The Brazos Review: AMD's E-350 Supplants ION for mini-ITX
by Anand Lal Shimpi on January 27, 2011 6:08 PM ESTAMD has been curiously absent from the value netbook and nettop segments since Atom’s arrival nearly three years ago. These markets are highly profitable only for component vendors, as the OEMs that sell netbooks and nettops must survive on very slim margins in order to hit aggressive price points. It wasn’t too long ago that we were shocked by $699 desktop PCs, but to now be able to get a fully functioning portable PC with display at below $300 is impressive. In order for the profit equation to work out however, you can’t simply scale down a larger chip - you need an architecture targeted specifically at the type of very light workloads you expect to encounter in these segments. Underclocking and undervolting an architecture targeted at high end desktops or servers won’t cut it.
Generally a single microprocessor architecture can cover an order of magnitude of power envelopes. You can take an architecture from 10W - 100W using clock speed, voltage scaling and disabling features (e.g. cutting cache sizes). You can’t efficiently take a 100W architecture and scale it down to 1W. Intel realized this with Atom, and what resulted was a new architecture designed to span the 0.5W - 5W range. Given the constraints of the process (Atom was built at 45nm) and a desire to keep die size down to a minimum (and thus maximize profits), Intel went with a dual-issue in-order architecture reminiscent of the old Pentium - but with a modern twist.
AMD came to the same realization. For it to compete in these value markets, AMD couldn’t rely on its existing Phenom II derived architectures. The Phenom II and its relatives currently span a range of TDPs from 9W to 140W, and at the lower end of that spectrum we’re talking about some very low clock speeds and performance targets. Getting down to 1W was out of the question without a separate design.
What AMD came up with was a core called Bobcat, initially targeted for netbooks, notebooks, nettops and entry level desktops. Architecturally Bobcat is a significant step ahead of Atom: while still dual-issue, it features an out-of-order execution engine making it the Pentium Pro to Atom’s Pentium.
It isn’t just CPU architecture that AMD surpassed Atom with, the first incarnation of Bobcat is an integrated SoC with on-die DirectX 11 GPU. AMD calls this combination a Fusion APU (Accelerated Processing Unit) as it places both a CPU and GPU on a single die. The possible CPU/GPU combinations for Bobcat based APUs are listed in the table below:
| AMD Brazos Lineup | ||||||||
| APU Model | Number of Bobcat Cores | CPU Clock Speed | GPU | Number of GPU Cores | GPU Clock Speed | TDP | ||
| AMD E-350 | 2 | 1.6GHz | Radeon HD 6310 | 80 | 500MHz | 18W | ||
| AMD E-240 | 1 | 1.5GHz | Radeon HD 6310 | 80 | 500MHz | 18W | ||
| AMD C-50 | 2 | 1.0GHz | Radeon HD 6250 | 80 | 280MHz | 9W | ||
| AMD C-30 | 1 | 1.2GHz | Radeon HD 6250 | 80 | 280MHz | 9W | ||
AMD avoided branding its first APUs, they’re simply the AMD E-series and C-series Fusion APUs. The emphasis isn’t on the CPU or the GPU in this case, just the company name and a model number.
AMD's E-350
| CPU Specification Comparison | ||||||||
| CPU | Manufacturing Process | Cores | Transistor Count | Die Size | ||||
| AMD Zacate | 40nm | 2 | ? | 75 mm2 | ||||
| AMD Thuban 6C | 45nm | 6 | 904M | 346mm2 | ||||
| AMD Deneb 4C | 45nm | 4 | 758M | 258mm2 | ||||
| Intel Gulftown 6C | 32nm | 6 | 1.17B | 240mm2 | ||||
| Intel Nehalem/Bloomfield 4C | 45nm | 4 | 731M | 263mm2 | ||||
| Intel Sandy Bridge 4C | 32nm | 4 | 995M | 216mm2 | ||||
| Intel Lynnfield 4C | 45nm | 4 | 774M | 296mm2 | ||||
| Intel Clarkdale 2C | 32nm | 2 | 384M | 81mm2 | ||||
| Intel Sandy Bridge 2C (GT1) | 32nm | 2 | 504M | 131mm2 | ||||
| Intel Sandy Bridge 2C (GT2) | 32nm | 2 | 624M | 149mm2 | ||||
These APUs do need the aid of an additional chip - the Hudson Fusion Controller Hub (FCH). The FCH adds support for things like SATA, USB, Ethernet and Audio.The Hudson FCH is very tiny measuring approximately 4mm x 7mm for a total die size of around 28mm2.
The Hudson FCH
The combination of these Bobcat based APUs and the FCH is called the Brazos platform.
Late last year AMD invited me to spend several hours with a Brazos system at its brand new campus in Austin, TX. While the preview gave us some insight into what we could expect from Brazos, I didn’t have enough time to really dive in as much as I would’ve liked to.
