Compute & Video Transcoding Performance

AMD's hope with Fusion is that the GPU transistors will regularly be used for more than just 3D gaming. Today we're still hunting for good uses for GPU compute, but we do have a few benchmarks that we can use to illustrate Llano's GPU compute prowess.

Our compute benchmark comes from Civilization V, which uses DirectCompute to decompress textures on the fly. Civ V includes a sub-benchmark that exclusively tests the speed of their texture decompression algorithm by repeatedly decompressing the textures required for one of the game’s leader scenes. Note that this is a DX11 benchmark and obviously can't run on Intel's Sandy Bridge GPU.

Compute: Civilization V

The Radeon HD 6550D performs as expected - it offers performance close to the 5570 and above that of the 6450. Memory bandwidth constraints are less of an issue here apparently as the 6550D delivers 89% of the performance of the 5570.

One area where we can compare the Radeon HD 6550D and Intel's HD Graphics 3000 is in video transcoding performance. For that we turn to ArcSoft's MediaConverter 7 and measure performance transcoding a 15Mbps H.264 encoded 1080p Quantum of Solace rip to a 4Mbps 720p iPhone 4 compatible file:

Video Transcoding Comparison
  AMD A8-3850 Intel Core i3-2105
Frames Transcoded per Second 63.7 fps 165.2 fps

While the A8-3850 can exploit its GPU cores for transcoding, Intel's Quick Sync continues to be the faster method of transcoding video as we found in our original Sandy Bridge review.

Performance in Older Games Power Consumption & Final Words
POST A COMMENT

99 Comments

View All Comments

  • mino - Sunday, July 03, 2011 - link

    Tell ya what. The benefit is we get paid trolls like you over here. Reply
  • jaydee - Thursday, June 30, 2011 - link

    How many monitors can you connect to this with a discrete gpu? Can you do 3 or more DVI/HDMI montors between the motherboard output and a discrete gpu? Reply
  • j_iggga - Thursday, June 30, 2011 - link

    The other dude hit it on the mark. The target for this is OEM parts for budget desktops. So the point about the discrete GPU being more cost effective is moot

    So given that...it's great that finally everyone can game adequately.

    1920 x 1080 on integrated? unheard of in my time
    Reply
  • HangFire - Thursday, June 30, 2011 - link

    OK, so maybe it's better than i3 for laptops. But what I was looking for in Llano was a greatly improved per-clock efficiency in the CPU, something that will drag AMD back into true competitiveness with Intel.

    Instead we get slight tweaking.

    If Bulldozer doesn't deliver a better CPU than oft-tweaked cores dating back to Hammer, AMD is dead on the desktop. Low-end laptops will be the only place they can compete, at least until Intel completes implementing DirectX 10 and 11.
    Reply
  • L. - Thursday, June 30, 2011 - link

    Bulldozer will deliver .. and it will kill Intel at it's price point, I bet your head on that ;)
    And Intel will be dead on the Server market, you can expect a 30+% perf/watt advantage for Interlagos on release day, only dampened when Intel will release their first 22nm Xeon ---

    More blood in the water, better market for us.
    Reply
  • BSMonitor - Thursday, June 30, 2011 - link

    Bet taken. Bulldozer will be an underclocked, overheating monster. AMD is 2 years minimum behind with Bulldozer.

    Ivy Bridge will be out before Bulldozer... Bet on that.
    Reply
  • L. - Thursday, June 30, 2011 - link

    AMD is 2 years in advance with bulldozer, as it is not designed as a desktop processor, but a byproduct of Interlagos, which will very likely take a lot of server market share from Intel.

    I'd like to see how AMD could be two years behind, when Intel has been stretching a core design from core1 to sandy bridge ;)

    Bet on the fact that AMD has always priced their stuff right, Bulldozer is in i7-2600k range, that means it will beat it hands down.

    Ivy will be a win in desktop for Intel, but then again, this depends on how fast both Intel and GF can get to 3d-gate 22nm. (which if we look at current trends would mean Intel 6 months before GF more or less).
    Reply
  • BSMonitor - Thursday, June 30, 2011 - link

    You are completely wrong. Bulldozer was scheduled for release in 2009. It is 2011. Hence 2 years.

    Considering Conroe processors still dominate Phenom II x2, x4, and x6 processors from AMD. I would say AMD is behind. About 3 generations.

    GF is just now shipping its first 32nm chips. Still not a single heavyweight chip at 32nm. Intel has 32nm 6-core processors for over a year. Ivy Bridge is out this fall at 22nm.

    Are you just completely drinking the AMD Kool-Aide or what?
    Reply
  • silverblue - Thursday, June 30, 2011 - link

    If I remember correctly, Bulldozer's design was torn up and started again from scratch in 2008. This would undoubtedly increase development time, especially if they completely changed the design.

    And for the final time, stop spreading BS about Conroe dominating Phenom II. Even Penryn doesn't. There are plenty of instances where K10.5 beats the Core 2 family and in most cases where Core 2 wins, the difference is marginal at best, not to mention that a) there aren't any hexacore Core 2s out there nor any with any turbo technology, and b) any sufficiently high performing Core 2 parts are massively more expensive than anything AMD is shipping.

    I dedicated a HUGE post on this topic to you on Toms using Anandtech Bench data and you've obviously decided to ignore it so... believe what you like. :)
    Reply
  • BSMonitor - Thursday, June 30, 2011 - link

    Yeah, I exaggerated. But the only way to deal with the diluted is to use such devices. Perhaps they will think and research.

    Phenom II vs Core 2 is not even close clock for clock.

    Phenom II only wins in scenarios where it is grossly clocked higher than the similar Core 2. The 9650 is a 3GHz part, Ph II 980 a 3.7GHz part.

    http://www.anandtech.com/bench/Product/49?vs=362

    The recently released 980 barely outperforms the 3 y/o Penryn Core 2.
    Reply

Log in

Don't have an account? Sign up now