Tuesday, April 26th 2016

AMD Polaris 10 "Ellesmere" as Fast as GTX 980 Ti: Rumor

At a presser in Taiwan for its Radeon Pro Duo launch, AMD talked extensively about its upcoming "Polaris" and "Vega" family of GPUs. The company appears to be betting heavily on two SKUs it plans to launch this June, Polaris 10 and Polaris 11. Polaris 10 is an internal designation to Radeon R9 490(X), based on the 14 nm "Ellesmere" silicon. It may be the biggest chip AMD builds on the "Polaris" architecture, but it won't exactly be a "big chip," in that it doesn't succeed "Fiji." That honor is reserved for "Vega," which debuts in early-2017.

The "Ellesmere" silicon is more of AMD's competitor to NVIDIA's GP104. It is rumored that the R9 490(X), based on this silicon, will offer consumers performance rivaling the GeForce GTX 980 Ti (ergo faster than the Radeon R9 Fury X), at a USD $300-ish price point. "Ellesmere" will be a lean-machine, physically featuring up to 2,560 4th generation GCN stream processors (2,304 enabled on Polaris 10), a possible 256-bit GDDR5X memory interface, and a deep sub-200W typical board power rating.

Source: GameDebate
Add your own comment

91 Comments on AMD Polaris 10 "Ellesmere" as Fast as GTX 980 Ti: Rumor

#1
Outlander_04
bug said:
It's basically about what margins they would have at $300.
Chips at 14/16nm will be much smaller, thus much cheaper than they are at 28nm. And it seems we won't get a big performance boost to go along, thus saving should go mostly into costs/price.
Intels 14 nm is not cheaper than the previous 22 nm despite size reductions . The manufacturing process and possibly yields mean the cost per die is marginally more than the previous generation [ unlike every other die shrink ]
So lets assume that Samsung/GF/AMD face the same challenge . Cost will be about the same as the previous generation , but performance will be greater . Probably they will open with a price hike to make current stock seem worth while. My guess is $340 US for the R9 490 . Add another $50 to that for the 480X . Still undercutting the 980ti by about $150
Posted on Reply
#2
G33k2Fr34k
efikkan said:
That would be an enourmous achievement. GCN is currently way behind Maxwell, and Pascal is a major architectural overhaul while 4th generation GCN is not. Keep in mind that Fiji is just performing close to GM200, even though it has 50% more theoretical performance.

A 200mm² chip is not going to be hig-end, and even AMD says they're targeting mainstream. You'll have to wait at least until Vega before AMD is even trying, if ever again..
First off, we don't know how these two chips perform. Pascal is just a marketing name. All Nvidia's chips since Fermi have had very similar micro-architectural designs.

Perhaps AMD managed to find major bottleneck/bottlenecks in their earlier "GCN" designs and fixed them with Polaris. We don't know. What we know is that they claim to have improved "shader efficiency", which means more performance with the same amount of processing power or SP count. They also claim to have improved Tessellation performance, which is a major weakness of AMD cards.
The same goes for Nvidia, we don't how well their new chips will perform. We need to wait to see.

Back when the 4000 and 200 series cards were released, the 4870 was position by AMD against the GTX980. Well... it turned out that the 4870 could wrestle with the GTX280 and beat it in some games.
http://techreport.com/review/14990/amd-radeon-hd-4870-graphics-processor
Posted on Reply
#3
efikkan
G33k2Fr34k said:
First off, we don't know how these two chips perform. Pascal is just a marketing name. All Nvidia's chips since Fermi have had very similar micro-architectural designs.
Not true at all, we know from GP100 that Pascal is a huge step forward in CUDA code performance, the biggest in may years. AMD on the other side has stated that larger changes in GCN has to wait until 2017 because Zen is a priority.

G33k2Fr34k said:

What we know is that they claim to have improved "shader efficiency", which means more performance with the same amount of processing power or SP count.
To keep up with Nvidia's improvements AMD would need to more than double it's performance, in addition to the shrink itself.
Posted on Reply
#4
medi01
FordGT90Concept said:
AMD will be shooting themselves in the foot if they launch at $300. NVIDIA will be forced to drop their prices to compete. Good for consumers but AMD can't exactly afford that.
Remember AMD 5xxx times?
Slower, more power hungry and more expensive NV chips still kept more than 50% of the market share.

=[

Assimilator said:
I should make some slides showing something that people desperately want to be true, then release them to websites that do no fact-checking.
They demoed a game though. So "980 Ti" levels comes from there.
I'm not talking about PS4k chip, though.

