[Alexko]

There's an AMD analyst day scheduled on February 2, where the company will surely talk about that.

[denev2004]

Originally Posted by sebbbi

Of course, but in the long run Brazos seems to be fading away. There has been basically no progress in one year now (0.05 GHz overclock with E450, and another 0.05 GHz coming up later this year). Z-01 at 5.9W (1/3 of Brazos TDP) is not that far behind in performance (and is really cheap to produce), and Trinity offers much better performance at same TDP. If either the Z-01 is scaled up or Trinity is scaled down (single module version would be considerably cheaper to produce and wouldn't likely need agressive cherry picking to reach 17W), there would not be much room left for Brazos anymore. Too bad Deccan, Wichita and Krishna all got canned (28 nm would have made the E-series much more competitive in the long run).

Actually I'm wondering about another problem, there also seems to be no improvement in intel's Z series Atom for a long time...

[mczak]

Originally Posted by denev2004

Actually I'm wondering about another problem, there also seems to be no improvement in intel's Z series Atom for a long time...

You missed the Z2460? Or are you talking about the cpu core? In this case yes not much progress has been made. That surely needs to change for Silvermont (and it is indeed supposed to) otherwise it will look silly against all those Cortex-A15 SoCs...

Originally Posted by Kaotik

I haven't seen any signs of Wichita & Krisha being canned, there were a lot of false news about how they need to "redesign them for TSMC" which was obviously false, it was known like 6 months before those news that at least part of Wichita and/or Krishnas will come from TSMC.

Even Wikipedia lists Krishna/Wichita as canned, it must be true .

[Kaotik]

Originally Posted by mczak

Even Wikipedia lists Krishna/Wichita as canned, it must be true .

Yeah, someone heard GloFo is having issues and that they need to be redesigned for TSMC "from scratch",the news spread like wildfire, and everyone ignored the 6 months earlier info saying TSMC is doing at least part of them anyway.

[mczak]

Originally Posted by sebbbi

or Trinity is scaled down (single module version would be considerably cheaper to produce and wouldn't likely need agressive cherry picking to reach 17W)

I guess a 1 module, 3 shader cluster Trinity would still be somewhere along ~150mm². Cheaper yes, but that's still quite big. Speaking of that, what happened to the dual-core Llanos? I mean true dual-cores not downrated four core ones. Do they exist? Some of the internal designations floating around for the graphics part (SUMO vs. SUMO2) seems to suggest there are indeed two parts, but that might also be a result from simply disabling some stuff. If they exist, I've never seen anything indicating transistor count or die size.

[mboeller]

a few Trinity presentation slides from CES:

http://www.computerbase.de/news/2012...te-des-jahres/

look at "Update 18:13 Uhr"

[Alexko]

The last slide claims 3 hours and 20 minutes of 3DMark 06 on battery power… that would be pretty impressive.

[mczak]

Originally Posted by Alexko

The last slide claims 3 hours and 20 minutes of 3DMark 06 on battery power… that would be pretty impressive.

Yes, but it doesn't say anything about the performance while achieving this. It might be possible some battery profile downclocks the gpu (or disables some units). Not saying that's happening but there are so many unknowns in these slides it's difficult to draw any conclusions.

[fellix]

I measured 236mm² die area, with the Euro coin reference at 23mm in diameter.

[nAo]

Originally Posted by fellix

I measured 236mm² die area, with the Euro coin reference at 23mm in diameter.

Is it built on a 28nm or 32nm process?

[mczak]

Originally Posted by nAo

Is it built on a 28nm or 32nm process?

Still the same as Llano, so 32nm HKMG SOI at GloFo. I don't know if the 236mm² would be more accurate than the previous die size number from semiaccurate which was 240mm².

[Gipsel]

Originally Posted by fellix

I measured 236mm² die area, with the Euro coin reference at 23mm in diameter.

So Charlie was right on the spot with his "the die size is within an hair of 240mm²".

Originally Posted by nAo

Is it built on a 28nm or 32nm process?

32nm of course.

[LordEC911]

Originally Posted by Kaotik

Yeah, someone heard GloFo is having issues and that they need to be redesigned for TSMC "from scratch",the news spread like wildfire, and everyone ignored the 6 months earlier info saying TSMC is doing at least part of them anyway.

