thelastword
Banned
R490 and R480 will be P10. P11 will be used on R470 and lower since they're smaller parts at 128bit with less power draw.
AMD Polaris architecture to succeed Graphics Core Next
- Thread starter wachie
- Start date
Frankfurter
Member
P10 is rumored to be R480 only, P11 R460. Vega 10/11 is rumored to be R490 and R470.
Edit: + honestly it'd be pretty unprecedented to use one chip for 4+ GPUs which are potentially in the $200-500 price range.
That would be super fucking dumb because they would get destroyed by Nvidia on every single chart until vega releases. If find it hard to believe AMD won't release something to compete with 1070/1080 this year.
Frankfurter
Member
Well, it's confirmed that they won't. Polaris 10 and 11 will release this year, Vega 10 and 11 next year. Chances are that Polaris 10 and the slowest GP104 GPU will be roughly in the same performance field, though.
Nvidia will be in a similar situation, as they won't have anything new to compete with R480 until their smaller chip releases.
dgrdsv
Member
Well, you will likely be able to get a $300 card which will be able to run VR in 2016. P11 cards should be in the $300 range I think.Well, if you are right, then I am going to have to face a huge dilemma when it comes to choosing a successor for my HD7770I was hoping to be able to get a ~299$ card in 2016 that will be able to run VR stuff eventually. Guess I am stuck waiting one more year
So it'll be _four_ cards on P10? 480, 480X, 490, 490X? I find this rather unlikely.
From recent leaks I see two possibilities:
A. P10 is 490 and 490X, performance level covering current 390X/Fury-non-X.
P11 is 480 and 480X, performance in line with current 380X/390.
B. P10 is 480 and 480X, performing around 390/390X.
P11 is 470 and 470X, performing around 380/380X.
Which is Aug-Sep.
They may aim for the low-mid section and OEMs. Afaik there is more money in it.
Might be wise to not directly compete with NVIDIA.
Bolivar687
Banned
Where?
thelastword
Banned
Polaris is also R490 from the leaks and writeups I've seen. Pretty much any R400 branded GPU will be polaris 10/11. Now Vega on the other hand is different and will be branded differently (probably R500 series).
Looking at the performance of 480 and 480x, I can't see how these gpu's will be polaris 11, I dont think 50 watts and a 128bit bus is enough to push the performance that we've seen of the 480 class GPU's.
THE:MILKMAN
Member
If P11 470 is the replacement for the R9 380 then it would be an epic efficiency upgrade. Going from ~190W TDP/TBP to <50W for similar performance sounds too good to be true!
Frankfurter
Member
The chip is simply too small, it's estimated to be ~200-250mm².
thelastword said:
How would they do that? Polaris 10 and 11 are only 2 chips and both are relatively small (~200-250mm² and ~130mm² respectively).
Irobot82
Member
Where did the information on Polaris's chip size come from? We've seen alleged images of GP104 for Nvidia but I don't remember seeing any Polaris chips themselves.
Sir Abacus
Member
Because that worked so well with CPUs /s.
Locuza
Member
From linkedin where an AMD employee stated to work on a 14nm GPU project F, which was 232 mm² big.
So people assume that this might be P10.
http://www.guru3d.com/news-story/amd-polaris-project-f-gpu-surfaces.html
Moreover there was a chip with 36 CUs found on sisoft:
http://videocardz.com/58639/amd-polaris-10-gpu-specifications-leaked
The AMD Polaris Linux driver at least pretty much confirms that P10 is indeed equipped with 256-Bit.
40 CUs coupled with 256-Bit Interface seem to fit well in 232mm² and being under GP104.
SneakyStephan
Banned
For my upgrade, I'm looking for the best 1080p bang for buck upgrade on a 660. I'd like to go 1060 but if AMD can give me something awesome for cheaper, I'm all ears.
Irobot82
Member
GP104 is 1070/1080 right? So P10 will compete with Nvidia on the same level? What's everyone freaking out about?
THE:MILKMAN
Member
AMD have their Q1 earnings call later today, could more details about Polaris come from that?
Probably not but thought I'd ask.
Probably not but thought I'd ask.
Locuza
Member
If every assumption till now is right then they are not directly on the same level.
GP104 should target a bit higher.
