AMD Bristol/Stoney Ridge Thread

[Flash831]

Stoney Ridge is a ~125mm² die, but I'm not certain if there will be harvested Carrizo / Bristol Ridges used for some of "Stoney" models.

AMD seems to be using the harvested dies for their embedded products.

Harvested Carrizo/Bristol Ridge dies -> I-family. Based on the fact it uses 4 CU igpu.
Harvested Stoney ridge dies --> J-family

 

[Dresdenboy]

So is confirmed...
http://wccftech.com/amd-bristol-ridge-processor-launch/
http://www.anandtech.com/show/10362/amd-7th-generation-apu-bristol-ridge-stoney-ridge-for-notebooks
Bristol and Stoney Ridge released on mobile!

Nice, my combination of Benchlife + USB leak info was spot on.

[/QUOTE]

 

[The Stilt]

AMD seems to be using the harvested dies for their embedded products.

Harvested Carrizo/Bristol Ridge dies -> I-family. Based on the fact it uses 4 CU igpu.
Harvested Stoney ridge dies --> J-family

And I thought the old naming / model scheme was already confusing :\

 

[Shehriazad]

Im looking forward for FM2+ part and later for AM4 part of desktops Bristol ridges

Unless I missed something...what makes you think they'd bother releasing Bristol to FM2+?

They would shoot themselves in the foot if AM4 is about to be released.

 

[The Stilt]

There is no point in "releasing" Bristol Ridge separately for FM2+, as the same silicon already exists in form of Athlon X4 845 for FM2+.

 

[Dresdenboy]

There is no point in "releasing" Bristol Ridge separately for FM2+, as the same silicon already exists in form of Athlon X4 845 for FM2+.

They mentioned process tweaks for BR. I wonder if they simply carry over to CZ.

 

[The Stilt]

They mentioned process tweaks for BR. I wonder if they simply carry over to CZ.

I wonder what these might be. The dies are identical (as you've seen) and most of the low-level software is identical / shared too (microcode, co-proc. FWs etc). I believe the difference is purely in SW configuration, if there is any in the first place.

The situation is different on Stoney Ridge, which is also a 7th generation APU. It is a new die and there are hardware changes (security processor, UVD).

 

[AtenRa]

Consumers also asking for 35W APU only sub $500 with Dual Channel DDR-4 and 14-15.6" but not AMD nor OEMs are lessening.

Can we have a A10-9630P APU only with dual Channel DDR-4 2400MHz at 14" format at sub $500 ???? That is what consumers want and are willing to spend.

 

[hojnikb]

Consumers also asking for 35W APU only sub $500 with Dual Channel DDR-4 and 14-15.6" but not AMD nor OEMs are lessening.

Can we have a A10-9630P APU only with dual Channel DDR-4 2400MHz at 14" format at sub $500 ???? That is what consumers want and are willing to spend.

That would sell like hotcakes, if performance was there.

 

[hojnikb]

There is no point in "releasing" Bristol Ridge separately for FM2+, as the same silicon already exists in form of Athlon X4 845 for FM2+.

Would APU (igpu+cpu) even work on fm2+ ?

 

[ShintaiDK]

Consumers also asking for 35W APU only sub $500 with Dual Channel DDR-4 and 14-15.6" but not AMD nor OEMs are lessening.

I dont think consumers ask for 35W, but rather 15W.

 

[The Stilt]

IMO anything above 20W TDP is too high for a laptop, even for the larger framed ones. For Carrizo this should be the sweet spot in terms of performance & power anyway.

For example in the case of AMD Carrizo reference system. It has good cooling, quality fans (ADDA) and a rather spacious frame. Still I always set the cTDP to minimum available value (12/20W) when I use it for anything else than testing.

 

[The Stilt]

Would APU (igpu+cpu) even work on fm2+ ?

No.
Hardware wise it would be doable, but the way AMD has built the power management in these chips it is impossible.

 

[NTMBK]

IMO anything above 20W TDP is too high for a laptop, even for the larger framed ones. For Carrizo this should be the sweet spot in terms of performance & power anyway.

For example in the case of AMD Carrizo reference system. It has good cooling, quality fans (ADDA) and a rather spacious frame. Still I always set the cTDP to minimum available value (12/20W) when I use it for anything else than testing.

Modern laptops should have a big fat hardware "TURBO" button, like the 386 desktops of yore. Quiet and battery efficient most of the time, then kick it into high gear when you want to play games/transcode/whatever.

Yes I know you can do similar things with power settings, but a physical button is much cooler than messing around in the taskbar.

 

[AtenRa]

IMO anything above 20W TDP is too high for a laptop, even for the larger framed ones. For Carrizo this should be the sweet spot in terms of performance & power anyway.

For example in the case of AMD Carrizo reference system. It has good cooling, quality fans (ADDA) and a rather spacious frame. Still I always set the cTDP to minimum available value (12/20W) when I use it for anything else than testing.

We had 35W TDP Core i5s at 12.1" format 3-4 years ago and you telling me we cannot have a 35W TDP APU at 14" format with low power DDR-4 in 2016 ???

When we see 15W TDP CPUs paired with 35W TDP mobile dGPUs its better than having a 35W TDP APU alone ???

There are laptops with dGPUs that are worse performance than their iGPUs, that is called stupidity from the OEMs part as well for those retailers that are selling crap like that to poor uneducated consumers.

