AMD will focus less on desktop CPUs, more on mobile APUs

Matthew DeCarlo

Posts: 5,271   +104
Staff

During its annual Financial Analyst Day last evening, AMD shared its products and initiatives for the next two years. Along with pledging to better execute its ideas, the company said it would focus on system-on-a-chip (SoC) designs, low-power solutions and heterogeneous computing. With Intel's iron grip on the conventional x86 desktop market, AMD has its sights set on the booming mobile segment.

To be clear, AMD isn't looking to dabble in the smartphone space which is "heavily crowded with low margins," but it's aiming at products such as thin-and-light notebooks and tablets. This isn't necessarily a massive shift in the company's strategy as it has gradually pushed into the mobile business in recent years. However, it's done so while struggling to remain competitive with Intel in the desktop realm.

amd direction heterogenous computing trinity hondo

Although it isn't exiting the desktop business, the company is less interested in being Intel's shadow and more interested in paving its own path. As part of its presentation, AMD showed a graph illustrating the annual growth rates of platforms. Whereas desktops and servers have growth rates of 2.8% and 5.3%, notebooks, tablets and embedded solutions are growing at a healthier 8.4%, 18.6% and 15.1%.

Conveniently enough, the company's lineup already includes an array of Fusion SoCs geared toward mobile computers. In fact, newly appointed CEO Rory Reed noted that AMD's most successful platform to date is Brazos, a Fusion range for ultraportables. It's reported that some 20 million Brazos chips were moved in the last year. As such, it wants to build on that momentum over the next 24 months.

Mobile APUs 2012-2013
The chipmaker plans to introduce at least three new APUs this year, including updated versions of Trinity and Brazos as well as a new ultra-low-power SoC dubbed "Hondo." The second-gen 32nm Trinity parts will remain in the performance segment packing two or four Piledriver CPU cores and a DX11 GPU. They'll span various power ranges, including standard 35W models and low-power 17-25W variants.

amd direction heterogenous computing trinity hondo

Brazos 2.0 will bring revamped C and E-series chips with native USB 3.0, two 40nm Bobcat CPU cores and a DX11 GPU in the 9-18W range -- the same general TDP as now, but with more performance. Scaling down further, Hondo will consume about 4.5W while packing one or two Bobcat cores and a DX11 GPU. Hondo is designed for Windows 8 tablets and should appear in such products later this year.

amd direction heterogenous computing trinity hondo

AMD plans a full 28nm refresh for 2013 with "Kaveri," "Kabini" and "Tamesh." Kaveri will have two or four "Steamroller" (evolved Bulldozer/Piledriver) CPU cores, Kabini will carry two or four "Jaguar" (evolved Bobcat) CPU cores and Tamesh will pack two Jaguar cores with ULP iterations dipping below 2W. All of them will have next-gen GPUs, while Kabini and Tamesh will have an integrated southbridge.

Desktop CPUs/APUs 2012-2013
Again, although AMD is honing in on the mobile sector, it won't completely ignore desktops. This year will bring second-gen 32nm FX-series processors codenamed "Vishera" with four to eight Piledriver cores along with 32nm Trinity-based A-series APUs and 40nm Brazos 2.0-based E-series Fusion chips. Likewise, there will be desktop iterations of the 28nm Kaveri and Kabini APUs sometime in 2013.

amd direction heterogenous computing trinity hondo

Server CPUs 2012-2013
AMD's server roadmap for the next two years isn't particularly exciting. Interlagos, a high-end Socket G34 chip with up to 16 cores, will hold down the 2P and 4P enterprise segment throughout 2012. "Valencia" (already out with Interlagos) connects via Socket C32 and carries six or eight Bulldozer cores, while "Zurich" (due this quarter) is the lower-end Socket AM3+ model with four or eight Bulldozer cores.

amd direction heterogenous computing trinity hondo

The following year will bring Piledriver-based versions of the chips above (same sockets), codenamed "Abu Dhabi," "Seoul" and "Delhi." The flagship will peak at 16-core offerings, despite previous talks about 10 and 20-core solutions dubbed "Sepang" and "Terramar." Although the server CPU lineup is somewhat dull, Reed said the company would take advantage of opportunities in cloud computing.

