AMD Carrizo Pre-release thread

[Abwx]

Some Carrizo numbers from the conference :

AMD A10-8700P with AMD Radeon™ R6 graphics, 2x4GB DDR3-1600, 256 SSD, Windows 8.1 64bit, Catalyst 15.10 beta scored 1100 in 3dmark Firestrike. Intel Core i5 5200u, 2x4GB DDR3-1600, 256 SSD, Windows 8.1 64bit, Driver 4170 scored 780 in 3dmark Firestrike. The result is 41% faster performance from AMD.

The 2015 AMD "Carrizo" performance platform projects 11.2 hours battery life running HD 1080p Local Video Playback while the 2014 "Kaveri" performance platform battery life under the same workload is projected to last 5.3 hours.

Testing by AMD Performance labs using reference platforms based on an AMD 6th generation Processor featuring an AMD FX-8800P platform, 2x4GB DDR3-1600, 256GB SSD, Windows 8.1, Driver 15.100 lasting 10.2 hrs The comparison system consisted of an AMD FX 7600P with 2x4GB DDR3-1600, 256GB SSD, Windows 8.1, 14.200.1004 drivers. Jan 15, 2015 lasting 7.1 hrs. Testing done with a 50 w/hr battery with Varibright mode set to aggressive, and airplane mode enabled, with MobileMark12

 

[krumme]

Some Carrizo numbers from the conference :

Its the battery life numbers that seems good.

Sad to see this high freq server arch ending its days in a low freq notebook apu. But considering a failed arch in a low freq aplication made by a very budget restrained team on an old 28nm process - well from that perspective it looks very well made - impressive its possible at all ...but crazy.


What about design wins?

 

[Enigmoid]

What is the TDP of the a10-8700P?

 

[Arachnotronic]

EDIT: Nevermind.

 

[Abwx]

Its the battery life numbers that seems good.

Sad to see this high freq server arch ending its days in a low freq notebook apu. But considering a failed arch in a low freq aplication made by a very budget restrained team on an old 28nm process - well from that perspective it looks very well made - impressive its possible at all ...but crazy

What about design wins?

The idle power match their Beema SoC but the Videoplay is much better, perf/Watt wise they should largely exceed the competition in both GPU and CPU usages, wich is quite a performance given the nodes difference, and this is where the CMT concept shine actualy because at equal process it would have a better perf/Watt ratio than big cores, Carrizo good perf/watt is also due to this caracteristic.

As for design wins they seem to be confident about the thing, you can read through the Carrizo dedicated slides, must be Papermaster notes or Lisa Su s ones :

http://ir.amd.com/phoenix.zhtml?c=74093&p=irol-analystday

What is the TDP of the a10-8700P?

Likely the same as the comparison, so 15W as base TDP with thermal headroom used to boost the chip up to 25W according to a SA member in the know, exactly like the Intel chips, i guess that they had to copy this unfortunate "feature" for marketing reasons..

 

[ElFenix]

Sad to see this high freq server arch ending its days in a low freq notebook apu. But considering a failed arch in a low freq aplication made by a very budget restrained team on an old 28nm process - well from that perspective it looks very well made - impressive its possible at all ...but crazy.

considering it's improved this much without a real die shrink is pretty amazing. with actual engineering resources and a real die shrink or two, this could have been a pretty interesting architecture.

 

[Asterox]

I posted in wrong place, so Mod delete this post.

 

[Justinbaileyman]

Will these be compatible with FM2+ socket or a new socket? If new socket will there be ddr4 support?

 

[Phynaz]

So much fud in this post, i bolded one of the most innaccurate statement..

Looks like the FUD was coming from AMD.

 

[jpiniero]

Will these be compatible with FM2+ socket or a new socket? If new socket will there be ddr4 support?

lol

Carrizo is BGA only. At least for now.

 

[DrMrLordX]

considering it's improved this much without a real die shrink is pretty amazing. with actual engineering resources and a real die shrink or two, this could have been a pretty interesting architecture.

Heck even Steamroller on 28nm SOI would have been something to see. It just took too darn long for AMD to fix their Construction designs. At least they have something good to show today.

