that AMD uses to try to compete with the 1070/1080.
I will really like to know why you believe the 64CU die will be the large one.
AMD doesn't have the HPC contracts to justify a die like GP100. They don't want to get burnt like they did with Fiji. It's probably about the size of GP102.
I think vega 10&11 should be much faster than the 1080/titan X. They will be released like 9 months later.
I think anyone who needs that level of performance will have already bought a 1080/titanX?
http://www.newegg.ca/Product/Product.aspx?Item=N82E16814500398&cm_re=gtx_1080-_-14-500-398-_-ProductThese
Vega cards will be competing with 1080/titanX refreshes in my opinion. mabe even Volta.
GP102 die is 470mm2. I believe a 64CU die could be smaller than 400mm2 and no way to be able to compete against GP102.
That's one way to look at it, but another way is to bring higher level of performance to more reasonable price levels -- in a way AMD brought $650 GTX780-class beating performance to a far more reasonably priced $399 R9 290. Right now, the cheapest good 1080 is about $629.99+, with many GTX1080 cards hovering in the $650-700 USD range. Even if AMD has some Vega chip that only matches GTX1080, but the card is priced at $499-549, that would be a huge win for consumers because we'd get a cheaper card with superior DX12 performance.
We'll have to see how the future unfolds but AMD has had extremely competitive offerings during HD4000, 5000, 6000, 7000 and R9 290/390 generations. It's only the Fury generation that can be considered the first major flop on their end. The last 2 generations, for anyone who keeps their GPUs longer than 18 months, AMD clearly won not only on price/performance but also on high-end performance:
HD7970/Ghz > 680
R9 280X > 770
R9 290 > 780
R9 290X > 780Ti
R9 295X2 > 780Ti SLI
Today, the high-end Hawaii R9 290X/390X trades blows with GTX980 and sometimes approaches 980Ti in DX12 games where 780Ti is beaten by an R9 290. Ouch. That's another point many are downright dismissing -- Vega may launch later but at least it will be a proper DX12 architecture whose GPUs can be used for the next 3-4 years.
I don't doubt that NV will retain the GPU performance crown but for 99% of PC gamers not on 1440 144Hz, 3440x1440 or 4K monitors, these GPUs are largely irrelevant if NV charges $800 USD for 1080Ti and $1200 for Titan XP.
If AMD just undercuts GTX1070 by $50 and 1080 by $100, those NV cards won't be worth purchasing due to lack of a proper DX12 hardware/architecture; and NV's lackluster driver support for GTX600/700 series. Since modern GPU generations last roughly 2 years, AMD will have at least a year to sell high-end cards. Ironically, as we've seen during HD4890, 6970, 7970/7970Ghz, R9 290X and Fury generations, AMD's high-end cards are rarely purchased by high-end consumers. Therefore, even if Vega are great GPUs, they will hardly be popular among enthusiasts who as you said either already purchased Pascal, are waiting to replace 980Ti with 1080Ti, and/or are by now locked into G-Sync anyway.
Where AMD needs to concentrate with new GPUs is for newcomers to PC gamers who don't have inherent biases with GPU purchases. I would guest-estimate that a lot of people on these boards are veteran PC gamers, at least in their early or mid-30s and older. There are also millions of young PC gamers between 16-28 who are either interested in 4K gaming/VR, may just be may be entering PC gaming for the first time or building their 2nd PC. That's where Vega comes in. By 2017, the current gen consoles will start to look very aged and usually towards the end of the console generation, a lot of gamers decide to jump ship to PC (hence why Sony is trying to prevent that with PS4 Pro).
Depends on a gamer's backlog and current pricing in their country. Newegg Canada sells the cheapest decent AIB 1080 for $925 CAD after taxes. We are talking almost $1,000 CAD for an upper-mid-range Pascal chip. Not exactly enticing to purchase considering that today a $200 USD GTX1060 matches or beats a barely 2-year-old $550 USD 980. The era of 'needing' $1,000 CAD GPUs is pretty much over. PC gaming today is nothing like the old days where if you had a high-end card, your PC gaming experience was dramatically better. Back in the days it was a difference between running the latest games and not. Today, a basic $250 GPU such as RX 480/1060 is going to be a great gaming GPU for the vast majority of PC gamers running 1080p 60 fps or lower.
Consumer Volta has never been on NV's roadmap for 2017 but we continue to read how Volta is coming in 2017. If 1080Ti launches in 2017, why would NV canibalize their gross margins by releasing a $450 Volta card that obsoletes the $800+ 1080Ti?
