• Hey Guest. Check out your NeoGAF Wrapped 2025 results here!

AMD's Fermi Done Right(?) | GCN™ - Graphics Core Next |

A

artist

Banned
AMD released details of their upcoming GPU arch at the Fusion Developer Summit few days back.

In a nutshell, AMD is ditching their VLIW approach for MIMD+SIMD. This is basically to address the compute performance while trying to make simplify the scheduling part of their GPU and effectively get max utilization of the core. For example the peak SP & DP of AMD GPUs are quite high but when it comes to real world app's compute performance the advantage is not really there, folding for example; http://i.imgur.com/rjHDs.png

Plus, they announced more new features such as revamped memory caches, OS virtualization, tools etc.

Anyway, here's the goodies.

X2mIB.png


Official AMD Presentation

PCPer: http://www.pcper.com/reviews/Graphi...ecture-Overview-Southern-Isle-GPUs-and-Beyond

Anandtech: http://www.anandtech.com/show/4455/amds-graphics-core-next-preview-amd-architects-for-compute

Few goodies from the above articles;

Memory Cache
It also opens up some new advantages to traditional graphics. “Megatextures” which will not fit on a card’s frame buffer can be stored in virtual memory. While not as fast as onboard, it is still far faster than loading up the texture from the hard drive. This should allow for more seamless worlds. I’m sure John Carmack is quite excited about this technology.
Click to expand...

Graphic's Performance
Graphics performance is still the primary goal of this new architecture. There will still be plenty of fixed function units which have not changed in ages. ROPS and Z-units will remain, and it is quite likely their numbers will grow with each shrink, thereby allowing more rendering power to push pixels to the screen. With performance sinks like Eyefinity and 3D, pixel fillrates are still very important.
Click to expand...

Physics, AI, Tessellation for iGPU
One scenario discussed was that of physics acceleration. Instead of the dGPU doing both rendering and physics/compute work, the iGPU on the CPU would handle those. The iGPU would have the advantage of being located on the CPU, sharing the same memory controller, and accessing the main memory very quickly, as well as greater memory localization of the data. This would reduce latency by a significant degree as compared to the dGPU doing the same thing over the PCI-E bus. By taking care of this business, the dGPU would better handle other operations such as geometry and pixel shading or tessellation.
Click to expand...

Ofcourse AMD didnt show all their cards; how many CUs (compute units), clocks, release time-frame etc. PCPer thinks that we'll see new GPUs based on this arch (may not have all these new features) this year while Anandtech's opinion is that we'll see products using this arch in early 2012.
 
Microsoft has to be tied into these technologies, but they were right there ready with their AMP presentation loud and proud.
 
Well at least you put the ? in your headline. Aside from birthing pains, I don't really see what was wrong with Fermi. Given AMD's lackluster performance with Bulldozer, they should consider themselves lucky if a new architecture switch only gives them issues in the first gen of cards.
 
Now look who creepin’ look who crawlin’ still balling in the mix
It’s that six six long dick Slim ni**a stick in yo’ chick
Pullin’ tricks, lookin’ slick and on times when I’m flippin’
Bar sippin’, car dippin’
Grand wood grain grippin’
Still tippin’ on four 4’s
Wrapped in vouges
Pimpin’ four hoes
And I’m packing four 4’s
Blowin’ on that endo
Gamecube Nintendo
5% tint, so you can’t see up in my window
These niggaz don’t understand me cuz I’m Boss Hogg on candy
Top down at Maxi’s wit’ a big glock nine handy
Pieced up creased up, staying dressed to impress
Big Boss belt buckle under my Mitchell ‘N’ Ness
Oh, Gucci shades up on my braids when I Escalade
When I’m riding Sprewells sliding like a Escapade
I got it made the Big Boss of the north
Ain’t shit changed I still represent Swisha House(Ha!)
 
You don't make a thread about graphics and have GCN in the title without folks thinking about Nintendo recently. :lol

That said, pretty hyped for what this means for AMD graphics cards going forward. :)
 
elrechazao said:
serious question - does this mean I should delay the build I have planned for year end to 2012 sometime?
Click to expand...

yes, the motto for PC gaming is "always wait"(tm).