Now let's see what OEMs can do with it.

 

[erunion]

lol

Carrizo is BGA only. At least for now.

Until 2016, but then it'll be called Bristol ridge.

 

[DrMrLordX]

Until 2016, but then it'll be called Bristol ridge.

That might not be Carrizo, though. Bristol Ridge, if it's still an Excavator (and it looks like it is), may not use HDL. My guess is, AMD would be criminally insane to push Excavator + HDL into a desktop socket in 2016, unless they have some trick to get clockspeeds higher.

I would expect at least a slightly different design: no HDL, possibly some other mobile-oriented features removed/disabled. They might even pull the FCH entirely and use the one native to AM4 (formerly FM3).

 

[Abwx]

Until 2016, but then it'll be called Bristol ridge.

Flooding yet another thread with your wishfull thoughts..?.

There s no Bristol ridge in the slides published but the 7th gen APU is presented as using Zen cores, starting from page 14 of Lisa Su s CG Segment Overview available here :


http://ir.amd.com/phoenix.zhtml?c=74093&p=irol-analystday

 

[Maxima1]

Flooding yet another thread with your wishfull thoughts..?.

There s no Bristol ridge in the slides published but the 7th gen APU is presented as using Zen cores, starting from page 14 of Lisa Su s CG Segment Overview available here :


http://ir.amd.com/phoenix.zhtml?c=74093&p=irol-analystday

Where does it present 7th gen APUs as having Zen cores? In the points table, it only mentions Zen for FX and instead shows efficiency as the big point for the 7th gen APU. A little further down, it shows arrows on a graph for GPU and compute. The FX is using the color they used for Zen and the 7th gen APU has the purple "scalability" color. It seems obvious that the APUs won't have Zen cores.

 

[Abwx]

Where does it present 7th gen APUs as having Zen cores? In the points table, it only mentions Zen for FX and instead shows efficiency as the big point for the 7th gen APU. A little further down, it shows arrows on a graph for GPU and compute. The FX is using the color they used for Zen and the 7th gen APU has the purple "scalability" color. It seems obvious that the APUs won't have Zen cores.

What is written in page 16 in respect of Zen..?..

For wich product is it dedicated..?.

I read "Increased CPU/APU performance" sticked with Zen written as main title, the roadmap that follow page 17 point to 7th gen APU using this core, Carrizo is marked as 6th gen in the same slide.

 

[Maxima1]

What is written in page 16 in respect of Zen..?..

For wich product is it dedicated..?.

I read "Increased CPU/APU performance" sticked with Zen written as main title,

At the top, it says, "Future Technology Innovation". I think it's much more likely only referring to the fact that future APU products will have Zen cores.

the roadmap that follow page 17 point to 7th gen APU using this core, Carrizo is marked as 6th gen in the same slide.

The roadmap only shows Zen in FX box.

Btw, didn't you see the FX arrow shoot up like a rocket on the compute side (similar to that other slide showing the IPC increase)?

 

[eRacer]

What is written in page 16 in respect of Zen..?..

For wich product is it dedicated..?.

I read "Increased CPU/APU performance" sticked with Zen written as main title, the roadmap that follow page 17 point to 7th gen APU using this core, Carrizo is marked as 6th gen in the same slide.

Just because Zen will eventually be used in APUs doesn't mean the 7th gen APU has to be Zen-based. I believe someone from the AMD team, probably Papermaster, mentioned that Zen APUs would be a 2017 product in the presentation today, possibly during the Q&A.

 

[Abwx]

The roadmap only shows Zen in FX box.

Btw, didn't you see the FX arrow shoot up like a rocket on the compute side (similar to that other slide showing the IPC increase)?

Yes but the whole mobile line up is also marked as 7th gen to replace both Carrizo and Beema/Mullins, probably that they dont want to disclose their plans completely since the FX will be first to market as proof of concept.

According to Hardware.fr article those 7th gen are Zen based and will be launched in a second wave.

Just because Zen will eventually be used in APUs doesn't mean the 7th gen APU has to be Zen-based. I believe someone from the AMD team, probably Papermaster, mentioned that Zen APUs would be a 2017 product in the presentation today, possibly during the Q&A.

