AMD’s Brazos vs. Atom Thermals, Revisited
by Jarred Walton on January 14, 2011 5:35 PM ESTAMD’s Brazos vs. Atom Thermals, Revisited
Last week, we met with AMD at their CES location to see some of their upcoming systems and laptops. While they’ve also recently released several new desktop GPUs, there wasn’t anything new to discuss in that area. The same applies to their desktop CPUs—we’re all waiting to see Llano and Bulldozer. So the focus at CES was understandably on Brazos, aka the “first APU” Vision C- and E-series processors.
We’ve been critical of some of the staged platform comparisons we’ve seen in the past—as Anand put it, the onus is on AMD in this case to provide a truly representative comparison between their new product and Intel’s competing offerings. After the demonstration of their Brazos netbooks on Thursday, AMD called us back and said they wanted to let us rerun the tests to make sure we accurately represented the two platforms. See, there was a slight snafu in the initial thermal imaging comparison. Specifically, AMD thought they put out a netbook with a C-50, but the test system was actually a C-30. So, we returned….
The reason for the mix-up was simple: they had both a C-30 and C-50 system from the same OEM, and they’re basically identical (one was dark blue and the other was light blue). Given that the two C-series parts are both 9W TDP, we didn’t expect much to change, and the new testing confirmed this. We did get some better images of both the top and bottom of the three test netbooks—Atom N550 vs. C-30 and C-50. Unfortunately, stupidity on my part resulted in the loss of said images (it’s a long story…), so all we have are the thermal shots from the keyboard area and screenshots showing CPU utilization during playback along with screen captures taken with FRAPS.
The above gallery shows essentially the same thing as our initial testing: Brazos using its GPU uses less power and runs cooler than Atom N550 doing the decoding in software. The difference between the C-30 and C-50 is pretty much non-existent, as expected. The testing environment was not conducive to doing any form of noise comparison, so while the N550 setup was clearly warmer we couldn’t say if it was quieter or not. Battery life is looking to roughly equal Atom, so that’s good to see. Now we’re waiting for final hardware to see if we can shed any more light on the situation, as well as running our full suite of tests.
We also took the opportunity to capture a video showing the 1080p playback comparison, as that’s part of the story. The video in question is Big Buck Bunny, an open movie demo created as part of the Peach movie project. (You can read more about it on their site, though it’s old enough now that if you haven’t heard of it already there’s not much to add. Suffice it to say, the lack of any licensing issues meant BBB was all over the CES floor, and I’m tired of the short now!) This particular version is a stereoscopic rendering, so instead of the normal 24FPS the frame rate is 48FPS according to FRAPS.
I believe during playback Arcsoft TotalMedia Theater 5 is skipping half the frames, as none of the netbooks come equipped with a 3D 120Hz panel. Does that actually matter? Not that we could tell—now that we’re home from CES, I ran the regular 24FPS version of Big Buck Bunny on a different dual-core N550 netbook, and frame rates still frequently dropped into the teens. Actually, it was worse than the netbook at AMD’s demonstration, but that’s probably more to do with lack of optimizations and some bloatware that came preinstalled; but I digress….
You can see during playback that the Atom N550 periodically stutters and drops below 48FPS—and more importantly, it’s far below 24FPS as well at times. In comparison, both the Vision C-30 and C-50 Brazos/Ontario chips manage a consistent 48FPS. The C-30 does flicker between 47 and 48FPS, but again, that may simply be an artifact of using a stereoscopic 3D video on a non-3D panel. Temperatures are in line with what we reported in our earlier coverage, and the two AMD netbooks are virtually identical. CPU utilization on the dual-core C-50 is lower by about half, as expected.
Once More, With Feeling
This is essentially the killer app of Brazos compared to Atom, and it’s important to keep things in perspective. These chips have a much better IGP than Atom, but at least on the nettop side of things the faster AMD E-350 isn’t miles ahead of Atom D510 in the CPU department. When we drop clock speeds down to 1.0GHz (dual-core C-50) from 1.6GHz (E-350) and compare that to the Atom N550 (1.5GHz)… well, 62.5% of the performance of E-350 compared to 90.4% of the performance of D510 means that in some tests the N550 will probably beat the C-50 for raw CPU potential. Yeah, that’s a concern for me. The GPU is the real difference, so naturally a video decoding test is the best-case scenario. I suspect C-50 will be underpowered for most 3D games, even if the DX11 GPU inside Brazos is fast enough—it will just be the AMD equivalent of Atom + NVIDIA ION, only without as many discrete chips.
We also have to consider performance of the next tier of CPUs and IGPs. Atom is the lowest of the low hanging fruit; we have much faster chips and IGPs from both AMD and Intel, and we don’t need to move up to current generation parts like 2nd Gen Core processors. Even the old Core 2 Duo CULV chips are a darn sight faster than Atom (2x-3x faster), and bad as GMA 4500MHD is, it could do an okay job at H.264 offload. It appears that the E-350 will end up delivering performance roughly equal to the old CULV chips (probably a bit slower, to be honest). That means it will also be around the same level as the Athlon II Neo K325, only with a better IGP and apparently improved power characteristics.
The biggest point in favor of Brazos isn't performance, though. It's going to be cost. If AMD can get partners to put out $400 netbooks (hopefully without Win7 Starter and with more than 1GB RAM), that will hopefully put the nail in the current iteration of Atom. We've seen the Brazos chips, and they're extremely small—smaller even than Atom—so pricing should be very compelling. AMD also doesn't appear concerned about protecting their more expensive mobile offerings (mostly because there aren't many), so they don't have to castrate Brazos in the same way Atom has been stagnant since the first N270 rolled out. Well equipped Brazos netbooks (and nettops) in the $500 range should also be a more elegant choice than Atom + ION/NG-ION, so again AMD looks set to win several matchups.
