Trinity vs Ivy Bridge

Maybe the planned (faster?) Llanos have been canned and Trinity comes out earlier?
It's not impossible scenario - AMD needs all the momentum it can get, if you have the ability to push your product out faster at the expense of older product moving out of the pipeline faster than planned, so be it.
I was more thinking of the slower Llanos which aren't out yet (A4 etc.) though actually they are out now so the family is complete (haven't seen any review of those parts yet).
I agree the faster they can get it out the better, it just doesn't seem likely - heck Trinity is enhanced Bulldozer and Bulldozer isn't here yet neither.
 
I was more thinking of the slower Llanos which aren't out yet (A4 etc.) though actually they are out now so the family is complete (haven't seen any review of those parts yet).
I agree the faster they can get it out the better, it just doesn't seem likely - heck Trinity is enhanced Bulldozer and Bulldozer isn't here yet neither.

In this thread over at XS you will find few results from A4 APU ;)

http://www.xtremesystems.org/forums/showthread.php?274882-Dual-Core-APU-in-stock-in-Hong-Kong
 
That was achieved by 2.15 billion vs. 956 million transistors. A 125% increase. That's my point. :)

The comparison is totally meaningless though, compare rv710 to Cedar (+20% transistors) and rv770 to Juniper (+10% transistors though it lost some functionality (DP support) and half the mem i/o too which probably wasn't that many transistors, if clocked the same and with similar mem bw they pretty much have the same performance). So I don't really see the huge increase in transistors just for dx11 features - a noticeable increase yes but not that dramatic.
Yes we know Juniper->Cypress doesn't scale that well considering the difference in transistor count, but saying Cypress has way more transistors for little performance gain compared to rv770 due to dx11 is just silly.
 
Oh interesting. The scores aren't that hot, though still a bit faster than HD3000. Ivy Bridge will almost certainly beat that, though luckily (well assuming it will really launch quite a bit earlier than Trinity in the first place) in that price bracket there most likely won't be any IVB parts with 16 EUs and the 8 EU parts probably won't beat HD3000...
btw are these A4 native dual cores or die-harvested parts?
 
Ivy Bridge GPU SKUs are 6 and 16 EUs. Some early samples also indicated 1600MHz Turbo GPU clock, meaning 20% increase. I wonder how they'll seperate that?
 
Actually I'm quite sure it was only 16bit, supporting ddr2/3 on all the 7xx chipsets.
Not sure about rs690 might have been 16 or 32bit (but didn't support ddr3 for sure).

Exactly. SidePort was only 16bit on RS780/RS880 chipsets, it's basically same engine in both chipsets only 780G/790GX was only dx10 and 785G/880G/890GX is dx10.1. And it always used one x16 memory chip. Older chipsets didnt have Sideport at all as it SP was firstly introduced for RV770 chips.

SidePort in IGPU chipsets helps alot in DX performance even if it's only 16bit (there was some good test from some user on 880G that even AMD liked in marketing their chipsets)
 
Exactly. SidePort was only 16bit on RS780/RS880 chipsets, it's basically same engine in both chipsets only 780G/790GX was only dx10 and 785G/880G/890GX is dx10.1. And it always used one x16 memory chip. Older chipsets didnt have Sideport at all as it SP was firstly introduced for RV770 chips.
There were definitely RS690 variants supporting sideport. Might not have been widely used though.

SidePort in IGPU chipsets helps alot in DX performance even if it's only 16bit (there was some good test from some user on 880G that even AMD liked in marketing their chipsets)
A 10-15% increase or so isn't really "a lot" in my books.
 
Curious thing, notice how "London graphics" is present in every APU from the ~9W Krishna (Deccan) all the way up to Trinity A8 (Comal) models.

What are the chances of all these using the exact same iGPU while just enabling more or less shader/TMU/ROP units, crossfire support and changing clocks?

Llano's Beavercreek goes from 240 shaders to 400, for example.

But it's strange that such a big GPU would go into a ~9W APU, right?


Brazos seems relatively unchanged unitl 2013, which is kind of a let down for my ideal future 10" Windows 8 tablet :(
 
Bobcat cores are, if my memory serves me right, being tuned though, they can also add more speed & cores to those in same or even smaller thermal envelopes thanks to moving to smaller processes - I don't see what other kind of upgrades you could possibly want for them.
 
Bobcat cores are, if my memory serves me right, being tuned though, they can also add more speed & cores to those in same or even smaller thermal envelopes thanks to moving to smaller processes - I don't see what other kind of upgrades you could possibly want for them.

Brazos-T isn't moving to smaller processes, it's staying on 40nm. That's the problem.



I believe PowerTune can help to fit Trinity APUs to predefined TDP limits.

The curious thing is Krishna, 2012's equivalent to C-50 and E-350, getting the same GPU as Trinity, which is supposedly more powerful than today's Llano (from 240sp, 12TMUs, 4ROPs to 400sp, 20TMUs, 8ROPs).

And then there's the matter of die size. Even a ~240sp VLIW4 GPU should be huge compared to 2 or even 4 bobcats.
Is it in AMD's best interests to have such a "big" chip for low-end laptops?

Let's hope Krishna has at least a dual-channel memory controller, or such GPU will be seriously bandwidth bottlenecked.
 
Last edited by a moderator:
Yeah, london is the family consisting of Thames, Chelsea, Heathrow, Wimbledon.. and supposedly the gpus on khrisna and trinity.
28nmmdgpumobile_1a_Dh_fx57.jpg.jpeg
 
Back
Top