The Brazos Performance Preview: AMD E-350 Benchmarked
by Anand Lal Shimpi on November 16, 2010 12:01 AM ESTFinal Words
The Brazos platform really poses the question of what is fast enough from a CPU standpoint. Netbook makers often argued that Atom was fast enough, and honestly they'd be right if Atom wasn't paired with such a heavyweight OS. Running Windows 7, Atom just wasn't fast enough but many made the sacrifice in order to get the form factor and battery life benefits that went along with the platform. The E-350 offers an alternative. You get faster than Atom CPU performance (particularly in single threaded tasks) and a fairly potent GPU. The only issue is that the E-350 doesn't compete against Atom. Most of the time you'll find it up against Pentium or low end Core i3 notebooks.
From a CPU performance standpoint, the E-350 can't compete with either of those CPUs clock for clock. Where AMD has the advantage is in power consumption and GPU performance. This is effectively AMD's answer to Intel's CULV platform, but with better graphics performance.
This is a huge statement from AMD, echoing what Apple has been saying for the past year: while CPU performance matters, GPU performance must scale at least as quickly, if not more to make up for lost time. The only thing we're really lacking are the killer apps (outside of 3D gaming and HD video) to support this mentality today.
The Radeon HD 6310 in the E-350 does very well, despite the memory bandwidth limitations. Its easily faster than Intel's HD Graphics, although some games (e.g. Starcraft 2) are still held back by the performance of AMD's Bobcat cores. Despite being a significant step forward for integrated graphics at this price point, I would be lying if I said I didn't wish it were faster. Ideally I would like to see > 30 fps performance across the board from the E-350.
There's also the GPU compute argument that AMD makes for Fusion, however the GPU compute tests I tried to run on Brazos didn't exactly work. Cyberlink Media Espresso would not use the GPU for encoding and the Civilization V compute shader test wouldn't run either. Both I suspect will be worked out by the time the platform actually ships.
What matters the most with Brazos is what the OEMs do with it. We'll of course get mini-ITX boards to turn into HTPCs, but on the mobile side Brazos needs to be paired with a sleek/well built chassis, good display, fast storage and a large battery to truly be attractive. I am also concerned about the performance of the Ontario based APUs. In many areas the E-350 isn't that much quicker than a dual-core Atom, which makes me wonder how the C-30 and C-50 would do.
Brazos ultimately has the same problem Atom had at its introduction. Compared to similarly priced notebooks based around 2.2GHz Pentium dual-core CPUs or Core i3s, the CPU just isn’t competitive. Unlike Atom however, in GPU bound scenarios the E-350 is actually faster than those two. The only issue, as I mentioned above, is that currently the majority of mainstream applications that are GPU bound are 3D games. Like Atom however, the Brazos platform is also destined for ultra thin formfactors - places a low end Core i3 can’t go. It’s in that comparison that AMD is most successful.
Compared to a Core i3-330UM, Brazos offers much better GPU performance and it’s even competitive in CPU dependent games like Starcraft 2. The i3-330UM is still faster CPU bound scenarios though, reinforcing the CPU speed for GPU performance tradeoff that you’re faced with when considering these two.
The E-350 delivers 50 - 60% of the multithreaded CPU performance of the Core i3-330UM, and nearly 70% of the single threaded performance. It’s a noticeable gap, but AMD attempts to make up for it by delivering up to twice the GPU performance in games.
Then there’s the issue of die size. For years AMD has had to give sell you a larger die at a lower price than Intel in order to remain competitive. The Brazos platform changes that. At 75mm^2, the Zacate die is smaller than just the CPU portion of a Core i3 and it’s less than 40% of the total die area when you include Intel’s HD Graphics. While more CPU performance would be nice to have, this is a good start.
AMD is committed to revving the Brazos platform yearly. We’ll obviously see updated graphics next year but I’m hoping for updates to the CPU cores as well. The biggest mistake Intel made with Atom was to be too conservative with its roadmap. AMD has a history of not making the same mistakes as Intel, so hopefully the Brazos roadmap looks good. Now we wait for the notebooks.
