AMD's 28nm Mobilty HD7000 Series - Coming 2012

Unfornately the demo doesn't do everything a game does, it is a cut scene, sure rendering some smoke and other effects on the fly in real time as the demo insists, but it didn't have AI running in the background, calculating their movements and drawing their animations, or typically any other taxing features not shown.

I will say it looked good , better than this generation of graphics for sure, but it still looks like a game, granted in the right artists and coders hands im sure they could slim the bloated code to nothing and run more on the same hardware while making it look amazing. But Im more particularly talking Infinite detail in textures, infinite polygons, more organic ai, and so on.

What im wondering is if anyone is balzy enough to go out and do it, would realistically destroy the GPU hardware market but im sure a royalty could be set for use of the engine or code or whatever it ends up being maxing out 3d graphics. Makes you wonder what Nvidea and AMD are really hiding in their secret layers.

 

Most of what you describes happens on the CPU and properly threaded code for native 4 or more cores should easily be able to handle it.

The infinite detail stuff is great, but it can't handle movement atm, which is a bit of a no no for games.

 

Ya linking voxels together and moving them in a mesh is a bit troublesome right now, alot fo the reasoning is because we bet on polygons and went full force into that technology, and in all honesty I would have taken that gamble now. A man can hope and dream can't he?? , I think what will happen is we will need to wait for quantum or biological processors to come along before we see a big leap in graphics, silicon is slowly hitting its limit, unless we figure something out with software.

 

I know 16nm is doable but it is very difficult. After that it's becoming increasingly more difficult to make things smaller and smaller. As far as I know, the limit is around 4nm. After which it's impossible to go any smaller. This limit should be reached around 2025.
22 nm — 2012
16 nm — 2014 (but I think 2015 is more likely)
11 nm — ???
8 nm — ???
6 nm — ???
4 nm — 2025? I think it was speculated but it might take up to 2030

From what I've read, 22nm is the last step which is easily done.

Here: http://en.wikipedia.org/wiki/16_nanometer

 

Makes me wonder if they would keep the same processor size now, at 40 nm lets say and just crame more of the smaller transistors on their? If that would make it leaps and bounds better.

 

And then we hit the singularity.

I'm sure rogue AI will have destroyed the human population somewhere in there.

 

It won't be just straight die shrinks...there will also be side steps to things like 3d circuits.

 

In all seriousness, past 22 mm, quantum effects start to dominate and classical circuit laws become less valid.

 

@ Phinagle Yes, I am thinking the same thing.
@ anexanhume Quantum tunneling FTW!

 

From anandtech's tri-gate article.

 

AMD [H]ardOCP GamExperience - July 16 - [H]ard|Forum

Could catch our first glimpses of HD7000 series hardware in just over a month....or maybe it's the HD6990M.

 

That's not next gen though.

 

I don't think I will be able to upgrade to an ATI/AMD card, so I guess I'll just have to wait to see what NVIDIA is going to counter this with....

Mr. Mysterious

 

That's true, a HD6990M wouldn't be next gen.

 

EDIT: Sorry )) old news.

 

Due to other interests occupying my time, I honestly haven't really cared much for keeping up the discussion on AMD's upcoming GPU arch, but I guess it's time to catch people up...

PC Perspective has probably the most comprehensive preview of what to expect from Southern Islands from last month's Fusion Developers' Summit AMD Fusion System Architecture Overview - Southern Isle GPUs and Beyond | PC Perspective

VLIW5 & 4 are gone outside Llano and Trinity APU....replaced by the Compute Unit and a completely redesigned memory architecture that will allow dGPU to not only better communicate with AMD Fusion APU but may also go so far as to allow AMD GPU to potentially appear alongside ARM processors.


Key quotes for the people that will end up only spreading the little bits of information they understand:

(In other words you shouldn't compare the 960 SPU on Blackcomb to the number of CU that will come on Wimbledon.)

More and more it's becoming necessary to include information on Fusion APU when talking about AMD's GPU in order to understand what their grand scheme is. The arch changes coming with Southern Islands will certainly bring more game FPS but they're also directly keyed to much more interesting changes in the future.


AMD talks about next generation software and Fusion | SemiAccurate

AMD, Intel, ARM, semiconductor, Fusion, FSA | Why AMD is opening up Fusion...

AFDS: AMD unveils FSA for OpenCL - BeHardware

AMD and ARM join forces at last | SemiAccurate

Farewell to DirectX? | bit-tech.net

 

amd has gone off the walls with the naming of their desktop/mobile gpus.

before a front number change meant its a brand new generation with superior performance and features. now....

anyways looking forward to see bulldozers/new gen gpu performance

 

Wait what? The only renames I know about are the 6770 desktop and 68x0 and 65x0 on the mobile side. Both were done for similar reasons.