Graphics 2012-2013
While AMD's 28nm APUs are still a year or more away, the company is already shipping Southern Islands (Radeon HD 7000 series) GPUs based on the fabrication process. These will be followed up by another 28nm series, known as "Sea Islands," in 2013. Along with the usual performance increases, the next Radeons will pack Heterogeneous Systems Architecture (HSA) enhancements for compute tasks.

amd direction heterogenous computing trinity hondo

Heterogeneous Computing
As you'll notice by the increasing lack of traditional client processors, AMD wants to continue blurring the line between CPUs and GPUs. The company spoke at lengths about heterogeneous computing and how it will make life easier for developers because programs won't have to bounce data between the CPU and GPU. This unification is part of the vision AMD had in 2006 when it purchased GPU-maker ATI.

That hasn't happened overnight, obviously. It took five years before AMD released its first Fusion APUs. However, the company outlined an extensive game plan moving forward. At first, its heterogeneous system architecture will allow GPUs to access the CPU's main memory. By next year, the CPU and GPU will share a fully unified memory architecture, while 2014 will bring full GPU compute context switching.

Too Long, Didn't Read
AMD has drafted an ambitious plan for the coming years that it hopes will hoist it to leadership status in the mobile and embedded segments. It will emphasize improved power management, increasingly consolidated SoC designs and the continued marriage of CPUs and GPUs. Together, those factors should produce APUs that are affordable, power-efficient and more graphically endowed than Intel's offerings.

amd direction heterogenous computing trinity hondo

**Slide images via Anandtech

Permalink to story.

 
yeah but if there is no competition for intel they dont need to release new processors so now its boring and + they can up the prices ....... monopoly they will release ivy bridge after that i doubt they will release a new processor in 1 1/2 years ......
 
yeah but if there is no competition for intel they dont need to release new processors so now its boring and + they can up the prices ....... monopoly they will release ivy bridge after that i doubt they will release a new processor in 1 1/2 years ......
So be it!! The software world needs to catch up with the processing power of today anyway. Honestly what percentage of consumers need more processing power than we currently have today? I will welcome the wait if it allows AMD a chance at a second wind.
 
First slide, "Skating to where the puck is going". Nice to see a great Canadian's influence. ;)
 
Guest said:
yeah but if there is no competition for intel they dont need to release new processors so now its boring and + they can up the prices ....... monopoly they will release ivy bridge after that i doubt they will release a new processor in 1 1/2 years ......

Unfortunately your hypothesis was debunked when the Ivy Bridge price sheet was leaked.
 
Princeton +1

Intel well not stop or slow down because they're trying to enter the smartphone/tablet market. They can't have some weird price/performance rift between their chips for phones and desktops.

Furthermore, ARM is entering the desktop and mobile market; so Intel definitely can't rest.
 
why do they feel they need to do this? i mean, its great to expand your business, but why would shift from what has been the focus of your business for years.
AMD processors aren't bad, don't know why they are doing this. FX was first time on a brand new architecture that even Microsoft wasn't prepared for. Piledriver will be better, especially when paired with windows 8. But whatever AMD. hopefully you will realize the tablet/slimbook market isn't as big as you think.
 
Guest said:
yeah but if there is no competition for intel they dont need to release new processors so now its boring and + they can up the prices ....... monopoly they will release ivy bridge after that i doubt they will release a new processor in 1 1/2 years ......

If they take that tact, then Intel will be out of business in 2 years... As others have commented, the fact is that Intel has more competition than it likes, but it's not the same as the old "AMD vs Intel" x86 war days. Intel might get to beam as top dog while AMD concentrates on other areas, but that is actually more bad news than good for Intel. ARM is becoming the biggest threat to Intel's business model right now, and AMD's new direction will have more in common with ARM's strengths and models.

We're in the midst of a big swing, coming back down from the "brute force" processing days and getting back to focusing on efficiency (both power and processing). Intel (and AMD) reached heady heights by throwing so much raw power into processors that it became easy to do tasks sloppily - even if it was not the best way to do something, there was always enough power to shove the square peg into the round hole in a program. Decreasing sizes, increasing interest in mobility and portability, and pushes towards more "green" in general are threatening the old x86 "more power, Scotty!" approach to things. Intel needs to stay relevant, by constantly improving and refining their products so that they are still desirable. A "monopoly" does no good, if you are the sole provider of a product nobody wants or needs anymore.
 