 

[LTC8K6]

Hard to believe that most users even have a clue what TDP means, let alone asking for a specific number...

 

[cbn]

When we see 15W TDP CPUs paired with 35W TDP mobile dGPUs its better than having a 35W TDP APU alone ???

For these 35W APUs competing against the 15W CPU plus dGPU, I would hope we could see an emphasis on the space saving effect of the 35W APU SoC.

As I mentioned here, maybe even have a cooling system that could replace the battery when connected to AC power.

So in this way the laptop could still be relatively small, but yet still have the cooling for 35W when a wall outlet is available.

Does that make sense?

 

[therealnickdanger]

Modern laptops should have a big fat hardware "TURBO" button, like the 386 desktops of yore.

I know what you meant, but the "turbo" buttons of yore were used to slow down the PC to make faster CPUs backward compatible with software designed to run at legacy 8086 speeds (~4.5MHz). Wow, let that sink in for a bit... 4.5MHz for a PC. LOL

Of course, it would be fun to take one of those old AT cases and use that button for a turbo mode in the sense you're talking about.

 

[Dresdenboy]

I wonder what these might be. The dies are identical (as you've seen) and most of the low-level software is identical / shared too (microcode, co-proc. FWs etc). I believe the difference is purely in SW configuration, if there is any in the first place.

The situation is different on Stoney Ridge, which is also a 7th generation APU. It is a new die and there are hardware changes (security processor, UVD).

The AT article contains some slides, incl. this one:

http://www.anandtech.com/show/10362/amd-7th-generation-apu-bristol-ridge-stoney-ridge-for-notebooks

What I had in mind was this:

 

[The Stilt]

The AT article contains some slides, incl. this one:

http://www.anandtech.com/show/10362/amd-7th-generation-apu-bristol-ridge-stoney-ridge-for-notebooks

That's interesting. I wonder if they've actually gone back to GF28A instead of GF28HPP on Bristol Ridge. I don't think it is likely that they spend a single additional penny on a dead design / microarchitecture at this point thou.
But it the otherhand, they did create Stoney Ridge too. Impossible to know what these people are thinking.

Technically they could achieve the same "improved process" through simply binning the silicon better. The best silicon for BR and the rest for Carrizo. I could produce similiar "improvement" chart on FX-8800P by simply undervolting.

In the other hand, the production scale in terms of VDD vs. Frequency on BR is the same or even higher than on Carrizo.

 

[dark zero]

As expected 28nm HPP was not the real deal and 28A (used on Richland if I am not wrong) is the correct option. Seems that AMD learned their lesson and finally is about to hit Phenom II performance again.

 

[Dresdenboy]

That's interesting. I wonder if they've actually gone back to GF28A instead of GF28HPP on Bristol Ridge. I don't think it is likely that they spend a single additional penny on a dead design / microarchitecture at this point thou.
But it the otherhand, they did create Stoney Ridge too. Impossible to know what these people are thinking.

Technically they could achieve the same "improved process" through simply binning the silicon better. The best silicon for BR and the rest for Carrizo. I could produce similiar "improvement" chart on FX-8800P by simply undervolting.

In the other hand, the production scale in terms of VDD vs. Frequency on BR is the same or even higher than on Carrizo.

I think, there are bigger differences between GF28A and GF28HPP to switch that easily. What I wanted to show with the older slides is, that a process could be tuned in a steady improvement process.
See http://www.anandtech.com/show/2018

But that might also simply be the yield learning curve.

Addendum:
There are more BR related slides at P3DNow!:
http://www.planet3dnow.de/cms/24116-amd-stellt-apus-der-7-generation-bristol-ridge-vor/
Teaser:

Stilt, do the tolerances make more sense now?

 

[DrMrLordX]

That's interesting. I wonder if they've actually gone back to GF28A instead of GF28HPP on Bristol Ridge. I don't think it is likely that they spend a single additional penny on a dead design / microarchitecture at this point thou.
But it the otherhand, they did create Stoney Ridge too. Impossible to know what these people are thinking.

If they ported Carrizo over to GF28a/28SHP then is it possible that they added more l2 as well? Or have we confirmed that BR still has 1 mb l2 per module?

 

[JDG1980]

The presentation for Bristol/Stoney Ridge talked a lot about HEVC support, but no specific details were given. Carrizo (and Fiji) supported HEVC, but only the 8-bit version, not the Main10 profile that is required by 4K Netflix and Ultra HD Blu-Ray. Was Bristol Ridge updated with the new UVD block to support both profiles? IMO, it would be a bit dishonest to talk up HEVC support to such an extent if it omits the profile needed for most real-world applications.

 

[The Stilt]

If they ported Carrizo over to GF28a/28SHP then is it possible that they added more l2 as well? Or have we confirmed that BR still has 1 mb l2 per module?

Sure.
The die shots "between" Carrizo and Bristol Ridge revealed zero differences between the two. Other than the altered coloring and different cropping AMD used in the shots :sneaky: A full Carrizo die PS has never been published (AFAIK), but with AMD released on on Bristol Ridge.

Cropped Carrizo on top of full PS Bristol Ridge.

BR: http://blogs-images.forbes.com/patr...5/7thGen-DIE-SMALL-Final-Recolor-1200x847.jpg

CZ: http://gallery.dpcdn.pl/imgc/Video/63433/g_-_-x-_-_-_63433x20150603021240_0.jpg