 

apparently according to anandtech, there is the rebrand of the 5400 series as the 6300 series, now I dont see the 5500 series anywhere, since the 6400 series is new

 

and not be mention the fact that 5870 is faster than 6870

 

So, the single biggest architecture change in years for AMD GPUs coinciding with a process shrink? This should be interesting.

I guess the biggest question is if the arch change favors games or GPGPU applications.

 

2012 is the year of joy!

 

A process shrink and new silicon.

The biggest changes in the arch are to how GPGPU compute is handled but I don't see AMD's goals as favoring, or being limited to, either GPGPU or gaming.

The memory arch changes has potential to improve features like switchable graphics, and Crossfire, as well as open AMD's tech to new markets and/or other chip makers. Better still, FSAIL could help put an end to consoles holding back PC gaming...Microsoft's C++ AMP could lend a hand there too.

Right now I can only guess at what AMD will actually get done with their plans, but, since I see a bunch of opportunities, I can make a bunch of guesses.

2012 will be more like the year of good mirth...then 2013 the year of giddiness. We're probably going to have to wait until 2014 for the year of joy.

 

i actualy don't know if i wanna upgrade to amd's equivalent of fermi

with the load of problem they are gonna have and i don't know if i can wait till 2014 to upgrade

 

Oh the gaming should be fine, just dont buy it for the compute.

 

May laptop won't last until 2014. I can hold out until 2012.

 

that's why i have 2 gpu and with my QX landing anytime between now and when ever te post office decide it i could stand it till late 2013 on the performance side but i'm not sure the component will last that long ......

not sure about that they now use a completly diferent architecture meaning they have to write all new drivers

on the good side amd's dependancy on the CPU should go away with that new architecture ow and now i have to divide by 5 the number of SP i hope to have in my upgrade to have the equivalent

 

What is your setup capable of, in terms of Vantage GPU score?

Stock or overclocked, I don't care which.

 

most exciting thing about this post is the mxm 3.0 ... means my M18x will live for a while

 

about 9700 but heck that's low i hit 11000 in the past maybe i was oced

Edit something was wrong with my system running vantage now will post updated bench
erf nvm it seem it stay around that

 

I'm right there with you.

 

Meh, managing single units means the compiler can be far easier to write, rather than trying to find 5 way parallel executions.

 

AMD Expects to Be First GPU Designer with 28nm Graphics Chips - X-bit labs

Looks like AMD won't be using HKMG at 28nm for it's GPU as was generally expected. That's very disappointing. Originally I went along with the speculation that it would be Nvidia to use bulk silicon for 28nm to get a first to market win but it's going to be the exact opposite. AMD's going to go with the lesser silicon process while Nvidia is going to wait for the high performance HKMG 28nm to be ready.

Even if Nvidia's 600 series is another gianormous design, the advantage of being on HP HKMG would help it stay perf/watt competitive with a smaller more efficient design on bulk silicon.

 

But AMD/ATi is trying to beat Nvidia on release date, they are betting on doing the same they did with 5870s, some people say that AMD might still deliver this year, while most say Nvidia will go till Q2 of 2011, so this time difference might make up for the performance, at least thats what i think AMD is betting. Last i read about ATI switch, was for power power efficient design, i would think this will benefit a lot mobile users, so im looking forward to see it turns out to be, maybe Nvidia will be faster and more powerful, but they might not be able to crank them up in mobile market due to heat and power restrictions, who knows..... to much speculation.

 

sad if i on of the head of amd i would have gone 28nm bulk for first release like 7970 and stuf then use the last unused digit for a HKMG port like 7975

that way they would beat nvidia in speed and still give them an other slap to the face in q2

 

I had first read that it was going to be a blend of LP and HKMG processes, but that obviously isn't the case if they are spreading fab duties. Why they would want to cripple their performances and double their process anomalies to contend with for the sake of ramp rate is beyond me.

 

TSMC 28nm Portfolio

Southern Islands would be built around 28nm HPL according to s|a. How much this would impact performance & consumption is currently unknown.

 

Maybe they have confidence that their architecture is more power efficient than Nvidia's so that it will compensate for the inferior lithographic process.

 

I don't have a clear definition of why the HPL process is inferior to the HKMG one. HPL is a blend of their low power process and their HKMG process. The low power part means low leakage, but it also means lower peak frequencies. However, since most mobile GPUs max frequency are slower than the fastest mobile SoCs, that part isn't a problem.

Thus, I don't really get what makes this an 'inferior' process, it's just different. Maybe Phinagle knows more about why the process is inferior?

 

Seems nvidia is going to try to push their 28nm for this year,

NVIDIA promises to ship Kepler this year