680 March 2012 -> 980 Sept 2014 = 2.5 years
980 Sept 2014 -> 1080 May 2016 = but with real world retail availability approaching close to 2 years
If NV has 1080 Volta successors next year, it's unlikely to launch before November 2017 at the earliest, but chances are NV will wait until 2018 to launch Volta because they can continue to milk Pascal for all its worth. NV can easily drop the price of GTX1080 to $499-549 and not even sweat it.
The other thing is some of the most anticipated titles released in 2014-2016 run well on a $250 RX 480/1060 6GB GPU at 1080p and lower resolution. That means the dGPU market for $500-1500 GPUs is extremely small. What AMD really is a refreshed RX 485 that convincingly beats the 1060 6GB, a more solid card between RX 460 and 470, shift RX480 8GB to $199 price level and introduce a good GTX1070 competitor. Anything above that quickly hits diminishing returns (i.e., because NV loyal buyers won't buy NV, while the fraction of brand agnostic/AMD consumers who are willing to spend $500 USD+ on a card isn't that large). Probably one reason AMD is even making Vega is so that they can shrink all those chips later to 7nm as next generation's mid-range cards and for brand image. If AMD has faster high-end cards, it will drive the sales of lower end RX 400 series.
ANd about the refreshes?
GP102 die is 470mm2. I believe a 64CU die could be smaller than 400mm2 and no way to be able to compete against GP102.
Even with less resources, AMD had not 1, but 3 cards superior to the 960 in the same generation: R9 280X / R9 380 / R9 380X >>> 960.
Haha. Tonga is blatantly inferior on every single or nearly every single technical metric vs. the 960. The only thing Tonga had going for it was that it was cheaper and had higher performance (because it was much much larger and used more power).
Tonga was 359 mm^2. The 960 was 227 mm^2. Tonga is 57% larger and uses about the same amount more power than the 960. Despite having >50% more resources Tonga is nowhere close to 50% faster than the 960 on average. In DX12 games it approaches >50% faster but with the resources Tonga has it should be consistently >50% faster.
The only reason Tonga is 'good' is that it competed very well on price.
(The same goes for the 980 and how people moan how the 390X is better at DX12. The 390X is a little big larger and uses significantly more power - if it isn't stronger than the 980 then the architecture has problems).
People who bought 680 780 980 didnt get their moneys worth and got far worse perf/ $ than they could have. Nv shareholders did get the customers $ and say haha.Haha. Tonga is blatantly inferior on every single or nearly every single technical metric vs. the 960. The only thing Tonga had going for it was that it was cheaper and had higher performance (because it was much much larger and used more power).
Tonga was 359 mm^2. The 960 was 227 mm^2. Tonga is 57% larger and uses about the same amount more power than the 960. Despite having >50% more resources Tonga is nowhere close to 50% faster than the 960 on average. In DX12 games it approaches >50% faster but with the resources Tonga has it should be consistently >50% faster.
The only reason Tonga is 'good' is that it competed very well on price.
(The same goes for the 980 and how people moan how the 390X is better at DX12. The 390X is a little big larger and uses significantly more power - if it isn't stronger than the 980 then the architecture has problems).
Pretty much what the average buyer wants. Did you really try to spin these as inconsequential attributes?Haha. Tonga is blatantly inferior on every single or nearly every single technical metric vs. the 960. The only thing Tonga had going for it was that it was cheaper and had higher performance (because it was much much larger and used more power).
Tonga was 359 mm^2. The 960 was 227 mm^2. Tonga is 57% larger and uses about the same amount more power than the 960. Despite having >50% more resources Tonga is nowhere close to 50% faster than the 960 on average. In DX12 games it approaches >50% faster but with the resources Tonga has it should be consistently >50% faster.
The only reason Tonga is 'good' is that it competed very well on price.
(The same goes for the 980 and how people moan how the 390X is better at DX12. The 390X is a little big larger and uses significantly more power - if it isn't stronger than the 980 then the architecture has problems).
so we're back to arguing that end performance doesn't matter as much as the technical guts, or as much as the price? I guess when your team needs to win everything, we pick and choose which categories for which specific products are relevant to that specific discussion.
Pretty much what the average buyer wants. Did you really try to spin these as inconsequential attributes?
No wonder these forums are so barren lately.
I think it's more that nVidia is deliberately holding GPU performance back to maximize profit. If AMD was more competitive nVidia could easily make new models to counter.