Like in the past the CEO might be more precise about the timeframe, giving an update and maybe telling us something about the upcoming design-wins (consoles for example).
dgrdsv
Member
Well, I haven't said that. I said that P11's performance is either 380-380X or 380X-390. 380/380X is Tonga which is 5B transistors and 360mm^2 on 28nm. A quick estimation would put the same 5B transistors chip on 14LPE at ~150-180mm^2 which seems rather in line for what is expected from P11. Depending on how good it will clock and how big the efficiency improvements will be P11 can either substitute Tonga directly or be able to reach the lower tier of Hawaii (390). 128-bit bus is an issue, yes, but if they'll use GDDR5X while improving their memory management again it's not too far fetched to expect such P11 to be on 380X+ performance level.
P10 seems to be a 14nm version of Hawaii (440mm^2 @28nm -> 200-220mm^2 @14nm) and considering the above it should be able to reach Fury's performance level - if only in less bandwidth intensive situations.
GP104 seems to be 300+ mm^2 while P10 seems to be ~230mm^2 so they will probably end up in different performance tiers.
Bolivar687
Banned
Looks like these things are good to go (also some speculation on board numbers):
http://wccftech.com/amd-gcn-4-0-c99-flagship-polaris-rra-certification/
http://wccftech.com/amd-gcn-4-0-c99-flagship-polaris-rra-certification/
Irobot82
Member
This is getting all so confusing. I really hope AMD don't push rebrands.
icecold1983
Member
Thats unfortunate
dgrdsv
Member
I agree, this is starting to be confusing, especially as it's sound like P11 will be notebook exclusive which would be very odd.
A year ago "mainstream" positioning was where R7 cards are:
This seems very low for a P10 even with just 36 CUs. It would make sense if there would be faster chips above it meaning that they just pushed the performance in cheaper segment but since there aren't - I'm wondering if that's just an unfortunate choice of words? At the very least P10 should be where Hawaii is and that's "Performance".
Another interesting quote from the report:
Is it a mistake or did they revise the previously announced 2.5x figure?
icecold1983
Member
i guess we can pray for p10 competing with the lowest form of gp104. that could maaaaaybe be considered mainstream
chaosblade
Unconfirmed Member
So it sounds like AMD is launching lower/midrange stuff first. Performance rumors still don't sound too bad, but pricing will be key.
General Lee
Member
232mm^2 is decidedly mainstream. P10 is what 7870 used to be, the second highest tier back when it released. Hawaii brought an extra enthusiast tier in the high end for AMD, and that was replaced by Fiji. How these will relate in terms of performance will highly depend on how 14 nm clocks and how the memory bandwidth requirements work out.
My guess is AMD's total lineup will be:
Polaris 11 = value, Cape Verde
Polaris 10 = mainstream, Pitcairn
Vega 11 = high end, Tahiti
Vega 10 = enthusiast, Hawaii/Fiji
My guess is AMD's total lineup will be:
Polaris 11 = value, Cape Verde
Polaris 10 = mainstream, Pitcairn
Vega 11 = high end, Tahiti
Vega 10 = enthusiast, Hawaii/Fiji
Indeed. 24-26" size, all the syncs, not TN and ultra wide. Not TN being the real selling point. Mustn't far away from this, hey?
dgrdsv
Member
The problem with this guess is that Vega is 2017 and they don't have much above Hawaii's performance which can be used for high end and enthusiast segments - unless they plan to use Fury cards against GP104.
General Lee
Member
My guess is they'll let Nvidia have that 300 mm^2 chip be top end for the half year period before first Vega comes out. I still find the whole idea of GP104 being 300+ mm^2 and releasing before GDDR5X is available kinda problematic. Something about the equation doesn't add up. Either the release schedule is wrong, or the expectations for the chip are wrong, or they have some secret memory sauce to get 980 Ti+ performance with a 256-bit GDDR5 solution.
Irobot82
Member
It would seem very strange that one chip would be used on more than two GPU's. But maybe things are changing. I swear there is a quote somewhere where Raja said all new GPU's will be Polaris and that there would be no rebrands. Ugh. Computex can't come fast enough. AMD/Nvidia just gimme the FPS mang I'm looking to upgrade this year.
I don't really see any reason to doubt it.
http://wccftech.com/amd-greenland-gpu-32-gb-hbm2-memory-15-18-billion-transistors/
icecold1983
Member
Nvidia chips spend far more of a frame bound by areas other than bandwidth when not at 4k res which gp104 will def not be targeting anyway
The 390x has a TDP of 275W. If Polaris has a 2.5x perf/watt advantage over the 390x then that would mean a Polaris board with 150W would be equivalent to a 375W 390X, that suggests Fury level performance from the top Polaris 10 GPU at around 150W TDP.
