• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

AMD Polaris architecture – GCN 4.0

AMD has been a lot clearer when it comes to showing us where the improvements to their new pascal architecture are, with AMD clearly showing that they have redesigned their Graphics cores, their Geometry units, their caching system and have improved the design of their memory controller.

lol
 
AMD is not competing with 1080, so they are not in a rush to release polaris.
the question is, do they have the performance to compete with 1070, if they do, they will release polaris before june10 at least on paper with full disclosed infos.
if by june 10 we still dont have more info about polaris 10, then performance is too low to be competing with 1070.
 
The dynamic load balancing image from DP's link pretty much says it all when it comes to Pascal - quite limited in terms of concurrent compute capabilities with nothing like the broadness of AMD's approach - but Pascal better able to bring GPU resources to the task at hand when the graphics queue has cleared compared to Maxwell.

Judging from the apparent changes AMD are making with Polaris compared to Fiji, etc. it looks like they've realised that real world optimal use of async lies somewhere between the 2 different approaches employed by nVidia and AMD.

As i understand it, and may be wrong.

Nvidia already use Preemption in Maxwell, Preemption and Scheduling are the same thing, instruction sets are organised in a way that avoids idle threading by pretiming call instructions to interleave.

AMD are adding Instruction scheduling to Polaris, ASynchronous shaders will be carried forward from CGN 1.1 and 1.2.

Preemption or Scheduling is used to reduce draw call overheads in DX11 and can be used in DX12.
AMD did not use Scheduling as it can introduce a latency caused by holding intrusions in a stack, which is necessary to organise said instructions.

In AMD's GPU's the instructions flow directly without stacking, the downside of that is a reduced flow of calls compared with Scheduling so it is less efficient, the up side is improved latency.

Cryengine actually has the same scheduling system in its extraction layer, this is why its able to run DX11 so beautifully balanced across upto 16 threads, which is why AMD's 8 core FX CPU's so unusually compared with almost all other games run's Crysis 3 ever so slightly better than a 3770K.
Nothing to do with AMD's partnership with Crytek in making the engine for Crysis 3 :p

Anyway...

AMD's solution was to have multiple schedulers in the hardware, 8 in GCN 1.1 and 1.2, those ACE units.

The problem is its a bit like 8 core CPU's in GPU form and DX11 cannot handle this.
Mantle was the first API to be capable of running multiple shader threads.
DX12 has the same architecture, but of course its not at all borrowed from Mantle, its just a coincidence :D

Nvidia have proven how useful Schduling is in DX11 and AMD will now do the same for DX11 while for DX12 the ASynchronous hardware remains the same.

Pascal is the same as Maxwell, the difference is the ASynchronous capabilities derived from Scheduling have been switched on for Pascal

AlamoX posted the video that explains this nicely.

 
Last edited:
We could really do with some leakages of info from AMD on P10/11 and it might help with stopping sales to NVidia.

Polaris is not going to compete with the 1080, maybe the 1070, but that one isn't going to be out till June 10th, which is after Computex when AMD will hopefully release more info.
 
8 days before more solid info comes along the way.
once AMD is ready to release.

So far it seems the end of May AMD thing in Macau will be closed for the press, they will tell them information about Polaris, but it will be under NDA until the launch...which will be E3 as it seems.
 
Greg, i think Nvidias pricing fiasco of the 1080 is doing enough to stop Nvidia sales, not completely however as theres a lot of people with more money than actual sense still lol :)

I have been ragging on AMD for a good while now, but something makes me think in all this silence and refusal to leak any solid info on actual target of the cards and performance figures, they might just "Drop the Mic" with this launch and leave a lot of people with egg on their faces and a lot of others just gobsmacked.
 
Hitman is an interesting one. When you use the developers own canned benchmark AMD comfortably win, even in DX11 which raises eyebrows. When you test independently from the AMD sponsored developers benchmark things look quite different. http://www.pcgameshardware.de/Hitma...Episode-2-Test-Benchmarks-DirectX-12-1193618/

I wonder how many threads we would have had on this if the results were reversed and this was a NV sponsored title.

Thanks for the link, that had completely passed me by.

I understand what you mean about people having different standards for Nvidia and for AMD, but AMD do need a bit of help at the moment. ;) Especially if Polaris doesn't manage to get out in September after all.
 
The problem is abosultely pure and simple.
Marketing marketing marketing.
It's what sells cards and it's where AMD is behind Nvidia. (My job is Marketing and Branding so I have some experience).

Everyone knows that on paper the 390 is better than the 970. And even pricing it LESS than the 970. The 970 goes on to be the best selling card in the world...
Doesn't speak very highly of us consumers does it?

Sheep following shiny glossy marketing. That's all it is.

AMD's Polaris could be twice as good as Pascal but if they don't market it right it still won't beat Nvidia's sales.
 
What Layte forgot to mention in posting it yet again is that the review that shows AMD win the canned benchmark was reference AMD cards against reference Nvidia cards. That site used 3rd party cards with better coolers, lower temps higher stock clocks and they were then further overclocked.

AMD won the reference vs reference scores, Nvidia won the AMD reference vs Nvidia 3rd party cooler heavily overclocked scores..... which isn't entirely surprising.
 
