backtop


Print 29 comment(s) - last by mindless1.. on Nov 11 at 12:51 PM


AMD's Zacate E-350 APU is pint-sized compared to AMD's monolithic previous gen V105 Geneva chip pair.   (Source: Hot Hardware)

AMD helps inform the public what an "APU" is (hint: What does GPU+CPU=?).  (Source: AMD via Hot Hardware)

Zacate adds support for two memory DIMMS (click to enlarge).  (Source: AMD via Hot Hardware)

Zacate and Ontario are only the first two members of AMD's APU invasion, which will soon hit netbooks and notebooks as well.  (Source: AMD via Hot Hardware)
APU is two Bobcat cores and a DX11 GPU

Intel's Atom processor has long enjoyed a reign as undisputed king of entry-level netbooks and ultraportables.  AMD is at last prepared to challenge that position with its long-awaited "Fusion" processor.

The chips are done and in 2011 AMD plans on launching 4 processors:



The E-xxx processors are parts in the Zacate series, while the C-xxx parts belong to the Ontario series, which is more heavily aimed at ultraportables.  Together these parts collectively belong to the Brazos family.

So is Brazos a leap forward for netbook computing or a step backward from the modest performance of Atom?  The answer is a bit of both.

Finally, a Solid IGP

The clear upside here is the integrated GPU.  In its high-end Ontario/Zacate models, AMD has packed two Bobcat cores.  Both the single-core and dual-core chips also feature an on-chip DirectX 11 GPU with 80 ALUs -- twice the number in Nile, AMD's previous integrated graphics solution.

The fact that AMD incorporated its integrated GPU directly on the same die as its CPU cores isn't exactly revolutionary, in so much as Intel already did this with its Pineview (Atom) chips launched earlier this year.  It is revolutionary in that it is the first on-chip GPU whose performance isn't horrendous and that it is the first on-die IGP to have a high-bandwidth link to the CPU (Pineview oddly opted for a slower FSB-like link on-die between the GPU core and CPU core[s]).

Intel's Pineview uses the Intel GMA 3150, which lacks dedicated vertex shader hardware.  As a result 3D performance is a miserable experience.  The GPU's overall weakness makes even playing trying to play back high definition video a painful prospect.

By contrast Brazos's Radeon GPUs should offer decent entry level gaming performance and should play HD video (including Blu-Ray) with ease. 

Memory and I/O

Pineview only supports a single 800 MHz DDR3 DIMM, Brazos offers the support for two 800-1066MHz DDR3 DIMMs.

While specifics on the smaller uni-core Ontario chips aren't available, AMD has revealed that its Zacate consists of a 19 x 19-mm, 413-ball BGA package with a 75 mm² die "advanced processing unit" (APU) (GPU+CPU) inside.  That's ever-so-slightly smaller than Intel Atom's single core entry that features a 22 mm x 22 m package and a 87 mm² die.

The APU is hooked up to AMD's chipset unit, which is named Hudson.  The Hudson chip handles part of the I/O duties, offering a wealth connections including four PCIe Gen1 lanes, four PCIe Gen2 lanes, six 6 Gbps Serial ATA connections, 14 USB 2.0 connections, and built-in fan control logic. 

Together the APU core and Hudson chipset form the Brazos platform.

The 6 Gbps SATA connections should allow for ultra fast SSD access, though it seems a bit strange to be considering pricey hard drive options with a budget-minded netbook chip.  The four PCIe Gen1 lanes can be used with the ethernet and wireless (802.11n) connections, freeing up the Brazos core's PCIe gen2 lanes for use with a discrete GPU.

Thus netbooks or mobile internet devices sporting the Brazos APU could in theory also offer a discrete GPU.  It's unclear at this point whether you could switch to the integrated GPU to decrease power consumption and extend battery life.

Power

Speaking of battery life, the one area that Pineview appears to have AMD's Brazos beat is in battery life.  Its dual core Atom N550 is clocked similar to the E-350, but has about half the power consumption.  Of course the lean power footprint is due to the garbage graphical performance.  We're not sure if that's something to brag about -- even in the mobile space.

Overall, though, when you consider chipset power (Pineview is paired with the NM10 chipset) the result is that an Atom-based netbook's internals will consume about 16 W, while AMD expects a platform TDP of 21 W for Brazos (possibly less for Zacate models).

At the end of the day AMD is claiming the E-350-equipped Zacate platform will last for about 8.5 to 9 hours on a fully charged 55Wh laptop battery, while an Ontario may get 10.5 hours.  An Atom platform netbook will last slightly longer.

Conclusion

At this point it's hard to draw definitive conclusions as a) Intel may have new Atom designs up its sleeve for 2011 and b) AMD hasn't delivered on Brazos chips yet.

That said Intel definitely has cause for concern here.  AMD's chips still are more power hungry than their Intel siblings (+5 watts for the entire platform), but the roughly 33 percent power increase reportedly allows a > 50 percent performance bump in GPU-intensive applications.  With even everyday programs like Firefox and Flash using GPU acceleration these days, this could offer faster performance even for non-gamers.

And Brazos is only the budget-minded beginning of AMD's APU invasion.  It will be followed by the Llano platform (dubbed Lynx on desktops, Sabine on notebooks), which will feature a beefier GPU and a more powerful modified K10.5 core design (not quite a Bulldozer, as some pointed out).

