Radeon HD 7000 Revealed: AMD to Mix GCN with VLIW4 & VLIW5 Architectures

[ajaidev]

"
Starting next week, AMD is going to organize Tech Days in several destinations around the globe, such as London or Paris - during which the company is going to present 28nm Radeon HD 7000 series.

There is a lot of rumors flying around the web, some of which are spun by AMD themselves to raise confusion, as the Radeon HD 7000 series is going to mix the existing VLIW4 and VLIW5 architectures with the "Graphics Core Next" (GCN), introduced during June's Fusion Development Summit held in Bellevue, WA.
"

http://www.brightsideofnews.com/news...itectures.aspx

yay

[EniGmA1987]

VLIW4 is better than VLIW5 isnt it? Something about the extra overhead or whatever.

[Sentential]

Also for the TL: DR Crowd:

Radeon HD7970XT:
2048 processing cores 1ghz
128 Texture Memory Units
3GB GDDR5 (384bit) @ 5.5ghz
MSRP $450

Radeon HD7970PRO:
1920 processing cores @ ???
120 Texture Memory Units
1.5GB GDDR5 (384/256bit) @ 5ghz
MSRP $350

For Comparison:

HD6970
1536 Stream Processors @ 880mhz
96 Texture Units
1GB GDDR5 (256bit) @ 5ghz


Advantages:
25% More Shaders
25% More Texture Units
12% Higher Clock Speed
33% Higher Memory Bandwidth

Honestly I had expected more, unless the efficiency is greatly improved it wont be the usual 50% faster than the previou gen, more like 35%

[iddqd]

^ Tahiti Pro @ 900 Mhz, and they want to launch them at the 2012 CES.

Any word on when the 7800 are coming out? A straight-up Cayman shrink shouldn't be too hard...

[AffenJack]

25% More Shaders
25% More Texture Units
12% Higher Clock Speed
33% Higher Memory Bandwidth

Honestly I had expected more, unless the efficiency is greatly improved it wont be the usual 50% faster than the previou gen, more like 35%

What are you calculating there? 7970XT against 6970 would be:
33% More Shaders
33% More Texture Units
13,6% Higher Clock Speed
65% Higher Memory Bandwidth
51% more Gflops

Add a bit higher Efficiency of GCN to it and we are at a normal 70% jump like from 4870 to 5870.

[tajoh111]

According to neliz at beyond3d, its rumored that we won't see an increase in transistor to performance from NV or AMD.

If those are the correct prices, this is what I had expected earlier with the 69xx series, but couldn't because of competition. AMD has wanted to raise the price for a while now, hence the name change and they are finally going to get to do it with the lack of NV.

[Dimitriman]

Sounds like a truly revolutionary step from AMD. Don't care if mid range is VLIW 4 because Cayman offers enough performance to cover that.
But GCN sounds like a very big chip from his description and with a lot of potential. Let's hope the thermals don't spoil it.

[Klarko]

What happened to my XDR2?

[Russian]

What happened to my XDR2?

You do realize how pointless XDR2 really is, right? Plus, there isn't that much volume production of it and the prices of it are much higher than GDDR5... MUCH higher.

[kuroikenshi]

I'd like to see a day that memory becomes something shared between CPU and GPU, need more bandwidth though.... maybe that hybrid memory cube from Samsung and Micron could be it.

I'd have the speed and bandwidth required, making video cards less expensive and cooler. (Utopian dreams here)

[hersounds]

i think that AMD can be put in market an special edition /overclocking edition with XDR2 memory why not?...many here can be purchase that.

[EniGmA1987]

i think that AMD can be put in market an special edition /overclocking edition with XDR2 memory why not?...many here can be purchase that.

Wouldnt the memory controller have to be reworked to run XDR2 instead of GDDR5? That would be pricey to do, probably greatly outweigh volume sold.

[Lanek]

Wouldnt the memory controller have to be reworked to run XDR2 instead of GDDR5? That would be pricey to do, probably greatly outweigh volume sold.

