Categories: AutoGlobal Tech News

AMD Concedes Die-Shrink Race to Intel, Considers ARM Cores

At Advanced Micro Devices Inc.’s (AMD) 2012 Financial Analyst Day in Santa Clara, Calif., the company’s latest CEO, Rory Read, outlined his vision for the firm.  The talk hit on many points, both in terms of the company’s roadmap of chip releases and in terms of its long-term technology direction.

From the upcoming Sea Islands family of GPUs to the third generation, 28 nm Bulldozer core, Steamroller, AMD’s plans are pretty diverse and ambitious.  Read on to discover more.

I. AMD an ARM Chipmaker, Soon?

Don’t expect AMD to be pumping out Cortex-A15 quad-cores this year, by AMD dropped some pretty clear hints that it was mulling testing the ARM waters.  For the uninitiated ARM is one of two main architectures that is today vying for global CPU supremacy.

Traditionally AMD and rival Intel Corp. (INTC) have supported x86, a complex instruction set computer (CISC) style of architecture.  x86 chips have dominated traditional personal computers, outshipping all other archictectures in this segment.

On the other side of the fence is ARM, a reduced instruction set computer (RISC) style of architecture.  ARM chips are produced by third parties based on the instruction set, and often, on the intellectual property (IP) core designs of England’s ARM Holdings plc (LON:ARM).  Today ARM dominates the mobile devices space and is the world’s most used CPU outshipping x86 chips in quantity thanks to its strong embedded market share.  Top mobile device ARM chipmakers in today’s market include Qualcomm Inc. (QCOM), Texas Instruments Inc. (TXN), Samsung Electronics Comp., Ltd. (KS:005930), Marvell Technology Group ltd. (MRVL), and NVIDIA Corp. (NVDA).

Microsoft Corp.’s (MSFT) Windows 8 will be the first mainline version of Windows to support ARM chips in its personal computers.  There’s also growing interest in ARM in the datacenter space, thanks to its strong power performance.

At the analyst day AMD showed off a slide, stating:

The slide is hardly subtle.  ISA in this context stands of instruction set archictecture.  So AMD is saying that it is considering non-x86 instruction sets for make SoC (system-on-a-chip) processors for datacenters, SoCs that can use “third party IP [cores].”  

Given that this explicitly describes the ARM chipmaking approach, it seems extremly likely that AMD is considering ARM server chips similar to those being produced by ARM Holdings subsidiary Calxeda for Hewlett-Packard Comp. (HPQ) (the world’s largest sever maker).

According to Anandtech‘s Anand Lal Shimpi, AMD even went as far as to name-drop ARM several times during the presentations, although stopping short of making a specific commitment to ARM.

An AMD defection (at least in part) to ARM would not exactly be surprising, given all the momentum ARM has, and the financial burden it would take off AMD’s chip-developing units.  But it would be a major blow to Intel, who would be left in the lonely position of being the lone proponent of x86, pitted against a unified alliance of virtually every other large chipmaker in the traditional and mobile personal computer space.

Could this defection spell the death of x86?  Well, it’s far too soon to declare the world’s most used traditional PC CPU architecture being put out to pasture, but this is — at the very least — a big blow to Intel in terms of confidence in x86.  It will become an even bigger blow when (and if) AMD takes its plans from paper/labs and puts out ARM product.

II. AMD Works to Unify Offerings in Core Design, Memory Access, and Process

Currently AMD makes a mix of CPUs, GPUs, accelerated procesor units (APUs: CPU+GPU), and chipsets.  While AMD discrete graphics sales have solid in recent years after AMD helped turn around its struggling ex-ATI unit, discrete graphics sales as a whole have slowed.  AMD and its GPU rival NVIDIA can blame integrated graphics — including Intel’s increasingly powerful on-die GPUs — for cutting into sales.

But AMD is looking to make GPU’s more of a value proposition by offering hassle-free advance GPU acceleration of everyday programs like web browsers and photo editing.  To some extent these technologies already exist [1][2].  But their quality is hindered by the GPU’s high memory latency, and the need to write custom logic in special APIs.

AMD’s slides describe a slow unification process between GPUs and CPUs that will help to remove the latency and specialization.  It says the first step landed last year (with Fusion) when AMD deployed chips whose on-die CPU and GPU shared the same power circuitry.

Next, this year and next year AMD will be slowly giving its GPUs greater access to the CPU’s memory pool (cache, RAM).  The final step lands in 2014, something AMD refers to as Heterogeneous Systems Architecture (HSA).  With HSA, the GPU is able to run CPU-like workloads for the first time, meaning that you won’t need to recompile or write custom logic to exploit the benefoits of GPU computing.

AMD is also looking to unify its process for various chips down to the 28 nm node by 2013.  Currently AMD offers a mix of 28 nm, 32 nm, and 40 nm offerings.

III. New Core Designs, Server Chips Revealed

AMD only plans to release one 28 nm chip family this year — the just released Southern Islands GPUs [1][2] (led by the high-end Tahiti family).  The chips is being produced on processes by Taiwan Semiconductor Manufacturing Comp., Ltd. (TPE:2330).