AMD processors aren't bad, don't know why they are doing this.
• Time to market slippage
• <5.5% market share in server (Orochi/Piledriver might look to retain parity in comparison with Intel's previous gen...unfortunately for AMD, time, tide and Intel wait for no man)
• Lack of R&D funds
• A management that lacked market direction
• A company that bet the farm on Globalfoundries process execution/ramping abilities. Too late at 28nm, Too late at 20nm, Too late realizing that gate-first was an evolutionary dead end.
FX was first time on a brand new architecture that even Microsoft wasn't prepared for.
Nice fairy story.
Bulldozer's development pre-dates Windows 7's. All AMD had to do was work with MS during the OS's development in the same way that MS and Intel already co-operate to have hyperthreading viable. The fact that AMD saw no need to involve MS during Bulldozers gestation points to either shortsighted management, a basic change in BD's architecture during the design phase, or AMD not knowing exactly how their product would behave.
Piledriver will be better...
It certainly better be. A more mature architecture on a more mature process. You'd have to be the Keystone Kops to screw that up. Do you think that Intels next offerings might also be better ?
 
dividebyzero said:
AMD processors aren't bad, don't know why they are doing this.
• Time to market slippage
• <5.5% market share in server (Orochi/Piledriver might look to retain parity in comparison with Intel's previous gen...unfortunately for AMD, time, tide and Intel wait for no man)
• Lack of R&D funds
• A management that lacked market direction
• A company that bet the farm on Globalfoundries process execution/ramping abilities. Too late at 28nm, Too late at 20nm, Too late realizing that gate-first was an evolutionary dead end.
FX was first time on a brand new architecture that even Microsoft wasn't prepared for.
Nice fairy story.
Bulldozer's development pre-dates Windows 7's. All AMD had to do was work with MS during the OS's development in the same way that MS and Intel already co-operate to have hyperthreading viable. The fact that AMD saw no need to involve MS during Bulldozers gestation points to either shortsighted management, a basic change in BD's architecture during the design phase, or AMD not knowing exactly how their product would behave.
Piledriver will be better...
It certainly better be. A more mature architecture on a more mature process. You'd have to be the Keystone Kops to screw that up. Do you think that Intels next offerings might also be better ?
Intels offerings have almost always been better, just not for bang to buck wise.
I see your point of the AMD/microsoft development stage

and what does market share have to do with the performance of a processor? LOL.
 
ikesmasher said:
and what does market share have to do with the performance of a processor? LOL.

Because the market votes on what is the superior product? If AMD CPUs were faster then they would have a large market share, because people would buy them. That's basic economics.
 
and what does market share have to do with the performance of a processor? LOL.

Small market share -> low revenue -> decreased R&D funding.

The latter constitutes employee layoffs, time to market/ timetable slippage (see my earlier post), and contraction of product lines. For example, this is the difference between what AMD hoped for with BD, and what the reality is (and no, just because AMD have added a few new names -Abu Dhabi/Seoul/Delhi- it still amounts to putting lipstick on a pig. No socket change, no chipset change, and no Steamroller in CPU for client or server)

Timing is everything. If Bulldozer had entered production when it was supposed to (i.e. two years ago when AMD were targeting Core 2 Quad and Nehalem) they'd have a sure-fire winner. Now, not so much.
So, basically, that's what market share has to do with the performance of a processor

EDIT: what princeton said
 
It's really quite simple if you want Intel to have competition in the X86 market you need to purchase AMD processors just giving lip service to the competition mantra is not enough.For many years I have purchased and used AMD products in almost all the builds I do whether for myself or customers and have never had a complaint about performance even on top of the line gaming rigs,so set up people put your business where your mouths are for once!!
 
Bull the sheeple follow the best advertising .....how else do you explain Apple same machines for twice the price.
 
So.. if that were the case alone GM would still be king and not Toyota.. etc etc...

Sorry but things do NOT always remain the same.
 
It's really quite simple if you want Intel to have competition in the X86 market you need to purchase AMD processors
Why ? You think that somehow a few people buying a few CPU's makes a difference? AMD has damn near 50% of the discrete craphics market -in the last year they count sales of HD 5000, HD 6000 and HD 7000 (the latter shipped for revenue in Q4 2011)- THREE successful and class leading series of cards...and what was AMD's operating income last year? $51m...a 67% decrease from 2010.