Slightly faster than 1080 isn't good enough since HBM2 is so expensive.
NV have still alot room with GTX1070/1080 refreshes.Ya, what about them? You make it sound like there is no point in AMD releasing anything basically. I don't get your point.
On these very forums, not long ago, I already read how AMD will never achieve > 20% dGPU market share, I heard AMD's GPUs were finished during HD2000/3000 dGPU era, I heard how NV "stomped" AMD when OG Titan and 780 came out. All of these have been proven wrong. Now I am hearing non-sense how AMD left the high-end but anyone who did research and had objectivity knew that RX 470/480 were R9 380/X replacements. This means AMD still had to launch a Hawaii replacement and then a Fury/X replacement.
RX 480 is 64% faster than R9 380 at 2560x1440.
https://www.techpowerup.com/reviews/MSI/GTX_1070_Gaming_Z/26.html
RX 480 is a direct class / ~ price segment replacement of R9 380.
If AMD replaces R9 390X with a card 64% faster, it would end up just 1% behind the 1080.
Considering AMD is fighting a 2-front war against NV/Intel, it's amazing they are still churning out fast products. Imagine if NV had to design an x86 CPU and have dGPU? No one wants to give AMD credit for still competing despite having fewer of everything!
Even with less resources, AMD had not 1, but 3 cards superior to the 960 in the same generation: R9 280X / R9 380 / R9 380X >>> 960. Based on that, I wouldn't count AMD out at all, especially since NV is now releasing severely cut down flagships. Unlike Maxwell, Pascal barely has 12-13% overclocking headroom. 980Ti's overclocking is what made is a clear winner over the Fury X but this time Pascal's overclocking headroom is nowhere near last gen's.
But in that argument that I was replying to, AMD's competitive products at that generation, as evinced by RS, were performing better and were cheaper than nVidia's offerings. Did nVidia not consider that competition at the time? I can see the argument that for them, it really didn't matter with their marketing and command of the market: if you can get away with charging more for less performance, then of course you would.
so we're back to arguing that end performance doesn't matter as much as the technical guts, or as much as the price? I guess when your team needs to win everything, we pick and choose which categories for which specific products are relevant to that specific discussion.
People who bought 680 780 980 didnt get their moneys worth and got far worse perf/ $ than they could have. Nv shareholders did get the customers $ and say haha.
Pretty much what the average buyer wants. Did you really try to spin these as inconsequential attributes?
No wonder these forums are so barren lately.
AMD's products being competitive for the consumer and AMD's products being competitive for AMD are two completely different things.
AMD's products have been very competitive for the consumer but extremely uncompetitive for AMD the company.
It seems people took my post the wrong way. It was pretty obvious that any criticism of AMDs lack of R&D funds showing, despite being wholeheartedly true was going to get this response.
My point is this. AMD is not winning on technical merit right now (similar to how nvidia's Fermi did not win on technical merit) despite what is being said about DX12. They are simply getting by with larger dies and higher power consumption than the competition and selling at lower prices (and absolutely killing their company doing so. Any sane AMD supporter should not support prices that AMD cannot profit off of).
This thread is about performance on the high end. However, at the high end die size becomes are very real issue as does power consumption. If you are at huge perf/mm^2 and perf/W deficits vs. your competition you are not going to be competitive in terms of absolute performance when you start running into the limits of die size or maximum power use.
Being competitive on the low end (and losing money doing so) does not imply that it is possible to be competitive on the high end.
AMD is flat out noncompetitive in notebooks and has been since Maxwell launched because of the lack of technical merit in the architecture. Pasal in notebooks roflstomped everything on the market. This is a big problem, a serious problem for the bottom line of the company, and all this forum seems to want to do is go " power consumption doesn't matter".
This forum is simply not realizing that being "cheaper and faster than the competition" is not an end goal that a company strives towards. Companies strive towards making profits and AMD simply hasn't been making money (sufficiently in the black) for years. AMD selling you a "cheaper and faster card" is not something AMD wants to do, it is something they must do or they go out of business. In other words, the products are just NOT good enough to sell at anything resembling their desired margins.
AMD is doing OK at the moment not through any degree of individual merit but because their competitors view them as basically inconsequential and don't want to deal with the headache of being a monopoly.
AMD's products being competitive for the consumer and AMD's products being competitive for AMD are two completely different things.
AMD's products have been very competitive for the consumer but extremely uncompetitive for AMD the company.