AMD Polaris architecture – GCN 4.0

Mtom · 15 Mar 2016 at 11:12

humbug said:
You would think Polaris 10 at that size would be at least Fury X performance, 232mm at 14nm is at least equivalent to 500mm at 28nm.

The one in the demo was Polaris 10 and able to match Fury-X easily.

Yeah thats what i thought, it ran at Fury X speed, but i think the game was FPS locked to 60. I assume it could be faster than Fury X/980Ti.

muziqaz · 15 Mar 2016 at 11:52

I keep reading about how Polaris 10 is this or that size and Polaris 11 is a small Polaris.
Since I dozed off around half 12 last night right when the stream started to cut off, did I miss some slides or information?
Where do they say that Polaris 10 is larger than Polaris 11 (ok, Anandtech is claiming it). So when AMD demoed Polaris in January, I think it was Polaris 10 they demoed, wasn't it?
here: http://techreport.com/news/29616/amd-will-introduce-two-polaris-gpus-this-year

Koduri says that Polaris 10 is for thin and light. Is this again brainfarting from AMD?

Also last I checked any of todays GPUs are struggling to keep minimum 60fps at 1440p.

Mtom · 15 Mar 2016 at 12:02

So the one they demoed in January was Polaris 11, and yesterday Polaris 10. You are right about the 60FPS 1440P, its at least on level with the top dogs of 28nm.

mahius · 15 Mar 2016 at 12:12

Wait, what? Polaris 11 is actually small Polaris? So there won't be a Polaris that grinds current high end into the dust then? Well that's disappointing. In which case, doom and gloom. If Polaris 10 is AMD's big hitter and is only around Fury X performance... they better have some miracle pricing or it's going to be seriously disappointing. I do like AMD pushing the 1080p60 standard forward to 1440p60, but only if Polaris 10 is priced like a 390/390x. I guess it'll be a long while yet till 1440p120 is possible on a single GPU.

But from the sounds of it, Vega will be the cards we will be hoping for. Or at least if it really is late 2016/early 2017 that puts hopes up that AMD will price Polaris 10 in sucfh a way that Vega doesn't kill it when it launches. It'd be like launching the 980 at it's post-980ti price and then later launching the 980ti at the 980ti price. Whatever the case, it's probably in folk's best interests to wait it out for Vega unless Polaris is priced very well.

As I said before, Raja mentioned Moore's Law in the context of price-to-performance. Here's hoping it means the Polaris 10 will be priced well.

muziqaz · 15 Mar 2016 at 12:14

Mtom said:
So the one they demoed in January was Polaris 11, and yesterday Polaris 10. You are right about the 60FPS 1440P, its at least on level with the top dogs of 28nm.

So Koduri was brainfarting

Either way, I think if AMD wants to get some mindshare and marketshare with Polaris 10 release, GPU needs to be able to easily outperform reasonably clocked 980ti.
At this moment, seeing 60fps minimum at 1440p on DX12 game which was closely developed with AMD is not very impressive, since reasonably clocked 980ti gets just a tad less. And we know that 980ti has broken async shaders.
Now if Polaris 10 launches with some ridiculously slow power consumption and very high OCing potential then it is another story, but until then I am very cautious :/

humbug · 15 Mar 2016 at 12:16

Mtom said:
So the one they demoed in January was Polaris 11, and yesterday Polaris 10. You are right about the 60FPS 1440P, its at least on level with the top dogs of 28nm.

Right, i remember now.

Polaris 11: was about 120mm (480X at 390X+ Speed?)
Polaris 10: is 232mm (490X at Fury-X+ Speed?)
Vega: (Fury-X replacement at +##% Speed?)

muziqaz · 15 Mar 2016 at 12:19

JediFragger said:
And that's from the biggest AMD fanboy on the forum!!!

Must have been baaaaaaaaaaaad

It wasn't bad. If kids these days cannot be entertained, then it is their problem. The conference was quite interesting. A lot of new info on upcoming titles, also AMD is back in bed with Ubisoft, and Watchdogs 2 is dx12 titles and suppose AMD will be supporting this title (deja vu).
Total War Warhammer is dx12. I always thought that Total War games would benefit the most out of Vulkan/DX12, since it is very easy to kill highest end CPU with previous games when you have supersized army attacking the cities.
Frostbite 3 is going DX12.

muziqaz · 15 Mar 2016 at 12:21

Orangey said:
What is nexgen mem? HBM3? Some kind of graphics phase change variant?

You know AMD, they are always 5 steps ahead on memory tech

They finalised HBM1/2 so now they are already working on next gen memory tech

Jono8 · 15 Mar 2016 at 12:33

flopper said:
If I owned a 980ti I wouldnt buy Polaris for that reason.
I buy Polaris if it runs better for VR and it will with low latency that 980ti cant do.

GDc conference should not been streamed.
I watched it and was thinking, bored.
BORED AMDMatt I was friggin bored.

worst PR in history and its AMD engineers behind it.
send then to make the gpu and then hire people who can actually make it interesting to own the Polaris.

I can only do so much for AMD here

Twilight zone right there.

FoxEye · 15 Mar 2016 at 12:33

humbug said:
Right, i remember now.

Polaris 11: was about 120mm (480X at 390X+ Speed?)
Polaris 10: is 232mm (490X at Fury-X+ Speed?)
Vega: (Fury-X replacement at +##% Speed?)