On its 32 nm (GlobalFoundries) node, AMD will be dropping its next-generation APU, high-power APU Trinity.  Aimed at < 18 mm ultrathins, Trinity will look to take on Intel’s Ivy Bridge.  While the Intel chip will clearly have a healthy lead on performance, AMD hopes to beat Intel on everything else — price, graphics, and power performance (the latter goal might be a bit of a struggle, thanks to Intel’s slick power-saving 3D FinFET technology).

Even if AMD can just beat Intel on price and graphics, alone, it may win the sales war (assuming it can produce enough chips).  In our recent poll 35 percent of readers said they would be more interested in a $500 USD or less Trinity ultrathin, versus only 17 percent claiming interest in a $700-$1,000 USD Ivy Bridge design.

On the 40 nm node AMD has two APUs planned — Brazos 2.0 and Hondo.  Hondo will be aimed at Windows 8 tablets, with a 4.5W TDP.  It will pack 1 – 2 low voltage Bobcat cores and an on-die DX11 GPU.  Meanwhile Brazos 2.0 is expected to see use in budget notebook and netbook designs.  It brings AMD’s TurboCore (built-in automatic overclocking) technology, along with support (at last) for USB 3.0.

On the server side, AMD will be sticking with 4/8/12/16 core Interlagos (Opteron 6200 series) chips and 6/8 core Valencia (Opteron 4200 series) chips.  It will fill in the low-end gap, offering in Q1 2012 (by March) Zurich, a 4/8 core chip series with support for only a single hyperthreading channel (vs. 2 in Valencia and 4 in Interlagos).

But in late 2012/early 2013 is where things will get really interesting.  First, in 2013 AMD plans to have another brand new GPU architecture (that’s its second in two years!), name Sea Islands.  

But on the CPU/APU front things are getting even more interesting.  Virtually all of AMD’s product line will be moving to down to 28 nm (versus its current mix of 32 nm/40 nm).  AMD will also be rolling out three new CPU cores — Piledriver (aka “enhanced Bulldozer“), Steamroller, and Jaguar.  

Piledriver will land first, sometime in late 2012 or early 2013.  Past slides indicated a 2012 launch, but it’s possible this date has slipped, given TSMC’s difficulty in scaling up 28 nm production and GlobalFoundries’ incomplete transition to the 28 nm node.  

Piledriver is the direct successor to Bulldozer, and, as is typical, will first see action in server chips.  It will drop in the Socket G34 4/8/12/16 core Abu Dhabi family of Opteron CPUs (the successor to Interlagos), Socket C32 6/8 core Seoul Opteron CPUs (the successor to Valencia), and Socket AM3+ 4/8 core Delhi Opteron CPUs (the successor to Zurich). (AMD’s Terramar and Seppang codenames are no more, replaced by Seoul and Delhi.)

Steamroller, acording to past slides, is the direct core architecture successor to Piledriver.  And Jaguar is the direct successor to Bobcat.  Steamroller will be paired with “Graphics Core Next” (GCN) GPU cores — the same kind found in AMD’s Radeon 79xx HD series — in APU designs.

IV. Consumer CPUs, APUs for 2013

AMD will four families of refreshed laptop and desktop chips (six, if you count laptop and desktop incarnations of the same APU family as separate).  The majority of AMD’s new designs are APUs.  Correspondingly, APUs will look to drive the majority of AMD’s desktop and laptop shipments in 2013.  

On the performance desktop end AMD will drop new 4/8 core Piledriver-based “FX” series CPUs.  Dubbed Vishera, these chips will be AMD’s only upcoming non-server line to not include an on-die GPU.  In other words, this is the sole non-APU design in the consumer mix.

On the opposite side of the spectrum (tablets), AMD will drop the 28 nm, 2 watt TDP, Temash APU.  AMD has not clarified whether Temash will include GCN GPU cores or an earlier design (this is an important hole in AMD’s roadmap information).  Windows 8 tablets look to be very big in holiday 2012 and 2013, and AMD is clearly hoping to hit a sweet spot in terms of price, power, and performance, staying competitive with ARM and Intel.

(Note: there is a discrepancy between slides… one spells this chip Temash, one spells it Tamesh.  The correct spelling is Temash (like the river in Belize).)

On the mid-range, AMD will deploy Kabini for laptops and desktops.  Based on the Jaguar core — the successor to Bobcat — Kabini will come in dual-core and quad-core variants.  It will pack GCN GPU cores, as well, onto a brand-new system-on-a-chip die, which will also be utilized as Temash‘s die.  The die has a built-in “Fusion Controller Hub” (FCH), AMD’s bridge circuitry that controls talk between the on-die CPU, GPU, and external I/O devices (RAM, external channels, PCIe, etc.).

Lastly, AMD’s 2013 high-end APUs will be code-named Kaveri and will come in dual-core and quad-core desktop and laptop variants.  Kaveri will pair Steamroller CPU cores with GCN GPU cores, like Kabini.  

