AMD Mobile "Carrizo" Family of APUs Arrive in 2015

[btarunr]

AMD (NYSE: AMD) today at its Future of Compute event announced the addition of its first high performance system-on-a-chip (SoC), codenamed "Carrizo", and a mainstream SoC codenamed "Carrizo-L" as part of the company's 2015 AMD Mobile APU family roadmap. In collaboration with hardware and software partners, these new 2015 AMD Mobile APUs are designed as complete solutions for gaming, productivity applications, and ultra high-definition 4K experiences. With support for Microsoft DirectX 12, OpenCL 2.0, AMD's Mantle API, AMD FreeSync and support for Microsoft's upcoming Windows 10 operating system, the 2015 AMD Mobile APU family enables the experiences consumers expect.

"We continue to innovate and build upon our existing IP to deliver great products for our customers," said John Byrne, senior vice president and general manager, Computing and Graphics business group, AMD. "AMD's commitment to graphics and compute performance, as expressed by our goal to improve APU energy efficiency 25x by 2020, combines with the latest industry standards and fresh innovation to drive the design of the 2015 AMD Mobile APU family. We are excited about the experiences these new APUs will bring and look forward to sharing more details in the first half of next year."

The flagship "Carrizo" processor will integrate the new x86 CPU core codenamed "Excavator" with next generation AMD Radeon graphics in the world's first Heterogeneous Systems Architecture (HSA) 1.0 compliant SoC. The "Carrizo-L" SoC integrates the CPU codenamed "Puma+" with AMD Radeon R-Series GCN GPUs and is intended for mainstream configurations. In addition, an AMD Secure Processor will be integrated into the "Carrizo" and "Carrizo-L" APUs, enabling ARM TrustZone across the entire family for the security commercial customers and consumers expect. Utilizing a single package infrastructure for "Carrizo" and "Carrizo-L," the 2015 AMD Mobile APU family simplifies partner designs across a broad range of commercial and consumer mobile systems.

"Carrizo" and "Carrizo-L," are scheduled to ship in 1H 2015, with laptop and All-in-One systems based on the 2015 AMD Mobile APU family expected in market by mid-year 2015.

View at TechPowerUp Main Site

 

[eidairaman1]

Waiting for the trash talk in 3...2...1...

 

[Over_Lord]

Waiting for the trash talk in 3...2...1...

This is going to be great (I see they are still on the 28nm process, in mid-2015, so... great!).

 

[Winston_008]

hmm I thought they would have gone 20nm for this.

 

[TRWOV]

I hope we get an AM1 version.

 

[Jstn7477]

Waiting for the trash talk in 3...2...1...

Waiting to see what this brings to the table, actually. Hopefully AMD is on track to start bringing higher performance CPUs (with possibly better power consumption too) to the masses.

 

[Steevo]

Did they fix their latency issues in the cache to help improve IPC?

 

[The Von Matrices]

I know this is a mobile/BGA release, but would this article suggest that either the FM2+ socket or AM1 socket is dead?

If both dies can be in the same package on mobile devices, then why would there be a need for two sockets on desktops?

 

[john_]

If AMD had money for R&D, this would have been the 2nd or 3rd generation of APUs, not the 5th.
Anyway Carrizo looks promising even without the HBM. Excavator and color compression will help both cpu and gpu performance. The pin compatibility between Carrizo and Carrizo-L is also a big plus and it looks like this is the future way of AMD building new SOCs. Remembering that the ARM SOC and the x86 equivalent will also be pin compatibles, maybe in a couple of years AMD will have one socket where someone will be able to insert anything from 2W TDP to 150W, ARM or x86 based.

 

[FordGT90Concept]

28nm? Srsly? Intel already did this on 22nm.. It would be interesting to see how they stack up but I'm pretty sure we all already know who will win.

Remembering that the ARM SOC and the x86 equivalent will also be pin compatibles, maybe in a couple of years AMD will have one socket where someone will be able to insert anything from 2W TDP to 150W, ARM or x86 based.

Very, very unlikely. These chips are tiny compared to their desktop or even laptop brethren. What you're talking about is like trying to move 80,000 lbs of cargo with a Peel P50.

 

[dj-electric]

28nm? Srsly? Intel already did this on 22nm.. It would be interesting to see how they stack up but I'm pretty sure we all already know who will win.



Very, very unlikely. These chips are tiny compared to their desktop or even laptop brethren. What you're talking about is like trying to move 80,000 lbs of cargo with a Peel P50.

Well. You know how things are.

If anybody today in 2014 thinks that AMD can compete with Intel over power and effiecency - he is being completely delusional.

I really hate seeing things go that way, but intel is so far ahead, with the sounds of the construction workers here building the world's first 10nm facility.

Yes - you can offer decent power, decent price and a fairly good platform. Just don't expect it to survive much longer with a sad technological infiriority. Intel is marching so fast forward that today you have science-fiction tier of power\effiecency ratio chips like the M5Y70. I am somewhat concerned about AMD's future in this category.

 

[buildzoid]

And I just got a Kaveri based laptop(it kicks ass). The biggest issue AMD mobile parts is that they are really really hard to find compared to intel ones. My laptop completely destroys it's intel based counterpart(intel pentium + Nvidia GT 820M) but this is the only laptop in the price bracket based on Kaveri everything else is pentiums and i3s.

 

[EarthDog]

Waiting to see what this brings to the table, actually. Hopefully AMD is on track to start bringing higher performance CPUs (with possibly better power consumption too) to the masses.

This is an apu...


Regardless, looking forward to see what it brings to the table!

 

[RCoon]

28nm? Srsly?