A 75W Polaris 11 would be pretty close to 380x performance based on the 2.5x perf/watt improvement.
A 75W Polaris 11 would be pretty close to 380x performance based on the 2.5x perf/watt improvement.
D
Deleted member 465307
Unconfirmed Member
I hope we get some detailed rumors about Vega before the year is out. Looking into everything, it seems like I'm either going to be getting one of the top Polaris 10s or the lowest tier Vega. I'm currently leaning pretty strongly towards P10, but it'd be nice to know what I'm deciding against before making a decision.
Emphasis on performance per watt and notebook GPUs are probably sensible and relevant to help make AMD some money, but it doesn't sound great for anyone wanting something to go up against Nvidia this year. I would have liked to see what they can do with a new architecture and die shrink, not only to compare performance, but also to keep nvidia honest. If there is nothing competing for 6+ months, nvidia can go to town on pricing and then adjust later after creaming off lots of profit.
Also starting to really look like 'wait 6 months' is becoming 'wait 12 months' for people looking to enthusiast/high end cards and wanting to see what both companies are offering.
Also starting to really look like 'wait 6 months' is becoming 'wait 12 months' for people looking to enthusiast/high end cards and wanting to see what both companies are offering.
Well it depends on when NV launch a consumer GP100. AMD has always had better perf/mm, especially on new nodes, so if Polaris 10 is around 240mm^2 I can see it competing with GP104 even if GP104 is 300+mm^2.
I do see the high end gp104 being around 980Ti + 20% or so though so I expect it to be similar to the 2xx series vs the 4xxx series where AMD was about 90% of the performance at a lower power draw and around 60% of the price.
If NV get GP100 out and there is 0 competition expect a $1500 prosumer Titan card.
Inspectah_Deck
Member
Hmm, I´m really curious how pricing will be on the p10 cards, if they only offer performance on 390/x level with better perf-per-watt.
But I also can´t believe, that AMD gives Nvidia the 390/X, 970/80 market until Vega launches.
I guess we just have to wait and see.
But I also can´t believe, that AMD gives Nvidia the 390/X, 970/80 market until Vega launches.
I guess we just have to wait and see.
If Polaris 10 is 150 Watts and 232mm^2 like the current rumours indicate I can see Fury level performance for around $350-400, depending on NV pricing and performance obviously. The Nano is around $500 at the moment so it would offer a decent perf/$ improvement.
dgrdsv
Member
300mm^2 is hardly a top end though and the problem for AMD here will be that NV will be able to scale the GP104 below to any level basically if they need to. Sure, their margins will be lower but they'll block AMD from gaining market share this way. I'm kinda wondering if that's why we'll get three SKUs on GP104 in June.
I don't know why people are back to GDDR5X isn't available when Micron stated more than a month ago that they've started shipping the chips to customers and we even know what chips these are. The supply may be limited but it is available.
daninthemix
Member
So is Polaris 10 basically a PS4 Neo?
There is a chance they miniaturized 36CUs of 28nm and added HDMI2.0/better DVR support... but Polaris is most likely what they are using.
I'd love to be wrong, but I think expectations like these will only lead to disappointment. AMD would need to pull off some serious magic for a 40 CU chip to outperform 64 CUs. I know there'll be plenty of improvements overall and Fiji's scaling over Hawaii isn't the greatest, but that's still decent gap to overcome.
I'd say >~390X performance for $300-350.
Inspectah_Deck
Member
Is there any official word about this?
Because all I can find is this dubious hothardware.com article which all the sites that run the story link to.
No official word from Micron.
Is this site reputable?
Last official word was in february:
https://www.micron.com/about/blogs/2016/february/gddr5x-has-arrived
So they didn´t even announce sample dates yet (at least not officially).
It would be REALLY close for Teamred and green to have these chips in their cards launching in june.
I think we can expect cards with GDDR5X earliest near christmas.
dgrdsv
Member
Micron Begins to Sample GDDR5X Memory, Unveils Specs of Chips
Why would anybody wait for Christmas if the chips are in the catalogue and are sampling already?
Thing is there are no other uses for GDDR5X but in videocards and with Micron production started early this year it makes no sense to not have the new videocards use them.
But Summer 2016 clearly means June. Then there will be some offset until board manufacturers integrate all the components. It could definitely be autumn until GDDR5X is seen in retail products.
Inspectah_Deck
Member
Hmm, if adding to the catalogue == samples available to partners, then I give them the benefit of the doubt.
Similar threads
- Poll