Shortcuts

AMD Spills New Details on Bobcat, Brazos Mobile Platform

rated by 0 users
This post has 5 Replies | 0 Followers

Top 10 Contributor
Posts 25,873
Points 1,172,935
Joined: Sep 2007
ForumsAdministrator
News Posted: Tue, Sep 7 2010 5:37 PM
In the wake of AMD's architecture discussions at Hot Chips last month, eyes have increasingly turned towards the company's upcoming Bobcat processor. Now, AMD's director of Fusion marketing, John Taylor, has written a blog post with additional details on the diminutive chip, including several photos.


We know the unveiling of this info was in Europe, but would it have killed someone to plunk down a quarter?

We've covered much of the company's plans before, so we'll skip straight to the new stuff. Taylor notes that the upcoming Fusion processor will include a "new" UVD block, though we're guessing this term refers more to the need to redesign a block for the APU as opposed to a new set of features/capabilities. AMD will introduce two new APU products this year, including:

"An 18-watt TDP APU codenamed “Zacate” for ultrathin, mainstream, and value notebooks as well as desktops and all-in-ones. And a 9-watt APU codenamed “Ontario” for netbooks and small form factor desktops and devices."

AMD notes that "Both low-power APU versions feature two “Bobcat” x86 cores and fully support DirectX11, DirectCompute (Microsoft programming interface for GPU computing) and OpenCL (cross-platform programming interface standard for multi-core x86 and accelerated GPU computing). Both also include UVD dedicated hardware acceleration for HD video including 1080p resolutions."

So, now that we know the TDP of these new devices, we can compare them to Pineview and Moorestown, right? No. AMD and Intel continue to define TDP (Thermal Design Power or Thermal Dissipation Power) two completely different ways. Intel's TDP is meant to represent average power consumption in a representative workload; Santa Clara's TDP values are the amount of power a heatsink+fan must be able to dissipate without the processor overheating. AMD's values are theoretical maximum power consumption, even if said value is much higher than any real-world workload can create.

What we can do, however, is compare the upcoming Zacate and Ontario to current AMD solutions. Presumably the TDP values quoted above refer to both the CPU and GPU cores; Taylor refers to both the power consumption of the "APU" and uses the APU codenames. At present, AMD offers the following 45nm mobile parts:

K665: (1.7GHz, dual-core, 15W)
K325: (1.3GHz, dual-core, 12W)
V105: (1.2GHz, single-core, 9W).

Keep in mind that the CPU TDPs above don't reflect the additional power consumption of a GPU. Based on what we know now, Bobcat looks as though it'll deliver improved performance performance per watt compared to current ultrathin parts. An absolute performance gain, however, is far from guaranteed. The current crop of ultrathin CPUs offer 1MB of L2 cache per core and run that cache at full processor speed—Bobcat, in contrast, will offer just 512K of L2 per core and will use an L2 clocked at 50 percent of core clock. (Note: While the half-clock cache was listed as a feature of the Bobcat architecture, there's no inherent reason we're aware of why AMD couldn't accelerate the cache to full speed, either in higher-power designs or as part of a later die shrink.

Considering Ontario's dual performance advantages compared to Atom, AMD will almost certainly target a sweet spot between power efficiency and performance relative to Intel's netbook processor as compared to its own product lines. The higher-end K-series chips referenced above are in a similar power envelope to Zacate once their own GPUs are added to the mix, which indicates it'll be Zacate that's more concerned with improving performance vis-à-vis mobile Shanghai hardware.


AMD only identifies this as a Fusion APU, but it's presumably Bobcat as opposed to Llano. AMD has taken to Photoshopping dies to mask certain features from snoopy Intel spies; for all we know this is a Sim City screenshot with rainbow colors and a big empty downtown.

Unless the chip manifests a heretofore-unknown design disaster, AMD should have no problem positioning Ontario across part of Atom's path. At the risk of sounding like broken records, we want to again emphasize the fact that Atom is designed to fit into devices and power envelopes that Ontario isn't. Remember, Intel's original design vision for Atom saw netbooks and nettops as little more than a stepping stone towards smaller and more powerful MIDs/UMPC's. Even if Ontario becomes a preferred consumer option, it'll still be taking market share in a space Intel never expected Atom to explosively populate.