:) nah man...do it up now and enjoy yourself. there will always be something better. just buy a real nice case and a solid PSU and the rest you can upgrade as new tech comes out or when you want more juice.
 
sk3tch said:
yes, the motto for PC gaming is "always wait"(tm).

:) nah man...do it up now and enjoy yourself. there will always be something better.
Click to expand...
Well, I was gonna build around new year, so potentially 2012. If this is coming out say, a month later, that's a bit different than the normal "everything is always better in the future" rule of upgrading.
 
elrechazao said:
Well, I was gonna build around new year, so potentially 2012. If this is coming out say, a month later, that's a bit different than the normal "everything is always better in the future" rule of upgrading.
Click to expand...

you'll know more information about the roadmap as you get closer to your actual purchase date. none of these big gfx card releases are stealth...there's always a good bit of warning.
 
elrechazao said:
serious question - does this mean I should delay the build I have planned for year end to 2012 sometime?
Click to expand...

This isn't coming out anytime soon in PC years and is mainly gpu computing stuff, gaming wise you'll probably get the same type of jump you get from a new gen.
 
elrechazao said:
Well, I was gonna build around new year, so potentially 2012. If this is coming out say, a month later, that's a bit different than the normal "everything is always better in the future" rule of upgrading.
Click to expand...
If theirs a new cpu socket standard incoming I would wait till after that, worst thing that happens when doing a new build is getting in at the end of socket cycle and having to get another new motherboard in a couple of years if you want to upgrade further.
 
1-D_FTW said:
Well at least you put the ? in your headline. Aside from birthing pains, I don't really see what was wrong with Fermi. Given AMD's lackluster performance with Bulldozer, they should consider themselves lucky if a new architecture switch only gives them issues in the first gen of cards.
Click to expand...

hehe .. from Anandtech's conclusion

Moving on, it’s interesting that GCN effectively affirms most of NVIDIA’s architectural changes with Fermi. GCN is all about creating a GPU good for graphics and good for computing purposes; Unified addressing, C++ capabilities, ECC, etc were all features NVIDIA introduced with Fermi more than a year ago to bring about their own compute architecture. I don’t believe there’s ever been a question whether NVIDIA was “right”, but the question has been whether it’s time to devote so much engineering effort and die space on technologies that benefit compute as opposed to putting in more graphics units. With NVIDIA and now AMD doing compute-optimized GPUs, clearly the time is quickly approaching if it’s not already here.
Click to expand...

Fermi <3

anyway I'll be getting the fastest single-gpu card that either company is willing to release next year regardless of power usage :D if I keep gaming on 1080p such a card should last quite some time
 
Yeah, this is all computational. There will obviously be an improvement with more shader units and higher clock speeds, but other than that it's not a revolutionary jump or anything. No mention of DX11.1 or DX12 either makes me think of this as more of a refresh.
 
It's not a refresh in the same sense that Fermi wasn't a refresh of the previous architecture. Fermi was faster than it's predecessors but not by a whole lot, but what Fermi added was important for those supercomputers that use Nvidia GPUs and that sort of thing. GPGPU is something that both Nvidia and ATI are anticipating will be big this decade, as CPUs are increasingly hitting a single-thread performance wall and there are many types of tasks that throwing more cores at will make no difference. But if you bring in a GPU as a kind of floating-point super-coprocessor, you can dramatically multiply performance gains in many kinds of compute tasks. That's what CUDA was about (and Adobe Creative Suite 5 uses CUDA!), that's what C++ AMP is about.
 
Izayoi said:
Unlikely. We'll see at least DX12 and SM6/7 in the 720.
Click to expand...

Depends on when the PS4/720 officially have locked hardware specs. Remember that specs are locked at least 1-2 years before a console starts mass-production. DX12 is supposed to be in Windows 8, right? That's going to be around 2012/13, so if 720's specs are locked before Win8, it won't have DX12.
 
Unknown Soldier said:
Depends on when the PS4/720 officially have locked hardware specs. Remember that specs are locked at least 1-2 years before a console starts mass-production. DX12 is supposed to be in Windows 8, right? That's going to be around 2012/13, so if 720's specs are locked before Win8, it won't have DX12.
Click to expand...
My money is still on new consoles being released in 2015, so 2013/14 would be the years that specs got locked. Even a 2014 release could leave the possibility of DX12, or at least pseudo-support (like the 360) where Microsoft works with ATI and integrates some yet-to-be-released features into the GPU.
 