Earlier this month, AMD officially launched Brazos with hardware expected sometime this quarter. For the past couple of weeks I’ve been testing a Brazos mini-ITX motherboard from MSI and today, it’s time to break the silence and share the results. They are quite good.
Facebook
Twitter
RSS
176 Comments
View All Comments
ocre - Friday, January 28, 2011 - link
hahaha, You need to educate yourself a little more on the advantages/disadvantages of both the x86 and ARM architectures before you go around posting ignorant statements like these. If we assumed your statement: is true: "E-350 has roughly 5 times the performance of Tegra 2" it would still be nothing to brag on becaue the same tegra 2 uses 7-9 times less energy than the E-350. This makes a mere 5 time performance a joke.But on top of that, there is no good reasonable way to conclude where these architectures rate towards one another. The software is radically different. There is not enough to go on for any useful conclusion and what little we do have is subject to a very special cases where there is SW that in the end results to similar obtained data. But the entire process is different and while we know the x86 is pretty well optimized, any ARM based counter SW is in its beginnings. X86 has the luxury of optimizations, ARM will only get better when SW engineers find better ways to utilize the system. But this is considering an actual case where there is SW similar enough to even reasonably attempt to measure x86 vs ARM performance. For the most part there the SW of each architecture is so radically different. ARM is extremely good at some things and not so good at others and this doesnt mean it cant be good, in some cases the SW just hasnt matured yet. All this matters little cause at the end of the day, everyone can see that the exact opposite from your statement; The superiority of ARM is just unmatched by x86 when it comes to performance per watt. This is undisputed and x86 has a long long way to go to catch up with arm (and many think it never will). Alls Arm has to do is actually build 18w CPUs, they will be 3 to 4 times more powerful than the e-350 based on the current ARM architecture.
jollyjugg - Friday, January 28, 2011 - link
Well your last statement is simply laughable. Because all we have from ARM right now is chips which run in smartphones and tablets. ARM based processors are not exactly known for their performance as much as their power. While it is true that performance/watt is great in ARM today, it also true as any modern microprocessor designer would say that, as you go up the performance chain (by throwing more hardware and getting more giga Hz and other tweaks like wider issue and out of order etc), the mileage you get out of the machine improves so does the drop in performance/watt. There are no free lunches here. I doubt a good performance ARM architecture will be a whole lot different that an x86 architecure. While the lower power is an achilles heal for x86 as is the performance an achilles heal for ARM. You mentioned software maturity. It is laughable to even mention this in this cut throat industry. Intel showed its superiority over AMD by tweaking the free x86 compiler it gave away to developers to suit its x86 architecture compared to its rival and the users got cheated until European commision exposed Intel. So dont even talk about software maturity. The incumbant always has the advantage. ARM first has to kill Intel's OEM muscle and marketing muscle before it can start dominating. Even if it did the former two, there is something it can never do, which is matching intel's manufacturing muscle. Intel by far is way better than even the largest contract manufacturer TSMC whose only task is to manufacture.Shadowmaster625 - Monday, January 31, 2011 - link
Nonsense. ARM is more optimized that x86. x86 code is always sloppy, because it has always been designed without having to deal with RAM, ROM, and clock constraints. When you code for an ARM device, you are presented with limits that most software engineers never even faced when writing x86 code. When writing software for Windows, 99.9% of developers will tell you they never even think about the amount of RAM they are using. For ARM it was probably 80% 10 years ago. Today it is probably less than 20% of ARM software engineers who would tell you they run into RAM and ROM limitations. With all this smartphone development going on today, ARM devices are getting more sloppy, but still nowhere near as bad as x86.Shadowmaster625 - Monday, January 31, 2011 - link
Best buy is still littered with them. Literally. Littered.e36Jeff - Thursday, January 27, 2011 - link
what review were you reading? The only bug that is actually mentioned is the issue with flash, which AMD and Adobe are both aware of and should be fixed in the next iteration of flash. Stop seeing anything from AMD as bad and Intel as good. For where AMD wants this product to compete, this is a fantastic product that Intel has very little to compete with now that they locked out Nvidia from another Ion platform.codedivine - Thursday, January 27, 2011 - link
Ok one last question. Is it possible to run your VS2008 benchmark on it? Will be appreciated, thanks.Anand Lal Shimpi - Thursday, January 27, 2011 - link
Running it now, will update with the results :)Take care,
Anand
Malih - Sunday, January 30, 2011 - link
I'm with you on this.I'm thinking about buying a netbook and may be a couple net tops with E-350, which will mostly be used to code websites, may be some other dev that require IDE (Eclipse, Visual Studio and so on).
micksh - Thursday, January 27, 2011 - link
how can it be that "1080i60 works just fine" when it failed all deinterlacing tests?Anand Lal Shimpi - Thursday, January 27, 2011 - link
It failed the quality tests but it can physically decode the video at full frame rate :)Take care,
Anand