AMD doesn't much care about this—when you've got around 11 percent of the mobile market, the 15-20 percent Atom currently populates looks pretty darn good. We're very optimistic about these new chips, but Ontario won't actually challenge Pineview / Moorestown in the chip's (intended) primary markets. Taylor states that AMD plans to ramp production in late 2010 with delivery in 2011—guidance that's on track with what we've heard before. 
  • | Post Points: 65
Top 10 Contributor
Posts 8,577
Points 103,200
Joined: Apr 2009
Location: Shenandoah Valley, Virginia
MembershipAdministrator
Moderator
realneil replied on Tue, Sep 7 2010 10:02 PM

Hope it works out for them, didn't Intel just release dual core Atoms?

AMD must be tired of playing catchup.

Dogs are great judges of character, and if your dog doesn't like somebody being around, you shouldn't trust them.

  • | Post Points: 5
Top 100 Contributor
Posts 1,072
Points 11,625
Joined: Jul 2009
Joel H replied on Tue, Sep 7 2010 10:15 PM

Dual-core Atom is no big shakes performance wise. All of AMD's Ontario parts will be dual-core, and dual-core Ontario will destroy dual-core Atom clock-for-clock. Due to the differences between an in-order (Atom) and out-of-order (Ontario) architecture, it's almost impossible for Ontario *not* to beat Atom in raw performance.

At the same time, Atom will always draw much less power and fit into much smaller devices. It's a trade-off.

  • | Post Points: 20
Top 25 Contributor
Posts 3,544
Points 54,465
Joined: Jul 2004
Location: United States, Massachusetts
ForumsAdministrator
MembershipAdministrator
Dave_HH replied on Tue, Sep 7 2010 10:27 PM

I'll drop a caveat in there for Joel and say, "in theory" because we haven't tested it... YET. :)

Editor In Chief
http://hothardware.com


  • | Post Points: 5
Top 500 Contributor
Posts 153
Points 1,705
Joined: Jul 2010
lonewolf replied on Tue, Sep 7 2010 10:32 PM

I think we are fast approaching a time when Intel will be again playing catch up. More and more we are hearing how the GPU is playing a more important role in computers now and in the future. People want rich multimedia and that is where AMD if they play their cards right should start to shine. Intel when then be playing catch up as they attempt to get a decent GPU out to market.

Again though on a clock for clock basis Intel is hands down champ but this is changing with every passing month.

I read an article in Computer Power User July 2008 edition and they compare an awesome GPU paired up with a crappy CPU and the result is horrible, and they do the same thing with the killer CPU paired with a crappy GPU yep you guessed it the results were atrocious.

My point is AMD has both CPU and GPU that are very respectable more so on the GPU side but their CPU is nothing to scoff at.

Stay tuned.

  • | Post Points: 5
Top 100 Contributor
Posts 1,072
Points 11,625
Joined: Jul 2009
Joel H replied on Wed, Sep 8 2010 1:01 AM

Alright, alright, I'll agree to "in theory," but I think this is about as solid a fact as we've ever been able to predict. Not only is one CPU in-order and one OoO, Atom was specifically designed for low-power. Atom's ALU and FPU are as trimmed down as they can be; the chip shares certain execution units with the FPU (despite a heavy performance penalty) because Intel focused Atom on low power above *all* else.

Another good example is the chip's lopsided L1 cache. Instead of having a matched cache (32K Instruction, 32K Data), Atom has a 24K data cache. When Intel unveiled the chip's cache sizes, the company said it trimmed 8K of data cache off to make the die smaller and lower the chip's power consumption.

Trimming 8K of cache off a modern chip is like slicing a carrot so thin that the individual slivers are translucent. The fact that Intel did it anyway is evidence of how seriously the company wanted to cut Atom's power consumption. Bobcat *isn't* tuned for the same low power consumption, and evidence of this is all throughout the architecture.

Unless AMD/TSMC catastrophically screws something up, the chips look like this:

Ontario *will* outperform Atom in absolute performance.

It *should* be able to do so while offering improved battery life compared to previous AMD chips and roughly comparable battery life compared to Atom netbooks.

AMD *may* be able to offer a Bobcat at a superior performance-per-watt ratio to that of Atom depending on how the chip's frequency and power consumption scales.

Remember, beating Atom is easy. Everything in AMD's existing product line beats Atom, everything else in Intel's product line beats Atom, and there's not much Intel can do to tweak the chip's performance--Moorestown's enhancements are almost all aimed at improving power consumption.

If Bobcat starts cutting into Atom sales, Intel could respond by poking clockspeed higher on the dual-core flavor (2GHz should be an easy jump) or by pulling the CULV options into lower price points. It could even ally itself with NVIDIA's ION to gain a strong GPU component, though I think a wee bit too much water has passed under that bridge.

  • | Post Points: 5
Page 1 of 1 (6 items) | RSS