I answered your point above, as said thoses chips will be launched later than the FX, seems that the server parts are in high priority, indeed AMD pointed that this was the most pressuring demand from OEMs, even before consumer products.

 

[eRacer]

I answered your point above, as said thoses chips will be launched later than the FX, seems that the server parts are in high priority, indeed AMD pointed that this was the most pressuring demand from OEMs, even before consumer products.

If server parts/performance desktop parts are a higher priority than APUs, which they should be, then launching Zen core APUs in 2017 makes sense as 2017 is later than 2016. It even allows AMD to launch a 7th Gen "Carrizo Refresh" APU lineup even before Zen desktop/server launches.

In the roadmap there is "All New" above the "AMD FX CPU", but no "All New" above either of the "7th Gen A-Series" APUs. And no "Zen CPU cores" box.

EDIT: Confirmed from my recording of the Q&A session today that Papermaster said Zen is coming to APUs in 2017.

 

[Abwx]

If server parts/performance desktop parts are a higher priority than APUs, which they should be, then launching Zen core APUs in 2017 makes sense as 2017 is later than 2016. It even allows AMD to launch a 7th Gen "Carrizo Refresh" APU lineup even before Zen desktop/server launches.

In the roadmap there is "All New" above the "AMD FX CPU", but no "All New" above either of the "7th Gen A-Series" APUs. And no "Zen CPU cores" box.

NEW CPU CORES ADD MAJOR PERFORMANCE UPLIFT

So there will be major perfs uplifts over Carrizo and Kaveri with these 7th gen APUs despite being still Excavator based..?.

Actualy there might even be one that would use a completely new Excavator based core if we are to follow your logic.

Not that a Carrizo refresh wouldnt make sense, but given that they had to push ARM based designs to 2017 i wouldnt expect them to waste ressources doping a core that will be largely outmatched by their new, and existing at this moment, design.

 

[R0H1T]

So there will be major perfs uplifts over Carrizo and Kaveri with these 7th gen APUs despite being still Excavator based..?.

Actualy there might even be one that would use a completely new Excavator based core if we are to follow your logic.

Not that a Carrizo refresh wouldnt make sense, but given that they had to push ARM based designs to 2017 i wouldnt expect them to waste ressources doping a core that will be largely outmatched by their new, and existing at this moment, design.

If it's on 14nm GF or Samsung then yes it's a real possibility, BD's Achilles heal towards the end was that AMD couldn't fit it inside a reasonable TDP. Sure the IPC was bad but there were many good things as well & it really needed a superior node to get it to where consumers would find the power bill/heat acceptable, certainly as review sites put it & ideally would make it the perfect perf/$ alternative to Intel's i3/i5 sans the IGP.

 

[AtenRa]

I dont believe they will replace Cat cores (Carrizo-L, Beema/Mullins) with Excavator.

 

[R0H1T]

I dont believe they will replace Cat cores (Carrizo-L, Beema/Mullins) with Excavator.

They're going from 28nm straight to 14nm right, so why is it hard to imagine an Excavator based mobile APU that matches the current Beema or even Mullins? They could simply put a single module & be done with it, AMD with their resources can't keep spending more on cat/construction cores at the same time. They will have to kill one or the other & depending on what their internal assessment is it might well be the cat cores, not sure I agree with that but the cat cores are really anemic given the market (netbook, convertibles, tablets?) they're currently serving.

 

[AtenRa]

They're going from 28nm straight to 14nm right, so why is it hard to imagine an Excavator based mobile APU that matches the current Beema or even Mullins? They could simply put a single module & be done with it, AMD with their resources can't keep spending more on cat/construction cores at the same time. They will have to kill one or the other & depending on what their internal assessment is it might well be the cat cores, not sure I agree with that but the cat cores are really anemic for the market (netbook, convertibles, tablets?) they're currently targeting.

If you going to port Excavator to 14nm then you can build a 14nm ZEN APU. Why spend resources, time and money to just port Excavator to 14nm when you have a better architecture ??

or

If you just port Excavator to 14nm, why spend resources, time and money to create ZEN in the first place ??