GigabyteFanBoy said:
There in no way shape or form do i believe that a "PS4.5" could be anywhere near the level of a 980ti
It's "kinda confirmed" that GPU speed will more than double. So it comes down to how do you rate 7870 in the "optimized for this card" context.
Posted on Reply
#5
Outlander_04
efikkan said:
Not true at all, we know from GP100 that Pascal is a huge step forward in CUDA code performance, the biggest in may years. AMD on the other side has stated that larger changes in GCN has to wait until 2017 because Zen is a priority.


To keep up with Nvidia's improvements AMD would need to more than double it's performance, in addition to the shrink itself.
efikkan said:
Not true at all, we know from GP100 that Pascal is a huge step forward in CUDA code performance, the biggest in may years. AMD on the other side has stated that larger changes in GCN has to wait until 2017 because Zen is a priority.


To keep up with Nvidia's improvements AMD would need to more than double it's performance, in addition to the shrink itself.
I am completely baffled by your post . To keep up with nVidia AMD would have to what?
Buy an R9 390 or a GTX 970 . The AMD is marginally cheaper . The two swap the lead in games at 1080p , but the R9 is better at higher resolutions [ usually ] . And the Radeon architecture looks to be the better option for DX 12 [ in the very few examples available ]
How does the AMD need to be twice as good?

What we do know is that polaris is using way less power than current generation nvidia graphics cards at similar performance levels . That is as you would expect . 14nm vs 28 nm .

As for which of the new architectures will perform better per watt and per dollar is anyone's guess
Way too early to call.
Posted on Reply
#6
efikkan
Outlander_04 said:
I am completely baffled by your post . To keep up with nVidia AMD would have to what?
...
How does the AMD need to be twice as good?
Because right now AMD needs over 50% more theoretical performance and a much larger chip to "compete", like Fury X compared to GTX 980 Ti, R9 390(X) to GTX 970/980 and so on. This is the primary reason why AMD is struggling with it's energy consumption. Pascal is the most significant improvement in a decade on top of that, so AMD would need to catch up both with both the Maxwell improvements and Pascal in one sweep, with 4th generation GCN which is only a minor architectural tweak. You are allowed to have dreams, so dream on.

Outlander_04 said:
And the Radeon architecture looks to be the better option for DX 12 [ in the very few examples available ]
As mentioned before most people forget that Nvidia brought the driver side changes of Direct3D 12 to all APIs, which is the reason why people think Nvidia does not support Direct3D 12 well.

Outlander_04 said:

What we do know is that polaris is using way less power than current generation nvidia graphics cards at similar performance levels . That is as you would expect . 14nm vs 28 nm.
Have you forgot that Nvidia is also doing a node shrink? Any improvement AMD gets from the new node Nvidia will get as well, and Nvidia is bringing major architectural changes as well.
Posted on Reply
#7
Outlander_04
efikkan said:
Because right now AMD needs over 50% more theoretical performance and a much larger chip to "compete", like Fury X compared to GTX 980 Ti, R9 390(X) to GTX 970/980 and so on. This is the primary reason why AMD is struggling with it's energy consumption. Pascal is the most significant improvement in a decade on top of that, so AMD would need to catch up both with both the Maxwell improvements and Pascal in one sweep, with 4th generation GCN which is only a minor architectural tweak. You are allowed to have dreams, so dream on.


As mentioned before most people forget that Nvidia brought the driver side changes of Direct3D 12 to all APIs, which is the reason why people think Nvidia does not support Direct3D 12 well.


Have you forgot that Nvidia is also doing a node shrink? Any improvement AMD gets from the new node Nvidia will get as well, and Nvidia is bringing major architectural changes as well.
Again your post baffles me.
Its almost like you have abandoned logic and reason to tell us how much better you think one brand is compared to the other.

So I will repeat a few points
1/ In the very few DX12 benchmarks available the existing radeon architecture is well out performing the current nvidia chips . This might be a trend , or it might be just a few results that do not indicate a trend .
2/ AMD say much lower power consumption , and much higher performance per watt . We can judge who has the most power efficient architecture AFTER next gen cards are reviewed and not before .
3/ It looks like nvidia is opening their 16 nm manufacture with high end cards that will be sold in very small number . AMD have aimed squarely at the performance enthusiast market where they hope to get the highest turn over and market penetration .
4/My "dreams" dont affect hardware performance . But then neither do yours so its best we wait for some actual benchmarks
Posted on Reply
#8
rruff
Outlander_04 said:
Again your post baffles me.
I think the point he is trying to make is that Nvidia got a lot more performance and energy efficiency out of 28nm because of better architecture. Now that both are going to smaller processes, there is every reason to believe that relationship will continue. Nvidia spends a lot more on R&D. Even if AMD defies the odds with Polaris and Vega and produces an architecture that is as good as Nvidia's, it will be difficult for them to sustain it. It would be going from highly unlikely to a miracle. More likely AMD will need to sell less efficient chips at lower profit margins to make them performance competitive, just as they have been.
Posted on Reply
#9
Outlander_04
rruff said:
I think the point he is trying to make is that Nvidia got a lot more performance and energy efficiency out of 28nm because of better architecture. Now that both are going to smaller processes, there is every reason to believe that relationship will continue. Nvidia spends a lot more on R&D. Even if AMD defies the odds with Polaris and Vega and produces an architecture that is as good as Nvidia's, it will be difficult for them to sustain it. It would be going from highly unlikely to a miracle. More likely AMD will need to sell less efficient chips at lower profit margins to make them performance competitive, just as they have been.
Its an interesting point but one without too much merit . AMD are not just running a die shrink so its a spurious argument .
Secondly if the margins seen in DX 12 apply to next generation cards as well then you could turn down AMD clock speed 20% cutting power consumption by more than 50% and still be competitive with nvidias architecture . This is also a spurious argument .
Arguably the 14nm finfet process is superior to TSMC's 16 nm . But that also might not matter .