For fans of knowing where these crazy codenames come from:
Abu Dhabi — The capital of the United Arab Emirates and richest city in the world.
Seoul — The capital and largest city in South Korea, largest metropolis in developed world.
Delhi — The capital city of India, second largest city in India, and eight largest in the world.

Vishera — A Russian river in the Ural mountains

Llano — A Texas River.
Trinity — A trio of holy figures in traditional Catholic Christianity, the name of several cities, and the name of the female protagonist of The Matrix series.
…Also, apparently, this is the name of rivers in California and Texas.
Kaveri — A large river in southern India.
Brazos — A river in Texas.
Kabini — Another river in southern India.
Hondo — Another river in Belize.
Temash — A river in Belize, a Carribean-facing Central American nation.

GPU Cores:
Northern Islands — Line of Marianas islands in the Pacific Ocean.
Southern Islands — A series of islands in Singapore.
Sea Islands — An oxymoronically entitled chain of barrier islands off the coast of South Carolina, Georgia, and Florida.
Tahiti — An French island in the Pacific.

CPU Cores:
Bulldozer — A large machine with a flat shovel scoop attach to the front to push earth.

[Image Source: (original: AnandTech; modifications: Global Tech News/Jason Mick]

Piledriver — A hydraulic piece of machinery that pushes pillars into the ground for large structures.
Steamroller — A vehicle with a heavy metal cylindrical roller, used to flatten concrete or earth.
Excavator — Similar to the “Backhoe”, this heavy machine is capable of digging with a scoop shovel.

[Image Source: (original: AnandTech; modifications: ArsTechnica]

Kabini — Another river in southern India
Temash — A river in Belize, a Carribean-facing Central American nation.
Bob — A river in Belize, a Carribean-facing Central American nation.
Temash — A river in Belize, a Carribean-facing Central American nation.

IV. Remaining Questions

Overall AMD’s strategy of pursuing accelerated development of APUs (no pun intended!) — seems wise given that they were its biggest success in 2011.  AMD has essentially conceded the die-shrink race to Intel, partially because it is out of its hands (as AMD has switched to third party fabs).

But AMD is wagering that Intel has overdelivered in the CPU department with Ivy Bridge, producing a powerful chip, but one that is too expensive to appeal to the majority of consumers.  If AMD is right (which it may be), it will buy itself time, as Intel won’t get out cheaper Atom-based 22 nm parts to 2013.  In this sense AMD will go from pitting 32 nm parts versus 22 nm parts to pitting 28 nm parts versus 22 nm parts — a slightly less bleak competition.

There are some big questions left by the roadmap, though:
+What is the timeframe for third party IP core-based server SoCs?
+Will the 32 nm Vishera drop in Q1 2013, versus Q3 2013 drops for Kabini, Temash, and Kaveri?
+Will Temash include GCN or some other graphics architecture?
+What is the status of the Steamroller-successor Excavator (Bulldozer Gen. 4)?

+Does Trinity pack a Bulldozer or a Piledriver (Bulldozer Gen. 2) core? 

Overall, AMD deserves some praise.  Despite dipping into the red, it has handled the messy transition from a first-party chip manufacturer into a fabless chipmaker better than expected.  It delivered two compelling products in 2011 — Southern Islands and Fusion.

2012 looks to be a bit of a slower year for the company, as it fleshes out its product lineup.  Probably the biggest single launch will be Trinity, as it promises strong sales if AMD and its partners can deliver high-volume sub-$500 ultrathins.  In close second will be Southern Islands, aka Radeon 7xxx HD series GPUs, which — for now — enjoy a key time-to-market lead on NVIDIA’s upcoming Kepler (GeForce GT_ 6xx/7xx) family.

2013 looks to be a far, far bigger year for AMD, though, with the new Piledriver and Jaguar families dropping early, and the Steamroller family dropping late.  Sea Islands will also land somewhere in the mix.  And, most importantly, 2013 will be the timeframe for AMD’s big exodus down to 28 nm.  It is critical that AMD, TSMC, and Global Foundries deliver on 28 nm in 2013 — their cumulative fate may depend on it, in the face of a hungry Intel.

Remember, these roadmaps, though, may change depending on how things go.

[All slides are courtesy of AMD via Anandtech.]

Recent Posts

AMD Dual-Core Optimization Utility Available

AMD Dual-Core Optimization Utility Available

Improving dual-core compatibility for gaming

5.7″ ZTE ZMAX “Phablet” Coming to T-Mobile Sept 24 for $252

ZMAX will come with a Snapdragon 400 processor and 720p display

100 Northern California Households to Receive Plug-in Priuses

UC Davis dares to go where Toyota won't with the Prius

Apple on Microsoft Ads: PCs Are “No Bargain”, Macs Are “Cool”

An Apple spokesperson fires back over Microsoft's latest commercials

Update: 13.3″ Dell XPS m1330 Notebook Details Leaked

Engadget gets the scoop on Dell's latest "ultra-portable" notebook