The Test

As we previously indicated, we need to use at least a Core 2 Duo E6400 in order to avoid dropping frames while testing graphics card decode acceleration under X-Men: The Last Stand. As we also wanted an accurate picture of how much GPU decode acceleration really helps, we needed to use a CPU powerful enough to avoid dropping frames even under the most stressful load without GPU assistance. Thus we chose the Core 2 Duo X6800 for our tests. Using this processor, we can more accurately see how each graphics card compares to the others and how much each graphics card is able to assist the CPU.

We tested CPU utilization by using perfmon to record data while we viewed a section of X-Men: The Last Stand. The bookmark feature really helped out, allowing us to easily jump to the specific scene we wanted to test in Chapter 18. In this scene, the Golden Gate is being torn apart and people are running everywhere. This is one of the most stressful scenes in the movie, reaching a bitrate of over 41 Mbps at one point.

Unfortunately, we haven't found a feature in PowerDVD or another utility that will allow us to count dropped frames. This means we can't really compare what happens to the video quality when the CPU is running at 100%. In lieu of dropped frames, we will need to stick with CPU overhead as our performance metric.

For reference we recorded average and maximum CPU overhead while playing back our benchmark clip with no GPU acceleration enabled.

Here is the rest of our test system:

Performance Test Configuration
CPU: Intel Core 2 Duo X6800
Motherboard(s): ASUS P5B Deluxe
Chipset(s): Intel P965
Chipset Drivers: Intel (Intel)
Hard Disk: Seagate 7200.7 160GB SATA
Memory: Corsair XMS2 DDR2-800 4-4-4-12 (1GB x 2)
Video Cards: Various
Video Drivers: ATI Catalyst 6.11
NVIDIA ForceWare 93.71
NVIDIA ForceWare 97.02
Desktop Resolution: 1920x1080 - 32-bit @ 60Hz
OS: Windows XP Professional SP2

H.264 Encoded HD Content: A Good Thing X-Men: The Last Stand CPU Overhead
Comments Locked


View All Comments

  • Tujan - Monday, December 11, 2006 - link

    So heres a Sony notebook. It probably uses less than 40 or 50 watts. Has an HDMI connector on it. And runs on a battery. No less.">

    So what is my question here. This is a Centrino Core Duo for a notebook. With graphics enough to run using only battery power .

    As well the notebook has a Blue-Ray drive wich can be written to.AND watch blue-ray titles.

    Is this mostly in the liscencing ? How can it be when the processor used,and graphics cards used are such absolute 'top notch'for the desktop. And the notebook puts the works of them to shame.

    Blue-ray,and HDMI on battery power.

    This was one of AnandTechs Adds.Incodently - Hi Anandtech(Add-Click),HI Sony.
  • cmdrdredd - Monday, December 11, 2006 - link

    I too wonder how a laptop can play blue-ray fine but a $400+ video card with a CPU probably 2x+ more powerful and more memory...can't.
  • fanbanlo - Monday, December 11, 2006 - link

    most efficient software decoder! Maybe we don't need Core 2 Duo after all!">
  • DerekWilson - Monday, December 11, 2006 - link

    my understanding is that coreavc doesn't work in conjunction with HDDVD/BD -- that it doesn't support AACS.
  • totalcommand - Monday, December 11, 2006 - link

    BluRay support will be added to CoreAVC soon.
  • KashGarinn - Tuesday, December 12, 2006 - link

    When CoreAVC will support HD-DVD and bluray H.264, I'd be very interested in seeing this article updated with the comparison.

    Regarding the article itself, I thought it wasn't up to normal anandtech standards.. skimping on the H.264 details which makes it better and giving the reason as "but these are a little beyond the scope of this article." - What is anandtech coming to? That's like saying "we're going to compare graphic cards with directx9 capabilities, but explaining what directx is, is a little beyond the scope of this article"

    Also, not comparing amd cpus? What's up with that?

    And I find it odd that you didn't comment on the strangeness that nvidia has better acceleration across the board than the ATI cards, especially as the ATI cards have better shader throughput, so probably most likely hampered by software rather than hardware.. so this: "ATI hardware is very consistent, but just doesn't improve performance as much as NVIDIA hardware." - only paints the incorrect picture.

