Performance

For most of its brief life, Atom was the only core that beat at the heart of Chrome devices. Samsung actually brought us the first Core-based CPU with the Celeron 867 powered Series 5 550, and also the first ARM-based Chromebook with their very own Exynos 5 Dual. We didn’t have a chance to compare the results of the slightly faster clocked Samsung 550, but we do have data for the Atom-based Chromebook. What we expect from this processor is nothing less than IPC dominance. The Intel Celeron 847’s Sandy Bridge cores in the Acer C7 Chromebook are clocked at just 800 MHz, well below the 1.1 GHz possible. The low clock speed helps with battery life, of course, but still places the C7 in a position of dominance over the latest Samsung Chromebook. On the GPU front this is also the most power Chrome OS has seen, with the HD2000 being not our favorite but certainly not our least favorite integrated graphics.

How you harness all that processing and graphics power is . . . a bit limited. You won’t be launching Steam for a pick-up match in TF2 (about all I’d consider playing with the power on hand), instead everything you do will be in the browser. With technologies like WebGL and HTML5 becoming more capable of leveraging local hardware resources we are much closer to experiencing desktop-like web applications. To that end, we have a fair number of tests that show off how well Chrome OS can handle those technologies and the JavaScript tests we’re used to from our mobile tests.

Chromebook Performance Comparison
SunSpider 0.9.1 BrowserMark RIABench Focus Tests Kraken
Atom N570 1.66GHz 1034.3 ms n/a 1968 ms 14229.5 ms
Exynos 5 Dual 1.7GHz 690.5 ms 3056.0 1192 ms

9733.2 ms

Celeron 847 (SNBx2) 800MHz 527.1 ms 3403.7 1194 ms 6817.2 ms

In Sunspider we see a distinct advantage for the Acer, which is no big surprise. The Exynos 5 certainly shows off an admirable advantage over the Atom-based Chromebook, but the C7 trumps that advantage soundly. The Browsermark benchmark has seen its first full update, and as always, that means we’re left with the task of reevaluating all our old hardware with the new suite. In this case we only have a few samples to compare to, so no big burden. The advantage of the C7 is smaller compared to the Sunspider score, but still quite clear. RIABench stands out as the one equalizer between the C7 and the Samsung, indicating that the bottleneck may be within Chrome’s code itself. In Kraken we see another decisive performance win for the Acer.

Chromebook Performance Comparison
IE10 Bubbles Test IE10 Fishbowl IE10 Maze Solver
Atom N570 1.66GHz 11 fps 5 fps 45 seconds
Exynos 5 Dual 1.7GHz 17 fps 8 fps 17 seconds
Celeron 847 (SNB) 800MHz 19 fps 17 fps 17 seconds

The Bubbles and Maze Solver tests do not involve the GPU in any considerable way, making them mostly tests of JavaScript rendering. The Fishbowl test uses HTML5 functions which can be GPU accelerated, resulting in a large performance advantage for the C7. It’s hard to say whether there is any GPU acceleration happening in the Samsung, but if so then the GPU advantage for the Acer is enormous.

Chromebook GPU Performance Comparison
WebGL Solar System WebGL Cubes (500) WebGL Aquarium (50)
Atom N570 1.66GHz 2 fps 10 fps 2 fps
Exynos 5 Dual 1.7GHz 22 fps 28 fps 38 fps
Celeron 847 (SNB) 800MHz 31.7 fps 30 fps 43.3 fps

The WebGL tests hint at that GPU advantage, but not so clearly as we’d like. In the Samsung Chromebook review, Anand discussed how desperately Atom’s GMA-3150 GPU needs to see a huge update to be competitive. Samsung’s use of the Mali-604T in their Chromebook gives it a huge advantage over Atom. The HD2000-based graphics in the Acer C7 show how far mobile SoC GPUs still have to go to compete with PC derived GPUs. We’re limited by an inability to disable Vsync, but if we could I think we would see a much larger advantage than what we see now. We want to explore these differences further soon, but for now all you need to know is that anything that can be processed by the GPU will do much better on the C7 than the Samsung.