ALL the memory controller and system will be allready remade this time, in GCN ( for GCN, not VLIW4 shrink ) ..x86 memory instruction storage data, and specially memory access directly from the cpu on the gpu memory system ........ GDDR5 or not, the memory controller is modified totally. ( second coffee wait me ).
( GCN is a lot about computing, graphic wise i don't know what it will give, but computing wise, it's an extreme step, i just hope this step will not be a miss for the graphism part )

GDDR5 have still not be used at his full potential ( they use 5ghz GDDR5 when 3 year ago have been launch the 8ghz GDDR5 ( hence the rumor about cayman and 8ghz memory clock speed ).
But the old memory controllers used by AMD ( and Nvidia ) was just not in the case they can use this clock speed.

An absolute change is needed in computing for bring an Open standard for it ( based on OpenCL ), and AMD need absolulely bring a perfect solution for it. ( GCN ).... all computing solution maker wait about it. and the return for have open system is absolutely enormous( CUDA is horrible to use for server computing but a dream for low system computing ( the system is too much distorted for complex system, and it's a sub coding system, this mean you don't control and can't mix system with it ( ).... I hope AMD don't miss this opportunity. But AMD is a little bit alone for developp this computing standard .


pfff, i go for a 3th cofee...

[PatRaceTin]

35% faster sound good

[tbone8ty]

wow the 7870 looks insane @ $249 basically a 6970 shrunk to 28nm and at half the power consumption

so this article was kinda sorta correct minus the XDR2 memory

http://www.nordichardware.com/news/7...us-memory.html


(if some body can post the pics in this article it would be greatly appreciated)

[SKYMTL]

Here we go with the wild, over inflated hopes and dreams of some.....again.

[tbone8ty]

Here we go with the wild, over inflated hopes and dreams of some.....again.

so yur thinking the HD 7000 will be a flop?

[SKYMTL]

so yur thinking the HD 7000 will be a flop?

Absolutely not.

What I am saying is this: there is no reason to start or listen to any wild assumptions about pricing, performance or market orientation. They only way ANYONE will know the true potential of an upcoming product is when the first REAL reviews go live with a full suite of applications and using final drivers. Up until that point, everything is just background noise which will likely set people up for a disappointment.

I've said it many times and I will say it again here. There is a natural progression of these threads that goes something like this:

- Site A ensures they get plenty of traffic by posting supposed "inside info" about a product
- The excitement and wild assumptions begin
- Forum poster A will make some posts that look legit and draw the attention of the community (eg: neilz with Cayman and Barts or a certain unnamed forum member here -not JF- with Bulldozer)
- Yet more chest beating and assumptions ensue
- The launch draws near and people have completely the wrong idea about said product, its price and performance.
- Reviews are posted and the same people who built up the product in their heads as the Second Coming cry foul. "Too expensive", "where's the promised performance?" and "I'll wait for xxxx" are the usual responses.

There is something very important everyone needs to remember about pricing in particular. If a graphics card outperforms another, it WILL be more expensive. Simple as that. For example, if (and this is totally an example) the HD 7800 or whatever it will be called performs just above a HD 6970, expect it to be priced accordingly (ie: proportionally above the HD 6970's relative performance). Anything else would be market suicide for AMD.

In addition, I am sure we will once again see a ton of misinformation thrown around by various sites and forum posters prior to any GPU launch. There will invariably be some that after launch point to themselves and a small portion of their pre-launch assumptions and say "I told you so". Well, there is a good analogy for that: even a blind squirrel finds a nut once in a while.

[[XC] gomeler]

Absolutely not.

What I am saying is this: there is no reason to start or listen to any wild assumptions about pricing, performance or market orientation. They only way ANYONE will know the true potential of an upcoming product is when the first REAL reviews go live with a full suite of applications and using final drivers. Up until that point, everything is just background noise which will likely set people up for a disappointment.

I've said it many times and I will say it again here. There is a natural progression of these threads that goes something like this:

- Site A ensures they get plenty of traffic by posting supposed "inside info" about a product
- The excitement and wild assumptions begin
- Forum poster A will make some posts that look legit and draw the attention of the community (eg: neilz with Cayman and Barts or a certain unnamed forum member here -not JF- with Bulldozer)
- Yet more chest beating and assumptions ensue
- The launch draws near and people have completely the wrong idea about said product, its price and performance.
- Reviews are posted and the same people who built up the product in their heads as the Second Coming cry foul. "Too expensive", "where's the promised performance?" and "I'll wait for xxxx" are the usual responses.