If this is the case... they better be cheap. Like £150 max for a P11.

The 290X/390X has been with us since Oct 2013. If they price the P11 at £200+, they can shove it

Mtom · 15 Mar 2016 at 12:46

mahius said:
Wait, what? Polaris 11 is actually small Polaris? So there won't be a Polaris that grinds current high end into the dust then? Well that's disappointing. In which case, doom and gloom. If Polaris 10 is AMD's big hitter and is only around Fury X performance... they better have some miracle pricing or it's going to be seriously disappointing. I do like AMD pushing the 1080p60 standard forward to 1440p60, but only if Polaris 10 is priced like a 390/390x. I guess it'll be a long while yet till 1440p120 is possible on a single GPU.

But from the sounds of it, Vega will be the cards we will be hoping for. Or at least if it really is late 2016/early 2017 that puts hopes up that AMD will price Polaris 10 in sucfh a way that Vega doesn't kill it when it launches. It'd be like launching the 980 at it's post-980ti price and then later launching the 980ti at the 980ti price. Whatever the case, it's probably in folk's best interests to wait it out for Vega unless Polaris is priced very well.

As I said before, Raja mentioned Moore's Law in the context of price-to-performance. Here's hoping it means the Polaris 10 will be priced well.

Only change is the top polaris was changed to Vega.
We knew the top cards will come at late 2016 at best. Also both companies will price the high-midrange cards (Polaris10 - 1080) accordingly. They will beat the current top so they will start higher than current mid range, then they will reduce price as the big ones release.

Mtom · 15 Mar 2016 at 12:46

humbug said:
Right, i remember now.

Polaris 11: was about 120mm (480X at 390X+ Speed?)
Polaris 10: is 232mm (490X at Fury-X+ Speed?)
Vega: (Fury-X replacement at +##% Speed?)

Thats about it yep.

bru · 15 Mar 2016 at 12:52

Ok so forgetting all the mixups over the naming scheme, (have they really gone with the bigger number being the smaller chip). I get the feeling that AMD might have pulled a blinder here.

Here is the dilemma, you are going to launch a new set of GPU's but you don't want to show the opposition just how fast they are, (now most of us have expected the bigger chips from this first round of new GPU's to beat the currant 980ti/FuryX, just like I'm sure both AMD and NVidia have.) so you show a benchmark that is locked at 60fps, one that the currant top tier cards can just about manage to max out and voila you have shown off your new card, shown that it is at least as fast as the currant crop, without actually showing just how fast it actually is.

It could be a genius move by AMD, but then returning to the naming stuff, when you have two of your top people saying different things about which card is which, maybe AMD just blindly stumbled into this benchmarking piece of genius by mistake.

Mtom · 15 Mar 2016 at 12:58

bru said:
Ok so forgetting all the mixups over the naming scheme, (have they really gone with the bigger number being the smaller chip). I get the feeling that AMD might have pulled a blinder here.

Here is the dilemma, you are going to launch a new set of GPU's but you don't want to show the opposition just how fast they are, (now most of us have expected the bigger chips from this first round of new GPU's to beat the currant 980ti/FuryX, just like I'm sure both AMD and NVidia have.) so you show a benchmark that is locked at 60fps, one that the currant top tier cards can just about manage to max out and voila you have shown off your new card, shown that it is at least as fast as the currant crop, without actually showing just how fast it actually is.

It could be a genius move by AMD, but then returning to the naming stuff, when you have two of your top people saying different things about which card is which, maybe AMD just blindly stumbled into this benchmarking piece of genius by mistake.

Yeah i thought about that, they may spinning the names and performance figures on purpose.

bru · 15 Mar 2016 at 12:59

CAT-THE-FIFTH said:
Hmm,so that indicates two families - Polaris will use GDDR5 and Vega,HBM2??

Vega appears to be either late 2016 or early 2017.

Orangey said:
What is nexgen mem? HBM3? Some kind of graphics phase change variant?

Just a guess, but I reckon that show us that Polaris will be small/mid size chip GDDR5, Vega will be large chip HBM2, Navi will be next gen (maybe even GCN5) small/mid size chip GDDR5X.
I certainly don't know of any other new memory technologies that are soon to be available/in development, also bear in mind that these chips take years to design, so any work for new memory types must be worked into the chips right at the beginning of the design phase.

CAT-THE-FIFTH · 15 Mar 2016 at 13:05

Interestingly Vega is meant to have better performance per watt than Polaris? Since it is the larger chip it needs to replace Hawaii for professional markets so has stuff dedicated towards DP performance.

JediFragger · 15 Mar 2016 at 13:27

bru said:
maybe AMD just blindly stumbled into this benchmarking piece of genius by mistake.

They sure couldn't have planned it!! Their PR Dept can barely find their own ass with both hands on the best of days

queamin · 15 Mar 2016 at 14:00

JediFragger said:
They sure couldn't have planned it!! Their PR Dept can barely find their own ass with both hands on the best of days

Maybe they hired Nvidia PR to do it

LoadsaMoney · 15 Mar 2016 at 14:40

CAT-THE-FIFTH said:
Looking at the chart I posted and the demo they made lets assume Polaris 10 has roughly R9 390X level performance but is a 150w TDP card instead of a 275w TDP one.

We can't assume that though, as the Polaris 10 is faster than the Fury X, which in turn is faster than the 390X.

Orangey · 15 Mar 2016 at 15:02

Another lean year for enthusiasts.