Just a note. If you could convince 10,000 TechSpot members to go out and buy an AMD processor right now, AMD still wouldn't recoup enough profit to recoup the $250,000 they gave Rory Read in relocation expenses.

BTW: AMD's main failing isn't in the client markets, it's in the OEM and server area's -as well as debt servicing and the GF millstone around their necks-. Your "good enough" mantra cuts little ice with people who measure productivity by performance-per-watt and performance-per-core (where a lot of software is licensed on a per-core basis).
 
@Guest GM is once again the #1 car maker due to China sales and Toyota is #3. On the not quite so high end, I found my newly built AMD X2 computer to start up much faster than my 3 year old Intel Dual Core, almost as if it had an sdd, which it doesn't. Is this AMD magic, or just ddr3 memory? The Intel was never this fast.
 
@Guest GM is once again the #1 car maker due to China sales and Toyota is #3
The guest's analogy probably stands, even if the example doesn't.
Assuming that greater competition should be the rationale for buying a vehicle, shouldn't wizardB be pleading with everyone to buy Chrysler, or Saab, or Alfa Romeo ?
On the not quite so high end, I found my newly built AMD X2 computer to start up much faster than my 3 year old Intel Dual Core
A new Athlon build POST's faster than a four-and-a-half year old G31 powered by a old Pentium that pre-dates even that ?....you wouldn't be pulling our collective legs would you? Next thing you'll be telling us that the Athlon's discrete graphics are better than the Intel GMA 3100.
 
No, AMD don't give up... we don't need another monopoly here, I mean it's a known fact that Intel is faster than AMD, but come on, if there is no competition of prices what will happen to the processor market (just look at the 3930K vs the 3960K, Intel charges over $600 extra to get just a relatively little performance boost). Ivybridge will be way to overpriced compared to the pricing of the i7 2000 series processors if there is no competition (and I wonder if they'll even change the prices for the 2000 series).
 
No, AMD don't give up...
AMD aren't going anywhere. They are however looking at the high volume entry- mainstream markets
we don't need another monopoly here, I mean it's a known fact that Intel is faster than AMD, but come on, if there is no competition of prices what will happen to the processor market (just look at the 3930K vs the 3960K, Intel charges over $600 extra to get just a relatively little performance boost).
That's a joke right ? You realise the last time AMD footed it with Intel at the same (or greater) level of performance they were charging $700 to $800 to $1000.

Now, if AMD are such an altruistic company- who by their very existance, supposedly ensure lower prices, why are they recommending a $549 MSRP for the HD 7970 ? Surely, if Rory Read is the epitome of the modern humanitarian, wouldn't AMD have priced the card at say $399 (or even price at GTX 580 MSRP) and ensured the same level of supposed competition ?

Nah! You know what's going to happen ? Exactly the same as what happens every other time with every other IHV: Wallet-rape the early adopters and lower prices when you're forced to. AMD have adopted cheap pricing (CPU) simply because they lag in every performance metric. Years of tinkering with an old architecture and offering it for cheap -banging the value-for-money drum, and setting themselves as some kind of tech moral compass -which in all actuality is more a product of management's deficiencies in strategic planning.
Ivybridge will be way to overpriced compared to the pricing of the i7 2000 series processors if there is no competition (and I wonder if they'll even change the prices for the 2000 series).
Whoops. There goes another poorly researched comment. So who do you like in Superbowl XLVI...Cleveland ?
 
AMD has mostly delivered the better product contrary to "reviews" and BS benchmarks.
 
I fully support AMD and have done so ever since the Athlon CPU (Slot A) came to the market. They beat Intel to 1 GHz when Intel was at the Pentium III stage. The Pentium IV was a joke and during that time AMD excelled with the Athlon XP series. I must have built over a dozen computers since 2000 and each and every one runs with an AMD CPU. Why? Because Intel is a corrupt company that uses illegal marketing practices both here and in other countries. I hope I never have to resort to building a computer with an Intel CPU. My latest Bulldozer machine is using the FX8150 and I am very pleased with it's performance. I hope AMD continues to be a major player in the desktop market.
 
My latest Bulldozer machine is using the FX8150 and I am very pleased with it's performance. I hope AMD continues to be a major player in the desktop market.
I can say the same thing about Intel but that doesn't mean I am going to downgrade the other player. If you like AMD fine, I'm happy for you but shut up about Intel, especially when you say you haven't used their CPU's at all the last 12 years.
 
Back