Lets wait and find out after the cards are released
Posted on Reply
#10
Steevo
efikkan said:
Because right now AMD needs over 50% more theoretical performance and a much larger chip to "compete", like Fury X compared to GTX 980 Ti, R9 390(X) to GTX 970/980 and so on. This is the primary reason why AMD is struggling with it's energy consumption. Pascal is the most significant improvement in a decade on top of that, so AMD would need to catch up both with both the Maxwell improvements and Pascal in one sweep, with 4th generation GCN which is only a minor architectural tweak. You are allowed to have dreams, so dream on.


As mentioned before most people forget that Nvidia brought the driver side changes of Direct3D 12 to all APIs, which is the reason why people think Nvidia does not support Direct3D 12 well.


Have you forgot that Nvidia is also doing a node shrink? Any improvement AMD gets from the new node Nvidia will get as well, and Nvidia is bringing major architectural changes as well.
AMD kept the actual DX12 hardware in, plus some DP hardware that makes heat, and their waaaaayyy outdated video output logic tied to the core/memory speed has been their other issue except they render video better due to its configuration.


So when we want to look at apple to apples power consumption, lets look at why its more on many chips, and the issues faced by Nvidia with multiple displays and corruption, poor video playback, and TDR issues with power switching.
Posted on Reply
#11
G33k2Fr34k
efikkan said:
Not true at all, we know from GP100 that Pascal is a huge step forward in CUDA code performance, the biggest in may years. AMD on the other side has stated that larger changes in GCN has to wait until 2017 because Zen is a priority.


To keep up with Nvidia's improvements AMD would need to more than double it's performance, in addition to the shrink itself.
I don't think you're being objective here mate. Despite being smaller, Polaris10 packs 7-8 Billion transistors, which is almost as many as the GP104 packs. Also, both GP104 and Polaris10 are rumored to have the same number of FP32 units, which is 31/32 * 2560 = 2480. Hawaii had 2880 * 3/4 = 2112 FP32 units (I believe). And so ideally both Polaris10 and GP104 should perform about the same at similar clock speeds.
Posted on Reply
#12
Slizzo
G33k2Fr34k said:
I don't think you're being objective here mate. Despite being smaller, Polaris10 packs 7-8 Billion transistors, which is almost as many as the GP104 packs. Also, both GP104 and Polaris10 are rumored to have the same number of FP32 units, which is 31/32 * 2560 = 2480. Hawaii had 2880 * 3/4 = 2112 FP32 units (I believe). And so ideally both Polaris10 and GP104 should perform about the same at similar clock speeds.
Except, you can't compare the cores like that. Take a look at FuryX core makeup vs. that of GTX 980Ti.
Posted on Reply
#13
G33k2Fr34k
Slizzo said:
Except, you can't compare the cores like that. Take a look at FuryX core makeup vs. that of GTX 980Ti.
Posted on Reply
#14
Fluffmeister
Ultimately I just feel sorry for Fury X, soo much hype, such a big fat let down.
Posted on Reply
#15
Slizzo
G33k2Fr34k said:

There's always improvements to efficiency in the GPUs released time after time. Still, you cannot directly compare the GPUs.

As I was alluding to, Fury X had 4,000+ cores vs. 980Ti with 3,000+ cores. They performed similarly. To think that the new Pascal architecture wouldn't also improve shader efficiency is foolish.
Posted on Reply
#16
TheHunter
On another forum this news is titled as "nearly as fast as 980Ti."


Which is Imo more accurate estimation.

This is just around AMD 390x and Nano or 980GTX perf... Same with GP104 Pascal between 980 and 980Ti..
Posted on Reply
Add your own comment