AMD are focusing all their 20nm gear for consoles first and foremost to keep people (console makers) happy. Then console maker can sell millions more "slim" versions of everything. Not that AMD make any profit from it at all though. Next gen consoles will probably run intel, or even other chip makers...

 

[Crap Daddy]

Well. You know how things are.

If anybody today in 2014 thinks that AMD can compete with Intel over power and effiecency - he is being completely delusional.

I really hate seeing things go that way, but intel is so far ahead, with the sounds of the construction workers here building the world's first 10nm facility.

Yes - you can offer decent power, decent price and a fairly good platform. Just don't expect it to survive much longer with a sad technological infiriority. Intel is marching so fast forward that today you have science-fiction tier of power\effiecency ratio chips like the M5Y70. I am somewhat concerned about AMD's future in this category.

This. Sadly all bases are covered by Intel. Power and efficiency is all that matter, all devices need to be thinner, lighter, longer battery life.

 

[GhostRyder]

Intel does not have a good GPU which is where these things shine. If these turn out the way they are being described the mobile market will get some updated devices with more power to handle their media machines without sacrificing serious battery life and expense of having a secondary GPU handle this type of power.

 

[Jorge]

Wow... It's amazing the foolish comments by some who fail to invest a few minutes to educate themselves.

Carrizo uses Excavator cores which have up to 30% greater IPC over Steamroller cores in addition to a host of other architectural changes and improvements. Smaller node size primarily buys you lower power consumption yet AMD has drastically lowered the power consumption on Carrizo even at 28nm. As we will see Carrizo is a significant step forward in APU performance.

http://wccftech.com/amd-launching-carrizo-apus-december-features-gen-gcn-excavator-cores/

 

[Aquinus]

Wow... It's amazing the foolish comments by some who fail to invest a few minutes to educate themselves.

Carrizo uses Excavator cores which have up to 30% greater IPC over Steamroller cores in addition to a host of other architectural changes and improvements. Smaller node size primarily buys you lower power consumption yet AMD has drastically lowered the power consumption on Carrizo even at 28nm. As we will see Carrizo is a significant step forward in APU performance.

http://wccftech.com/amd-launching-carrizo-apus-december-features-gen-gcn-excavator-cores/

I will believe it when I see it. I'll wait to see what the numbers say instead of making any erroneous claims. I have reservations about 30% in single threaded tasks, but I could believe that if it's in multi-threaded workloads. I feel like I've heard that before and the real numbers came up short of those expectations.

I'll leave this at: I'm glad AMD is rolling out something new but we should all wait and reserve judgement for how it actually turns out instead of running on the expectations we already have.

 

[HumanSmoke]

AMD are focusing all their 20nm gear for consoles first and foremost to keep people (console makers) happy. Then console maker can sell millions more "slim" versions of everything. Not that AMD make any profit from it at all though. Next gen consoles will probably run intel, or even other chip makers...

Almost, but not quite. AMD receives around $100/$120 per console SoC, and according to AMD's own CFO, Devinder Kumar, margins are 15% after manufacturing and packaging by TSMC, so $15/$18 per console depending on which one we're talking about.

Wow... It's amazing the foolish comments by some who fail to invest a few minutes to educate themselves.

Speaking of which...

Carrizo uses Excavator cores which have up to 30% greater IPC over Steamroller cores...

AMD have said no such thing. What AMD have said is that at 15W, Carrizo offers 30% better performance than comparable possible/probable(?) Steamroller based APU - note that the comparison doesn't specify what Excavator is being compared to...nor the workload where the 30% is achieved. Given that Carrizo features a AVX 256-bit instruction set, a tailored bench suite could easily produce a 30% increase while overall performance is incremental over the previous architecture. The same case can be made for Intel's inclusion of AVX in its own processor line. Dominant performance using AVX, marginal increase without.

 

[Nordic]

I am most interested to see what they can leverage with HSA 1.0

 

[Aquinus]

I am most interested to see what they can leverage with HSA 1.0

Me too. Shared memory is a huge plus for this. The problem is that leveraging those GPU cores needs to be easier. Having to compile a kernel (application) to run on the GPU every time you want to do something on it is absurd for OpenCL applications. As a developer, I need better constructs for developing software that can leverage this technology. Working on a server, it's easier for me to leverage a multi-processor server than a server with GPGPU-like components. If they want this to take off, working with the GPU side of things needs to be easier than it is now.

 

[Frick]

Waiting for benchmarks with great anticipation. It can only be better.

 

[TheGuruStud]

28nm? Srsly? Intel already did this on 22nm.. It would be interesting to see how they stack up but I'm pretty sure we all already know who will win.



Very, very unlikely. These chips are tiny compared to their desktop or even laptop brethren. What you're talking about is like trying to move 80,000 lbs of cargo with a Peel P50.

Intel still prefers marketing over actual science. Their 22nm bulk shit (finfet) isn't actually 22 (more like 26). Die size doesn't add up.

Lithography is irrelevant, now. We're not going to receive any performance boost from it. They sell you fancy lower numbers b/c finfet is cheap to produce and every time they can go a little smaller it makes them more cash. High performance is actually dead (see power consumption when OCing).

 

[Hayder_Master]

this manger can't even convinces me to buy a sandwich when i am starving.

 

[Aquinus]

Intel still prefers marketing over actual science. Their 22nm bulk shit (finfet) isn't actually 22 (more like 26). Die size doesn't add up.

Bingo. Intel and AMD measure this differently and as a result a 22nm Intel CPU would not be the same thing as a 22nm AMD CPU forgetting the fact for a moment that AMD CPU IIRC are SOI and Intel's are HKMG which are two very different processes to build microprocessors. I don't know if this hold true for APUs though because Radeons have traditionally been done on a HKMG process like Intel's CPUs.

All in all. The real point is that we want to see what this sucker can do.