Display and User Experience Performance Upgraded
POST A COMMENT

63 Comments

View All Comments

  • JasonInofuentes - Tuesday, January 22, 2013 - link

    In the terminal I was able to run cat /proc/cpuinfo during lots of scenarios and the output was always 800MHz. Since there's no way to run monitoring programs in a Chrome OS instance, and since the clock and voltage tables aren't exposed as they were in the Samsung Chromebook, we don't know whether this is an intermediate state, but based on the results I'm confident of that figure as the max clock. Reply
  • Exophase - Tuesday, January 22, 2013 - link

    Are you sure it said 800MHz for both cores? Can you test it while running a workload you know is pegging the CPU indefinitely, preferably a backgrounded command line task like dd?

    Frankly I think the scores don't really add up for 800MHz. Look at what the lower end Samsung Chromebox achieves here:

    http://www.tomshardware.com/reviews/chromebox-chro...

    That uses a Celeron B840 which is clocked at 1.9GHz, and since it isn't running on a battery they'd have no reason to underclock it. If you scale its Sunspider score of 296ms from 1.9GHz to 1.1GHz you get 511ms, which fits perfectly with the values you got in this review. At 800MHz you would expect a number closer to 700ms. This is assuming linear scaling, but given Sunspider is single threaded and these are fairly low clocks for both I doubt you'd get much worse. I also doubt that V8 made huge improvements in its x86 performance since then.
    Reply
  • sonnyrao - Sunday, January 27, 2013 - link

    Hi, it is definitely a 1.1Ghz processor, but has cpu frequency scaling enabled like any other modern system. I'm not sure what scenarios you ran but looking at /proc/cpuinfo isn't really the best way to determine frequency. At the minmum, you should be looking at /sys/devices/system/cpu/cpu*/cpufreq/scaling_cur_freq
    but running something like powertop or i7z would be even better (you can copy them onto the C7 from another 32-bit Linux system)

    I'd do something like this, put it into dev mode, open a window for crosh (ctrl-alt-t) and run the shell from there sudo to become root
    then run
    watch -d cat /sys/devices/system/cpu*/cpufreq/scaling_cur_freq

    and then run your benchmarks in another window.
    Reply
  • jabber - Tuesday, January 22, 2013 - link

    I've been trying out Lenovo Q180 Nettops recently. The ones I've been rolling out have the top end 2.1Ghz Atom in them, with 4GB of ram and a HD6450 GPU in them.

    Tally this together with a Sandisk Extreme 120GB SSD and the computing experience is pretty good. In all the usual computing tasks I couldn't tell it was an Atom based machine. It felt as fast and fluid as any full size x86 PC I have around. I've rolled a few of them out to small businesses and feedback has been excellent.

    The problem with Atom is that makers feel they have an excuse to pair them up with even crappier hardware. Throw in an Atom chip with some decent supporting hardware and you have a pretty good low power setup.
    Reply
  • lmcd - Saturday, January 26, 2013 - link

    The graphics tied with Atom are probably the most hindering aspect. Reply
  • max347 - Tuesday, January 22, 2013 - link

    Great review!

    I think the pics could use a little less bokeh though :-)
    Reply
  • nathanddrews - Tuesday, January 22, 2013 - link

    LOL - first thing I thought of. DOF is cool, but not if you can't even tell what's on the IO panel. Reply
  • JasonInofuentes - Wednesday, January 23, 2013 - link

    Fair enough, I'm trying to get a nice light set-up that would facilitate more bokeh-free shots, but in the meanwhile I'm generally just trying to keep people from seeing the mess my house is generally in. :) Thanks for the compliment.

    Jason
    Reply
  • Notmyusualid - Wednesday, January 23, 2013 - link

    And it was so dog dam slow, that I GAVE IT AWAY, as soon as I got back from my deployment in the USA.

    Really, it gave me chest pains waiting for it to do anything.

    I tried upgrading the RAM, and that made very little difference.

    I realised quickly that I was wasting my time.

    The 10yr old niece was happy to receive it though....
    Reply
  • dj christian - Thursday, January 24, 2013 - link

    Justin where did Anand cover the Chrome OS? It would be nice if you could provide us with a link to it.

    Thanks!
    Reply

Log in

Don't have an account? Sign up now