AMD gave details on its next GPU architecture at AFDS, goes Scalar and Out of Order.

[grimpr]

http://www.hardware.fr/news/11648/afds-architecture-futurs-gpus-amd.html

 

[velis]

Is this meant for 7000 series or 8000 series? The article doesn't say or at least I couldn't find it...

 

[Arkadrel]

Velis I believe this is for the 7xxx series.
Seems like their going for more computational power, than their previous GPUs.

GPUs that are OoO.... I wonder what that ll do for the GPU.

 

[Skurge]

Acording to that article, there should be a demo of the new architecture.

 

[Tempered81]

Is this meant for 7000 series or 8000 series? The article doesn't say or at least I couldn't find it...

In the article it says this is not for Trinity (hd7000). It's likely 8000 or even 9000 branded radeons.


It's a complete rework, and focuses on compute - a lot like Fermi's concept.

VLIW4->MIMD/SIMD.

 

[Troll Trolling]

In the article it says this is not for Trinity (hd7000). It's likely 8000 or even 9000 branded radeons.


It's a complete rework, and focuses on compute - a lot like Fermi's concept.

VLIW4->MIMD/SIMD.

Trinity is Llano's successor, not an graphic series.

 

[blastingcap]

Velis I believe this is for the 7xxx series.
Seems like their going for more computational power, than their previous GPUs.

GPUs that are OoO.... I wonder what that ll do for the GPU.

I wondered out loud a long time ago if AMD's CPU guys could help out their GPU guys with this. *If* this is true, it looks like AMD did what NVDA did not or could not do with Fermi... http://www.anandtech.com/show/2918/2

 

[podspi]

In the article it says this is not for Trinity (hd7000). It's likely 8000 or even 9000 branded radeons.


It's a complete rework, and focuses on compute - a lot like Fermi's concept.

VLIW4->MIMD/SIMD.

The name of the "fGPU" doesn't have much relation to what it actually is, IIRC. The 6xxx integrated into Llano is based off of the 5xxx series, right?

I can't imagine OoO would do much for GPU tasks, but I could be wrong...

It sounds to me like this arch is actually very versatile, it can be geared towards compute OR pure graphics rendering, depending on the proportions of resources AMD sprinkles on there...


I wonder if they are going to start releasing compute-only cards with this arch, perhaps for Cray? :awe:

 

[grimpr]

Well, we'll find out soon enough, one hour to go for Eric Demers keynote.

http://developer.amd.com/afds/pages/webcast.aspx

 

[Kenmitch]

The example's of coding look interesting to me....Not that I can read GPU Geek but it does look like it takes a whole lot less code to perform the same function.

 

[Arzachel]

In the article it says this is not for Trinity (hd7000). It's likely 8000 or even 9000 branded radeons.
It's a complete rework, and focuses on compute - a lot like Fermi's concept.
VLIW4->MIMD/SIMD.

Trinity will use a HD6XXX graphics core. It seems like the HD7XXX line will use this architecture.

 

[KingstonU]

DO WANT...Nowwwwwww!!

 

[Genx87]

Velis I believe this is for the 7xxx series.
Seems like their going for more computational power, than their previous GPUs.

GPUs that are OoO.... I wonder what that ll do for the GPU.

Bring it more in line with Fermi. Including die size and energy consumption.

 

[Phynaz]

DO WANT...Nowwwwwww!!

May I ask why you are so excited?

 

[Lonyo]

Bring it more in line with Fermi. Including die size and energy consumption.

I would hope not, because that would put AMD with 2009 tech heading into 2012.

 

[Arachnotronic]

May I ask why you are so excited?

Because it's going to be ridiculously fast and it's going to put AMD into a position of kicking Intel and NVIDIA's butts in the HPC world?

 

[Nemesis 1]

SO what makes you think AMD has't found out the performance of Nights ferry in HPC work. YOU do know Nights corner is coming on 22nm right its for HPC market and developers have it(nights ferry) . Looks to me like AMD is scrambling .

 

[tviceman]

So, is this basically saying they're moving to a unified core architecture somewhat similar to Nvidia?

 

[Lonyo]

SO what makes you think AMD has't found out the performance of Nights ferry in HPC work. YOU do know Nights corner is coming on 22nm right its for HPC market and developers have it(nights ferry) . Looks to me like AMD is scrambling .

Because they have been planning this for longer than Nights Ferry has been in the works and they couldn't exactly make significant changes this late in the game.

 

[Skurge]

So, is this basically saying they're moving to a unified core architecture somewhat similar to Nvidia?

is fermi out-of-order aswel?

 

[Skurge]

Because they have been planning this for longer than Nights Ferry has been in the works and they couldn't exactly make significant changes this late in the game.

yeah. If it is launching this year then it would have been in development for the last 4yrs. That would be a year or so after AMD bought ati, the CPU division might have had a lot of influence on this design.

 

[blastingcap]

is fermi out-of-order aswel?

Nope.

http://www.anandtech.com/show/2918/2

 

[Ryan Smith]

Nope.

http://www.anandtech.com/show/2918/2

And either is Graphics Core Next. The ability to reorder and preempt threads (or rather wavefronts in this case) is not the same thing as reordering instructions within a wavefront. This is no more out of order than an Atom is.

We'll have more tomorrow; Anand & I are still working on our article.

 

[3DVagabond]

And either is Graphics Core Next. The ability to reorder and preempt threads (or rather wavefronts in this case) is not the same thing as reordering instructions within a wavefront. This is no more out of order than an Atom is.

We'll have more tomorrow; Anand & I are still working on our article.

I'm in NZ. It's already tomorrow here. So you can tell me.

 

[Genx87]

I would hope not, because that would put AMD with 2009 tech heading into 2012.

On the compute side they want\have to catch up.