Unknown Soldier said:
Depends on when the PS4/720 officially have locked hardware specs. Remember that specs are locked at least 1-2 years before a console starts mass-production. DX12 is supposed to be in Windows 8, right? That's going to be around 2012/13, so if 720's specs are locked before Win8, it won't have DX12.
Click to expand...

Both microsoft consoles had parts that were ahead of PC at release

NV2a (Xbox GPU) was a dual shader pipeline unit... GeForce 3 line never had dual pipeline (which is what nvidias flagship at the time)

Xenos? xenon? (Xbox 360 GPU) was, at the time, a radical architecture change going from fixed shaders to ALU/streams

they were both really powerful for the time, and in both cases took upwards of a year to have a PC comparable

edit: while i'm saying that yes, it's locked far in advance... they just have a significantly higher R&D budget because of the scope of the processors needed. heavy R&D then pays off for their other releases as they fine tune
 
sk3tch said:
Now look who creepin’ look who crawlin’ still balling in the mix
It’s that six six long dick Slim ni**a stick in yo’ chick
Pullin’ tricks, lookin’ slick and on times when I’m flippin’
Bar sippin’, car dippin’
Grand wood grain grippin’
Still tippin’ on four 4’s
Wrapped in vouges
Pimpin’ four hoes
And I’m packing four 4’s
Blowin’ on that endo
Gamecube Nintendo
5% tint, so you can’t see up in my window
These niggaz don’t understand me cuz I’m Boss Hogg on candy
Top down at Maxi’s wit’ a big glock nine handy
Pieced up creased up, staying dressed to impress
Big Boss belt buckle under my Mitchell ‘N’ Ness
Oh, Gucci shades up on my braids when I Escalade
When I’m riding Sprewells sliding like a Escapade
I got it made the Big Boss of the north
Ain’t shit changed I still represent Swisha House(Ha!)
Click to expand...

This is that shit right here, *daps*
 
Southern Isles isn't a simple 28nm shrink of the Northern Islands? By jolly!
 
The interesting thing is their roadmap, which aims to dissolve the barriers between the CPU and the on-die GPU. These are bold moves, which will require OS-level support to work properly. I'm not even sure the shared GPU virtual memory and GPU context switching (aka: GPU compute multitasking) can be implemented in XP drivers.

duk said:
dx12? i wish, but i think we'll see dx11++
Click to expand...
It'll take much longer for SM 6.0 to arrive. SM 5.0 added so much flexibility the current hardware still cannot fully utilize it (performance-wise) and it's the 2nd generation.

SM 5.1, however, is very possible to show up with these new cards to standardize the new features AMD is coming up with.
 
DopeyFish said:
Both microsoft consoles had parts that were ahead of PC at release

NV2a (Xbox GPU) was a dual shader pipeline unit... GeForce 3 line never had dual pipeline (which is what nvidias flagship at the time)

Xenos? xenon? (Xbox 360 GPU) was, at the time, a radical architecture change going from fixed shaders to ALU/streams

they were both really powerful for the time, and in both cases took upwards of a year to have a PC comparable

edit: while i'm saying that yes, it's locked far in advance... they just have a significantly higher R&D budget because of the scope of the processors needed. heavy R&D then pays off for their other releases as they fine tune
Click to expand...
NV2A was roughly the same with the flagship at the time, overtaken in 3 months by GF4
Xenos had a unified pipeline but was still weaker than the flagship at the time
RSX was clearly inferior to the flagship at the time

I think the fundamental archs was not radically different except for Xenos even which had a "lead" because ATI goofed up so badly with R600 that it got delayed for ages.
 
You must log in or register to reply here.

Similar threads

  • Poll
Will PS6/Xbox next finally offer true next-gen graphics?
2 3 4 5 6
264
3K
Baemono replied
F-Zero X and F-Zero DD Decomps Almost 100% Done! PC Port Next?
Retro Trailer Indie 
14
391
cartman414 replied
Intel and AMD next gen CPU delay to 2027
2
97
2K
Gameplay Gods Bless replied
Former Rockstar graphics programmer led next-gen procedural breakable glass system for GTA VI
Rumor Game Dev 
24
644
genesisblock replied
What is happening to consumer Graphics Cards?
26
124
existensmaximum replied
Top Bottom