Exactly.
AMD's first 28nm APU taped out almost a year ago at TSMC...
There was at least one more in the pipeline as well, though I'm not sure when that one taped out.

[itsmydamnation]

http://www.computerbase.de/news/2012...te-des-jahres/

now its upto 25% x86 performance increase....

[no-X]

Compared to Llano (this time), not to BD module

[TKK]

Originally Posted by no-X

Compared to Llano (this time), not to BD module

Well, those "up to 25%" are for 35W TDP. The fastest mobile 35W Llano is the 3520M, with only 1.6 GHz base clock (900 MHz turbo, but iirc, reviews indicated that Llano's turbo rarely kicks in).

Beating the 3520M by "up to 25%" shouldn't be that hard, even with lower IPC.

[sebbbi]

http://www.dailytech.com/article.aspx?newsid=23777

Another AMD CES interview. Clears up some things I was wondering: Originally Brazos was targeted for Fall 2011. Brazos got out half year early. That pretty much explains why the follow up hasn't yet shown up. AMD also stated that they are targeting Trinity for sub 500$ laptops (slightly lower price point than Llano's goal of 600$). That would overlap with the top of the line Brazos laptops (the most popular high end Brazos laptop, the HP dm1z was 400$-550$ based on configuration). That would explain why the Bobcat based APUs aren't agressively scaled up (scaling Bobcats down to tablets would be more important for AMD, as 17W Trinity can cover the high end netbooks and ultraportables).

[ToTTenTranz]

They also DOUBLE-CONFIRM the cancellation of Krishna/Wishita in favor of the 17W Trinity.
Given that the 17W Trinity is in fact a 2-module/4-core APU, they should be able to eventually release a 1-module/2-core model while lowering the power consumption by a handful of watts (I don't know if that's in the current roadmap or if it's interesting to them in any way).


But the best thing is that they're actually trying to get Trinity to replace higher-end Brazos models in ~400€ subnotebooks (HP DM1, EeePC 1215B, etc).
This means we're looking at some 400% GPU and CPU performance difference between a 2011 model and a 2012 one.
That's a lot!


I may feel a bit too compelled to replace my 2,5 year-old Ferrari One 200.

[mczak]

Some error in that article it mentions Brazos and Llano came in "traditional" packaging, whereas Trinity will be available in BGA. Well Brazos certainly was BGA only.

Originally Posted by ToTTenTranz

But the best thing is that they're actually trying to get Trinity to replace higher-end Brazos models in ~400€ subnotebooks (HP DM1, EeePC 1215B, etc).
This means we're looking at some 400% GPU and CPU performance difference between a 2011 model and a 2012 one.
That's a lot!

I think that's a tad optimistic. Surely low-end versions of the chip (such as used in these sub-500$ subnotebooks) would not be fully enabled Trinity (separate die or not). Just like the current E-series Llano (which I don't know if someone is actually using it). I'd expect more like one module, low clocked, with half (at best) of the simds enabled. That would be more like twice as fast (for both cpu and gpu) as your typical Brazos. Still, certainly a notable improvement!

[itsmydamnation]

why only one module ?

I'd expect more like one module, low clocked,

there is a 35watt 8 core bulldozer @1.6 with turbo to 2.8

so you get rid of 8mb of L3,2 modules an updated core and a chance to learn from any mistakes they made in the physicals of bulldozer. We will see much better power features then Llano and likely an improvement over bulldozer as well.

put that all together and i think we will likely see 4 core somewhere around 1.5ghz, but i agree it wont have all SIMD's, unless having a super low clock with more SIMD's is more power efficient then having a low clock with less SIMD's.

cheers

edit: i dont think bobcat will be going anywhere, if i was AMD my role for bobcat would be a dual role. 1. high performance tablets, do whatever needs to be done to hit that market and FFS put a 3G/LTE transceiver in it! Second would be take that core and clock it as high as you can for a given TDP ( i guess something around 40watts) and sell that product just in emerging markets.

if bulldozer does low power well, you dont want to be selling big chips to "poor" people when it can be put in expensive *books and a little chip can service the "poor" people.

poor== its all relative right?

edit2: clarification, i don't care what bit of silicon the LTE/3G transceiver sits on, i just want it in the damn tablet.