Facebook
Twitter
RSS
207 Comments
View All Comments
svojoe - Wednesday, November 17, 2010 - link
I'm sure it was not easy to do a full suite of testing, But I would *really* like to see the Brazos pitted against the ATOM/ION or the CULV/ION Platforms. Despite atom's really slow performance for the same price ranges even today I am still amazed that my single core ATOM (hp mini 311) paired with ION is hard to beat in the 11.6" and below form factor for multimedia and gaming at a sub $400 price tag.I know that I am a small niche in the market I want netbook/ultraportable form factors that are potent at gaming/multimedia that are also cheap. Can the Brazos answer in this category too or will the dual core ATOM/ION system's still rule when it gets down to raw cost and fps?
Cloakstar - Wednesday, November 17, 2010 - link
The chip may be memory bandwidth limited, but AMD was kind enough to demonstrate what now is best termed as a ~30% memory overclock at IDF. As long as the retail mobtherboards have even a few clock options, the raw TDP looks like it should provide a ton of headroom for the rest of the chip to clock up to consume that memory bandwidth.Brazos TDP: 21W ... Load power: 6.5W
"The Brazos platform was configured with 4GB of DDR3-1066 memory. The IDF system had memory running at DDR3-1333, however AMD had to decrease clocks presumably to meet validation requirements for final silicon."
"The Radeon HD 6310 in the E-350 does very well, despite the memory bandwidth limitations."
silverblue - Wednesday, November 17, 2010 - link
I put 2GB extra in my PC the other week and had to up a couple of the timings to make it work. I wonder how memory timings will affect the GPU's performance.Without a full rundown on what the system had, we won't know if AMD could've put better RAM in there, for example. I doubt they'd be sandbagging.
GeorgeH - Wednesday, November 17, 2010 - link
They're only a good value option because Intel doesn't care to offer their superior products with similarly thin profit margins.If Intel were to release a P4 for $0.50, it would be a fantastic value option for a lot of applications and users, but that doesn't change the fact that a P4 core is substandard when compared to a Nehalem core.
GeorgeH - Wednesday, November 17, 2010 - link
I wouldn't call it a myth so much as slightly disingenuous. Painting with a broad brush, there are two types of compute tasks - those that are arithmetically intense and those that aren't. Typically those that aren't are integer/boolean, and the rest are float. GPUs are great with most arithmetically intense operations, so they get conflated with float problems while leaving CPUs for the rest.It isn't completely precise to say CPUs are for integers and GPUs are for floats, as both can work with and excel at using either, but it's a good approximation when you start talking about real-world tasks a typical user will want to complete that take a significant amount of time to run.
krazyderek - Wednesday, November 17, 2010 - link
by the looks of things, AMD has made a good cut in power usage, at the expense of looking CPU competitive.Why don't they just release a E-460 with 3 bobcat cores at 1.6ghz, or 2 cores at 1.8ghz that will leave Atom in the dust and truly compete with i3 ?? looks like that would only take it up to ~27w which would be on par...
maybe it's just me, but when something new comes out, i don't care if just battery life has gone up, or any other one metric for that matter, i want to see better everything... battery life, graphics, and cpu, otherwise i just wait till the next generation when they've all been addressed.
sinigami - Wednesday, November 17, 2010 - link
uh, am i missing some charts on page 4 or something? because i only see three, and the zacate wins only the first one, ties on the second one, and loses on the third. With only one win, the headline should NOT say it's faster!you could say that the Clarkdale IGP wins or ties the zacate on two out of three gaming benchmarks, and declare it faster.
sheesh, is that some biasedness, or just wishful thinking?
StormyParis - Thursday, November 18, 2010 - link
you're testing games on this eminently non-gaming platform, but not video ?sinigami - Friday, November 19, 2010 - link
not just is it weird that they tested games, it's even weirder that they tested three games, and the headline said this platform was faster than intel's....WHEN IT ONLY WON ONE TEST!
did anyone else even read to page 4?
silverblue - Friday, November 19, 2010 - link
One thing to point out is that the i5-661 has a 900MHz GPU, the i3-530 a 733MHz GPU and the 890GX a 700MHz GPU. This particular flavour of Brazos is a mere 500MHz. Also, Anand is right in that, if most games are GPU bound, Brazos is easily superior to most integrated solutions and definitely so at the same GPU clock speed.