    I would give this article a 2 out of 5.. 1 for at least covering the basics (h.264 is a better codec than mpeg2) and 1 for showing that ati needs to improve it's decoder.. even though you don't point it out.

  • ninjit - Monday, December 11, 2006 - link

    I had a question about why you chose the golden-gate bridge scene to stress test the decoding capabilities of the various setups.

    You said that you chose that point in the movie because it had the highest bitrate (41Mbps), indicating a more complex scene.

    To me though that would indicate LESS encoding done by H.264, and subsequently LESS decoding work needed to be done for playback of that particular scene.

    I justify that by thinking with a very complex scene the codec cannot compress the stream as much because it would introduce too many artifacts, so the compression rate is dropped and the data rate increased to compensate for that particular section in time.

    Is my reasoning correct? If not, can someone explain to me why?

    I don't think choice of scene should change the graphs in terms of relative performance between setups, but it would affect absolute numbers - an easy way to check whether my thinking is wrong or not is to see if there are more dropped frames in the Golden Gate scene on the software-decoded E6600 vs. other less busy scenes.
  • DerekWilson - Monday, December 11, 2006 - link

    we tried to explain this a little bit, so I'm sorry if we didn't get it across well enough.

    I'm not an expert on H.264 by any means, but I can talk about other types of decoding as they relate to still images.

    The issue isn't really less compression -- when using H.264, we are always using H.264 complexity to encode the bitstream. We don't fall back to just saving raw pixel data if a scene is overly complex -- we encode more detailed information about the scene.

    For instance, with still images, run length encoding can be performed with huge compression especially in images with large blocks of identical colors (like logos or images on a solid background color). Basically, the idea is to list a color and then a number of pixels that use that color. For an image that is a single solid color, you could list the color and then the number of pixels in the image. This is a very small file with little processing requirement that represents a full image. If, on the other hand, we have a checker board pattern with every other pixel being a different color, we have to list the color of every pixel, BUT we also have to process every color to makes sure of how many consecutive pixels it represents (even if it only represents one). Thus, we end up donig more processing than we would on a smaller (lower "bitrate") file.

    This example is very fabricated as sophisticated run lenth encoding can handle more complex patterns, but it serves to illustrate the point: when using a specific type of encoding, higher bitrates can (and usually do) mean more complexity and processing.

    As we mentioned, using no encoding requires zero processing. MPEG-2 can compress the data to lower the bitrate while increasing computational complexity. But higher bitrate MPEG-2 means more data to process per frame -- which means more CPU overhead for higher bitrates under MPEG-2. The same is true with H.264 -- bitrates are genearlly lower than MPEG-2 and require more processing power, but as H.264 encoded movies use more bitrate (more data per frame), more processing is required.

    I hope this helps.

    Also, to clarify -- the spot at the video that reaches 41Mbps corresponds to the highest CPU utilization (we can see this on a the perfmon timeline).
  • ninjit - Monday, December 11, 2006 - link

    Thanks for the explanation Derek. That was very helpful.
  • jeffbui - Monday, December 11, 2006 - link

    The PS3 is able to play Blu-Ray back at 50% over normal speed without dropping frames. That gives an idea of how much power these consoles are capable of.

    Some interesting tidbits from a translation of an article interviewing PS3 developers.

    -H.264 decoding itself was not very difficult for Cell with moderate optimization and they could play a movie in realtime at the first try unlike very difficult SACD optimization. However, because they began the development without knowing the final Blu-ray standard, they set the goal very high for decoding 2 full HD H.264 streams at 40Mbps simultaneously. Besides the clockspeed of the devkit was lower than the final product which made the development difficult. The current decoder can decode full HD H.264 with 3 SPEs.

    -An SCE developer recommends trying 1.5x fast-forward playback in the PS3 BD player to see the power of Cell. When it's connected to a display via 1080/60p, it becomes very smooth as Cell has an enough margin for video decoding. In 1.5x fast-forward playback it decodes all frames then inserts them into 60fps with sped up audio.

Log in

Don't have an account? Sign up now