There is something very important everyone needs to remember about pricing in particular. If a graphics card outperforms another, it WILL be more expensive. Simple as that. For example, if (and this is totally an example) the HD 7800 or whatever it will be called performs just above a HD 6970, expect it to be priced accordingly (ie: proportionally above the HD 6970's relative performance). Anything else would be market suicide for AMD.

In addition, I am sure we will once again see a ton of misinformation thrown around by various sites and forum posters prior to any GPU launch. There will invariably be some that after launch point to themselves and a small portion of their pre-launch assumptions and say "I told you so". Well, there is a good analogy for that: even a blind squirrel finds a nut once in a while.

You are screaming into the wind.

[SKYMTL]

You are screaming into the wind.

I know. But it feels good.

[Lanek]

You are screaming into the wind.

Still better of the other way ...

[paulbagz]

Absolutely not.

What I am saying is this: there is no reason to start or listen to any wild assumptions about pricing, performance or market orientation. They only way ANYONE will know the true potential of an upcoming product is when the first REAL reviews go live with a full suite of applications and using final drivers. Up until that point, everything is just background noise which will likely set people up for a disappointment.

HA!

I tried saying that about Ivy Bridge and I got slayed for it.

-PB

[Iconyu]

HA!

I tried saying that about Ivy Bridge and I got slayed for it.

-PB

No, we were telling people not to have high expectations about IB performance per clock, you said that we hadn't seen any numbers and we couldn't know for sure, but numbers had been leaked months ago. Ivy Bridge performs as expected.

[Lanek]

Here we go with the wild, over inflated hopes and dreams of some.....again.

Its 6hours here, and i have quickly read this before go work (drinking my traditionnaly cofee ) ..... but they say nothing we was know allready it seems, GCN for high end, and VILW4 in 28nm for the rest ? something new i have miss ? ( outside it is now confirmed, all the press will be brief during december )

So it has been confirmed the 7970 has a 384 bit memory bus? Alright, about time too ATI, its been a while since you moved away from 256 bit. I thought they were going to use XDR2? Whatever happened to that?

Me and all, have allways tell you XDR2 have never been confirmed... this is and was a pure rumors..... what happened with this rumor.. what happend with many rumor ?

[LordEC911]

As much as I hate doing it... I am about to pull a saaya.

VLIW4 is better than VLIW5 isnt it? Something about the extra overhead or whatever.

Compute capabilities.

But GCN sounds like a very big chip from his description and with a lot of potential. Let's hope the thermals don't spoil it.

Smaller than Cayman.

Also for the TL: DR Crowd:

It is more BS from BS.

A straight-up Cayman shrink shouldn't be too hard...

There is no straight Cayman shrink.

According to neliz at beyond3d, its rumored that we won't see an increase in transistor to performance from NV or AMD.

That generally doesn't happen unless something went terribly wrong the previous gen... *cough*R600*cough*
As you move away from fixed units to slightly more generalized units you have to, unfortunately, give up some transistor efficiency.

What happened to my XDR2?

It was never going to happen.

Plus, there isn't that much volume production of it and the prices of it are much higher than GDDR5... MUCH higher.

How about no volume production...
It is hard for a certain company to get "partners" to manufacture their product when they sue them left and right.

Here we go with the wild, over inflated hopes and dreams of some.....again.

Don't worry, you will get your card soon enough.

So it has been confirmed the 7970 has a 384 bit memory bus?

Pretty much, if you believe Charlie.

Neliz? That guy promised a bunch of stuff in regards to the Cayman GPUs verse the GTX 580 when it launched. I'm not really going to bother paying attention to his input after that.

Quotes please. He was pretty much on the money with most of his info.
When you have better tidbits than him please post them.

A rumor is never a solid fact.

True but some rumors are better than others.

so yur thinking the HD 7000 will be a flop?

http://www.xtremesystems.org/forums/...=1#post4999710

... and VILW4 in 28nm for the rest ? something new i have miss ? ( outside it is now confirmed, all the press will be brief during december )

No. Pitcairn is GCN. The midrange/lowend is what is questionable.
There are at least 3 28nm ASICs and the 40nm lowend rebrands are 40nm.