Greg, i think Nvidias pricing fiasco of the 1080 is doing enough to stop Nvidia sales, not completely however as theres a lot of people with more money than actual sense still lol :)

I have been ragging on AMD for a good while now, but something makes me think in all this silence and refusal to leak any solid info on actual target of the cards and performance figures, they might just "Drop the Mic" with this launch and leave a lot of people with egg on their faces and a lot of others just gobsmacked.

Yer, quite agreed about the pricing of the 1080 FE card. Bit of a steep swallow that is :(

As for P10/11, the silence has me worried the other way. I remember the 290X launch and tid bits here and there and AMD bashing on the Titan launch doors and telling the punters to come this way (I am sure it was the 290X launch), but not much info being leaked at all and only snippets of cometing with the 950 on power and running 60 fps on Hitman at 1440P (but not divulging at what settings). I am hoping I am wrong though and you are right.
 
The problem is abosultely pure and simple.
Marketing marketing marketing.
It's what sells cards and it's where AMD is behind Nvidia. (My job is Marketing and Branding so I have some experience).

Everyone knows that on paper the 390 is better than the 970. And even pricing it LESS than the 970. The 970 goes on to be the best selling card in the world...
Doesn't speak very highly of us consumers does it?

Sheep following shiny glossy marketing. That's all it is.

AMD's tech is fine. They just need to up their marketing game.

Where does all this stuff come from anyway, 970 best selling card ever? it sold more than a low end card, I call complete and utter bull.

Second people get confused and think retail pricing at say OCUK is the same as Dell buying units by the 100k's direct from Nvidia or a AIB and such deals can be signed and done long before the cards actually get into computers and sold.

So at the time 390x wasn't as cheap or maybe not even out and Dell need a midrange cheap gaming card and the 970 is the actual best option at that point in time. At that time Dell create 3-4 different computers that use the 970 and sign a deal to get 2million over the next year at 15% cheaper then the supplier, lets say Gigabyte, was selling that model for individually. 3 months later the 390x is both faster and cheaper, but that initial deal is still done and happening.

You can't look at pricing today and these entirely silly statements about being the best selling GPU ever, and decide that it's all marketing. The market isn't as simple as that, deals done 2 years ago effect sales numbers today. Playing off different companies to get a good volume deal effects what cards a company chooses to put into computers.

Home upgrade sales, retailers like OCUK/newegg in the states, whoever else, they account for a tiny fraction of all GPU sales.
 
As i understand it, and may be wrong.

Nvidia already use Preemption in Maxwell, Preemption and Scheduling are the same thing, instruction sets are organised in a way that avoids idle threading by pretiming call instructions to interleave.

AMD are adding Instruction scheduling to Polaris, ASynchronous shaders will be carried forward from CGN 1.1 and 1.2.

Preemption or Scheduling is used to reduce draw call overheads in DX11 and can be used in DX12.
AMD did not use Scheduling as it can introduce a latency caused by holding intrusions in a stack, which is necessary to organise said instructions.

In AMD's GPU's the instructions flow directly without stacking, the downside of that is a reduced flow of calls compared with Scheduling so it is less efficient, the up side is improved latency.

Cryengine actually has the same scheduling system in its extraction layer, this is why its able to run DX11 so beautifully balanced across upto 16 threads, which is why AMD's 8 core FX CPU's so unusually compared with almost all other games run's Crysis 3 ever so slightly better than a 3770K.
Nothing to do with AMD's partnership with Crytek in making the engine for Crysis 3 :p

Anyway...

AMD's solution was to have multiple schedulers in the hardware, 8 in GCN 1.1 and 1.2, those ACE units.

The problem is its a bit like 8 core CPU's in GPU form and DX11 cannot handle this.
Mantle was the first API to be capable of running multiple shader threads.
DX12 has the same architecture, but of course its not at all borrowed from Mantle, its just a coincidence :D

Nvidia have proven how useful Schduling is in DX11 and AMD will now do the same for DX11 while for DX12 the ASynchronous hardware remains the same.

Pascal is the same as Maxwell, the difference is the ASynchronous capabilities derived from Scheduling have been switched on for Pascal

AlamoX posted the video that explains this nicely.


So are you now going to agree that I have been right all along?
Or you just going to admit you have changed your opinion now?
 
Yer, quite agreed about the pricing of the 1080 FE card. Bit of a steep swallow that is :(

As for P10/11, the silence has me worried the other way. I remember the 290X launch and tid bits here and there and AMD bashing on the Titan launch doors and telling the punters to come this way (I am sure it was the 290X launch), but not much info being leaked at all and only snippets of cometing with the 950 on power and running 60 fps on Hitman at 1440P (but not divulging at what settings). I am hoping I am wrong though and you are right.

Every launch is different. There was loads of information floating around about Fury X and the hype was massive. It turned out to be disappointing. The 4870 on the other hand was more like this and was a nice surprise. So far in my experience the more hype the more disappointment. The gtx980/ti and gtx970 just about came with no hype and we're decent cards.
 
To compete with the gm204 and to save r&d amd respun hawaii, rebadged it as grenada and managed to achieve a 10-15% improvement in driver overhead.
Effectively amd had improved their shader throughput per engine by improving their sceduling efficency. I predicted concerns with fiji and my theories were posted way before most people started looking into its performance problems. fiji is a perfect example of a front end problem, which leads me onto my final point.

It's worth mentioning the improved memory speed and bandwidth being a contributing factor to grenada's performance as well as just driver improvements.

When buying the 290s, there was a vram lottery which in the worst case caused some of the blackscreen issues we saw initially.
 
Back
Top Bottom