We’re working to get Brazos hardware in for testing as soon as possible, but it looks like the biggest beneficiaries will be users that want good H.264 decoding in a 10.1” form factor, or an alternative to ION. If you’re looking for the ultimate HTPC chip, we’ll have to investigate that area in further detail, as bitstreaming support and other features are still a question mark. Right now, Brazos is shaping up to be what we all wanted from Atom last year; whether that will be enough in 2011 remains to be seen.
151 Comments
View All Comments
bjacobson - Saturday, January 15, 2011 - link
actually that's not entirely true, transparency still doesn't work without crashing eventually in Windows 7-- but it's much faster without transparency so I leave it turned off.Hacp - Friday, January 14, 2011 - link
Perhaps you miss the point of bobcat? It is to offer cheap netbooks with good battery life and good enough performance. Will it be able to do everything? No. But is it cheap? YES! Will it have good battery life? Yes!jharper12 - Saturday, January 15, 2011 - link
Thanks for the good article. Have you considered the trend towards GPU use in browsers while pondering the future of this platform? IE already does this, but as a user of Chrome I know it's currently a lab feature. Google plans to unload more and more to the GPU, whenever possible. That bodes exceedingly well for browser experience with a Fusion chip.Dan Fay - Saturday, January 15, 2011 - link
Jared,In future benchmarks of Brazos platform-based machines, could you provide some more single-threaded benchmarks? Versus Atom, that's where the Bobcat processor gets interesting, as its OOOE engine should allow it to get higher single-threaded performance than Atom. For heavily multithreaded applications, it's not at all surprising that Atom's and Bobcat's performance are so close together, as Atom's SMT allows it to better keep its in-order pipeline filled.
Single-threaded performance is important in these comparisons because a major criticism of Atom is that it doesn't "feel" very fast, mainly because it doesn't respond quickly enough to OS/UI events, which are usually single-threaded in nature.
Finally, what might help capture Facebook performance is to run something like SunSpider. SunSpider is pretty easy to run anyway, so why not?
JarredWalton - Friday, January 14, 2011 - link
I'm not sure what you're getting at... obviously, doing high-end CPU-intensive tasks on Atom (or Brazos or CULV) isn't really the intended use. It's why I include PCMark, and I'm open for suggestions on other tests that truly show the difference. I've used enough laptops and desktops to say that for basic tasks, even Atom still feels very sluggish, but it might be tolerable in a pinch. Brazos looks to be about the same, only now you can get HD video decoding as a benefit. But calling the lack of Blu-ray 3D support "catastrophic"... well, I'll just respectfully disagree.Despite all the hype and the effort, everyone I hear thinks the 3D experience is marginal at best. A few people really like it, most are ho-hum, and there are a LOT of people that really dislike 3D. Active 3D gives a lot of people headaches, for example. Passive is better in that regard, but it still requires more effort on the part of your eye muscles. Normally I'd say exercising your muscles is good, but I'm not sure that's the case with the eyes. Either way, it's still uncomfortable for enough people that it's a concern.
Ultimately, for netbooks and ultraportables, I really don't care too much about lack of 3D support. It's something probably less than 5% of the market cares about. On HTPCs it's a bigger issue, but I'm not reviewing the HTPC aspect, hence the statement, "If you’re looking for the ultimate HTPC chip, we’ll have to investigate that area in further detail" -- and be "we" I mean "someone else at AnandTech, probably Ganesh. :-)
nitrousoxide - Friday, January 14, 2011 - link
Really looking forward to your test! Have seen many of the tests posted on other sites but they are not convincing enough as most of them are choosing the wrong test suite and wrong counterparts. Hope Anandtech can come up with a test suite designated for netbooks/ultraportables and, never ever compare APUs to the Mobile Cores (1st or 2nd Gen) or any mighty chips on the desktop, I'm just tired of seeing APUs being stated "incredibly slow compared to a X2 250or Pentium G6950" :)omelet - Saturday, January 15, 2011 - link
What do you expect them to compare it to? Obviously Atom will be in the comparison, but what else should they include? Brazos is going to be slow cpu-wise compared to anything except for Atom.knedle - Sunday, January 16, 2011 - link
Problem with "ordinary surfing" is that you can do it on PIII 1GHz or Core i7 and it will still work the same.ninjaquick - Tuesday, January 18, 2011 - link
lol? Was there a punchline I missed, or was it that P3 can run today's content rich web?Edgar_Wibeau - Friday, January 14, 2011 - link
One thing. Atom is strong when two similar threads run on the same core. On a dual core that's 4 threads naturally. Only in this scenario D525 is in the same league as E-350. But please tell me, which are the scenarios when 4 similar threads run under high load on a netbook or TnL notebook? You write that in your Flash scenario only one core is heavily loaded. But when only one heavy thread is running, or even one on each core, then E-350 leads by a big margin. And it likely still runs noticeably faster with three threads.So what ist the everyday-task on a netbook that runs four similar* threads at full load that justifies to say an atom D525 is practically as fast as an E-350?
Oh, and please try Flash 10.2 when you're testing Flash games, see blogs.amd.com/fusion
I don't know anybody playing facebook games in my country btw. Facebook yes, several. But FB games?
*similar means, I've only seen benchmarks for one program at four threads. What happens when four absolutely different tasks (or e.g. two programs with two threads each) are run on a dualcore hyperthreading CPU, I don't know.