[Albuquerque]

Originally Posted by itsmydamnation

why only one module ? there is a 35watt 8 core bulldozer @1.6 with turbo to 2.8

Yes, but Bulldozer is the CPU only. Trinity also must combine the GPU elements, and consider that GPU is a considerable power consumer for the package. If you could 'only' cut the CPU in half, you'd still be over the 17W power budget. The number of cores will necessarily decrease by almost 1/4 of Bulldozer's 8-core design in order to have enough power envelope to squeeze in the power budget for the GPU.

[ToTTenTranz]

The 17W APU shown in CES is 2-module/4-core. It's been confirmed in several places, and the pictures taken by the press show the 17W BGA chip with exactly the same size as the 35W laptop socket version.

AFAIK, what AMD said is that they will only be targeting the sub-$500 notebook market, as they'd have a very hard time competing with intel in the high-end/high-margin market. To me, this means the 2-module part is going for sub-$500 machines as AMD simply isn't going to compete in the market for higher priced models.


What is still unknown is how much power the rest of the system will consume. IIRC Trinity has a 2-channel DDR3, which should consume a bit more than Brazos' memory subsystem, and we're yet to know how much the Trinity's "southbridge" consumes. This means a 17W Trinity laptop may have substantially lower battery life than a 18W Brazos one.

[itsmydamnation]

Originally Posted by Albuquerque

Yes, but Bulldozer is the CPU only. Trinity also must combine the GPU elements, and consider that GPU is a considerable power consumer for the package. If you could 'only' cut the CPU in half, you'd still be over the 17W power budget. The number of cores will necessarily decrease by almost 1/4 of Bulldozer's 8-core design in order to have enough power envelope to squeeze in the power budget for the GPU.

that assumes
no advancement in process
no advancement in power consumption from bulldozer to piledriver
no advancement/optimization in the floor plan
ignoring the power consumption of the L3.
advancements in power sharing between CPU and GPU

I wouldn't be writing off 4 cores just yet....................

[Raqia]

I wonder if they ever considered doing a version of Trinity on TSMC's 28nm process. The excellent yields for 28nm (and high clock headroom for the 7970) makes me think it might be worth a shot to branch out for the subsequent high-end design for the capacity alone despite the different design rules at the two fabs.

In any case, the excellent 28nm process at TSMC should atleast make the next gen Jaguar core based fusion quite interesting. I'm thinking they could get a massive IPC boost by making cache full speed but more judiciously powering it up when needed. I also wish that more of the lower end fusion netbooks could get nicer IPS monitor as options atleast.

[Albuquerque]

Originally Posted by itsmydamnation

that assumes
no advancement in process
no advancement in power consumption from bulldozer to piledriver
no advancement/optimization in the floor plan
ignoring the power consumption of the L3.
advancements in power sharing between CPU and GPU

I wouldn't be writing off 4 cores just yet....................

Based on all currently available data regarding AMD's lithography processes, voltage needs and power draw, I see no reason to amend my statement yet.

If you can show me anything that AMD is currently producing that can change my mind regarding the power they're chewing through and the performance they're not getting from it, then please feel free to show me.

The 17W APU shown in CES is 2-module/4-core. It's been confirmed in several places, and the pictures taken by the press show the 17W BGA chip with exactly the same size as the 35W laptop socket version.

I saw one place claim it was 17W, and a bunch of others parrot that claim. I've seen no confirmation of that anywhere (ie, measurement of power draw from the unit in some definable way.)

I can take a picture of a 17W Sandy Bridge package and a 45W Sandy Bridge package and you will never be able to tell the difference. You know why they will look the same? Because they are the same. Showing a picture of a CPU die is meaningless to power consumption, so when you come back with your double-confirmed proof, please be aware that a picture of a BGA package isn't part of that proof.

Nowhere am I saying that AMD cannot accomplish a 17W Trinity, but I am saying that nothing we have to go on today indicates that it will be a 'fully featured' part.