At this point Intel's most compelling alternative to Brazos is to pair itself with NVIDIA's ION chip.  But that approach would likely negate much of Intel's power advantage, and further it cuts into Intel's bottom line, as it would have to either cut the cost of its chips for OEMs or end up double-charging customers for their graphics.

Atom has enjoyed a virtual monopoly on the netbook world, but in 2011 it will likely meet its x86 match in Brazos.  And if that wasn't enough to keep the Santa Clara chipmaker's executives up at night, ARM processors, fresh off their tablet takeover, will likely continue to trickle into the netbook space.  And those ARM processors have the potential to blow both Atom and Brazos awy in terms of processing power-per-watt.


Comments     Threshold


This article is over a month old, voting and posting comments is disabled

Need correction!
By stmok on 11/9/2010 6:04:44 PM , Rating: 5
quote:
And Brazos is only the budget-minded beginning of AMD's APU invasion. It will be followed by the Llano platform (dubbed Lynx on desktops, Sabine on notebooks), which will feature a beefier GPU and the more powerful Bulldozer core design.


Llano does NOT use Bulldozer core . It uses a modified version of the K10.5 core . Its cache configuration is basically the same as the current Athlon II series...In fact, Llano replaces Athlon II line from mid-2011 onwards. (Delayed due to yield issues.)

A little history that I pieced together from various web sources...

Mid-2007:
AMD announces "Falcon" series APU
(Bulldozer-based; covering 10W to 100W.)

End of 2007:
AMD announces "Swift" APU
(Falcon is quietly canceled. Bulldozer core replaced by K10-based.)

End of 2008:
Swift suffers from massive yield issues with 45nm process. Project is canceled. New project (Llano) announced to replace it; aims to use 32nm process instead.

=> Swift was made from 45nm process and used Radeon HD 4350/4550 class GPU. (80 stream processors). CPU was based on K10 cores, with up to dual-cores.

=> Llano is made from 32nm process and will use Radeon HD 55xx/56xx class GPU. (480 stream processors). CPU is a highly modified variant of the K10.5 architecture, with up to quad-cores.

Squeezing a powerful GPU on-die with a complex x86 design is a painful process. Even at 32nm nodes, AMD has encountered yield issues. But it looks like they've worked them out at the cost of a 6 month delay. It forced AMD to push Ontario/Zacate forward in place of Llano. That's why those APUs are made by TSMC on 40nm process. (Intel was planning to counter Llano with the LGA1155 version of Sandy Bridge, by releasing it early in 2011. They're obviously still going forward with this plan; as AMD won't have a real answer until mid-2011 or so...Intel is trying to nullify a popular GPU-accelerated application of video transcoding, by incorporating hardware acceleration circuitry into Sandy Bridge.)

We can see though, that Swift was split into two projects that resulted in what we have today; Llano and Ontario/Zacate APUs.

The next stage for Bulldozer (currently referred to as "Bulldozer NG"), is to integrate all the I/O into the processor. GPU technology won't be incorporated until much later. (A good 5 yrs+ away.) ...At that point, you will see a hybrid solution that is a blur of x86 and GPU-like cores on the same processor die. It won't have distinct sections of x86 cores and GPU-based IGP like we see today. (Hopefully, software infrastructure like OpenCL would have matured enough to be useful!)

So it all looks like this in 2011...

=> Ontario goes after Intel Atom and Celeron.
Mini-ITX, embedded boards(?), netbooks/nettops, etc.

=> Zacate goes after Intel Pentium.
Mum/Dad boxes; affordable desktops.

=> Llano will target processors in the Core i3 and i5 areas. Mainstream desktops...Replaces Athlon II line.

=> Bulldozer will aim for Core i5 and i7 processors. Performance/Enthusiast desktops...Replaces Phenom II line.

Whether or not AMD's lines beat their Intel market equivalents is another story. I don't think the general public gives a damn if they serve their purpose without offering a sluggish experience.




RE: Need correction!
By justjc on 11/9/2010 6:26:56 PM , Rating: 3
Actually there will be a Bulldozer APU "We’ll bring our “Bulldozer” CPU cores into APUs with “Trinity”, targeted for both the mainstream and performance notebook markets. We will also offer a “Trinity” APU for mainstream desktop;" and "Also in 2012, we plan to continue offering high-performance desktop CPUs for the enthusiast market with the “Bulldozer” core-based “Komodo” CPU."

Info is from the "simply put it’s all about velocity" post on AMDs Fusion Blog where a new roadmap is presented.


RE: Need correction!
By stmok on 11/9/2010 6:54:32 PM , Rating: 2
Yeah, I just saw the 2012 roadmap after I posted...


RE: Need correction!
By IntelUser2000 on 11/10/2010 2:45:18 AM , Rating: 2
No point of integrating all the I/O. Only things like PCI Express, and memory controller are necessary. Rest probably keep the CPU from being clocked high enough and the benefits are minimal.


RE: Need correction!
By foolsgambit11 on 11/10/2010 3:55:52 PM , Rating: 2
I doubt it would really affect clock speeds. Many processors already clock different parts of the chip at different speeds. Except for certain markets, though, I don't see there being much of an advantage to integrating the I/O onto the die. I guess it might shave a couple dollars off the cost of a total system. The only usage where integration makes overwhelming sense is for very small form factor devices like cell phones and MIDs where space is at a premium.


“And I don't know why [Apple is] acting like it’s superior. I don't even get it. What are they trying to say?” -- Bill Gates on the Mac ads














botimage
Copyright 2014 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki