The Ultimate Visual Experience. ATI Avivo HD: Addressing the Challenges of HD Playback

Transcription

1 The Ultimate Visual Experience ATI Avivo HD: Addressing the Challenges of HD Playback

2 Table of Contents Introduction... 2 The Challenge of HD Playback... 3 The GPU Advantage... 7 The Current Solution...7 Unified Video Decoder (UVD)...8 ATI Avivo Video Post Processor (AVP) HD content protection challenges...12 HDCP support on ATI Radeon HD HDMI & Audio support on ATI Radeon HD Summary...16

3 Introduction High Definition (HD) content is gaining in popularity, driven by the increasing availability and affordability of HD capable television sets and a desire by consumers for a more immersive entertainment experience. As the uptake of HD viewing increases so too will the demand for delivery mechanisms of HD content. Given that the PC has been the vehicle of choice for delivery of different types of digital content, including Standard Definition (SD), it is expected that PCs will also provide a complete ecosystem for fully featured and efficient HD content processing. HD content will be delivered in much the same way as SD content is today, i.e. broadcast, optical discs and content on demand / downloadable content. All of these delivery mechanisms affect the PC; however, the initial driver will be the increasing availability of high density optical media such as HD DVD and Blu-ray. HD DVD and Blu-ray drives are currently shipping in both desktop and notebook OEM configurations and are available as add-in or external units for upgrade. Even though consumers may be able to upgrade their current PCs with new HD DVD and/or Bluray optical drives it is not a given that the rest of the components will have sufficient processing capability for fully featured and acceptable HD content playback. HD content presents many challenges, including: Large quantities of data processing; up to 6 times the rendering required of SD content Computational complexity of algorithms for decode (i.e., CABAC) and processing (i.e., advanced de-interlacing) Power consumption when executing in software or GPU assisted decoding scenarios Content protection mechanisms Many notebooks available today are capable of playing back a typical SD movie on a single battery charge; however, left unchecked, the increased processing requirements of HD content will have ramifications in power utilization. These increased processing demands have ramifications on desktop PCs as well; not just making them less efficient but also potentially increasing their noise as any fans in the system may need to dissipate more heat. This is of particular concern for PCs that are used in a home theatre and/or personal video recorder environment where consumer electronics (CE) experience is expected. Here we aim to look at the particular challenges associated with decoding HD content on the PC and the specific process implemented in AMD s next generation GPUs in order to provide efficient and high quality playback HD playback on the PC. 2

4 The Challenge of HD Playback SD sources, such as DVDs, typically deliver content at resolutions of up to 720x480 (345,600 pixels) for NTSC discs or 720x576 (414,720 pixels) for PAL. HDTVs are generally required to support 720p (1280x720 progressive scan) and 1080i (1920x1080 interlaced). However movie titles released on HD DVD or Blu-ray have tended to be encoded in 1080p (1920x1080 progressive scan) resulting in a source resolution of 2,073,600 pixels. Thus, the number of pixels that must be decoded and rendered for HD content is five to six times greater than that for the SD equivalent, as shown in the following table: Frame Resolution Data Rate (pixels) Scan Mode (pixels / sec) Width Height fps* Interlaced 10,368,000 fps Interlaced 10,368,000 Ratio to SD fps Progressive 55,296,000 ~ x5 fps Progressive 49,766,400 ~ x5 fps Interlaced 62,208,000 ~ x6 * fps: Frames Per Second HD content playback also means an increase in the bit-rate (the quantity of data encoded per second). Although a higher bit-rate may ostensibly indicate higher quality, other factors, such as the compression ratio of particular frames, affects the bit-rate of the encoded video source. The following table provides sample bit rates for popular video sources: Source Maximum Video Bit-rate DVD ~9.5Mbps Ratio to SD HDTV (ATSC) ~18Mbps ~ x2 HD DVD ~30Mbps ~ x3 Blu-ray ~40Mbps ~ x4 Sources : DVD Format LLC, Blu-ray Disc Association The reason HD source resolution data rates (up to six-fold over SD) and bit-rates (up to fourfold over SD) do not track in a linear relationship is due to the use of newer, more efficient encoding methods that trade off of the increased processing power afforded by modern processors. These encoding methods permit higher compression ratios while maintaining the quality provided by previous SD encoding methods such as MPEG-2. Although MPEG-2 may be used to encode high definition sources, both HD DVD and Blu-ray require support for the newer VC-1 and H.264/AVC (Advanced Video Coding, or also known as MPEG-4 Part 10) schemes. Both VC-1 and H.264/AVC employ more complex encoding stages than MPEG-2, further increasing the processor workload when decoding the video for playback. 3

5 The advanced encoding stages used by VC-1 and H.264/AVC are: Entropy Decoding. During the encoding of the video source, the binary representation of the video is compressed and thus must be decompressed for playback. Methods that may be used for the compression process include: VLC (Variable Length Coding): Frequently occurring elements are compressed to shorter codes, while infrequent ones are assigned longer codes. CAVLC (Context Adaptive Variable Length Coding): Here multiple different coding schemes are used and the input elements are coded from any of these and keyed with the type of coding that is used. CABAC (Context Adaptive Binary Arithmetic Coding): A complex compression mechanism that allows input elements to be encoded as non-integer numbers and a greater level of adaptivity. More of the input elements are coded in context adaptive fashion than CAVLC resulting in a higher compression ratio. Frequency Transform. During the encoding process, the video frame is divided into blocks and each of these blocks are compressed with a transformation method. In many cases Discrete Cosine Transform (DCT) is used, whereby each block generates a block of coefficients indicating the amount of horizontal or vertical frequency within the block. During this process, the encoder performs a quantization of the block, dividing each coefficient by a constant and rounding the result to minimize the amount of encoded data, thus increasing the compression ratio but losing image details. During the decoding process an Inverse DCT (idct) formula is applied to reconstruct the original data. Although an exact representation of the original data is theoretically achievable with idct, it is not realistically feasible due to the rounding introduced by quantization during the encoding process. To reduce quantization errors, postprocessing can be performed during the decoding process, serving to minimize the visual artifacts introduced by over-quantization. Pixel Prediction (Motion Compensation). Not all frames are retained when encoding a source video into a compressed video file format. In fact, in order to achieve better compression ratios, many compressed frames are largely on information from other frames. First, the encoder compresses a full frame (reference) and estimates the motion that has occurred between blocks of the reference and the next frame (intermediate). It then encodes the parameters of the changes, called motion vectors. Depending on video formats, a new reference frame is compressed after a defined number of intermediate frames. At decode time, the intermediate frame is reconstructed and its pixels blocks predicted by combining the reference frame and motion vectors. This is known as interframe motion compensation. Intraframe motion compensation can also be used in order to better improve the process by using more adjacent information within a frame. Depending on the decoding mechanism being used, the motion vectors from various frames can be used with a bilinear (or an even more complicated bicubic) calculation to determine the final color value of the resultant pixel. Deblocking. Each encoding stage is performed on blocks of pixels smaller than the entire frame. Deblocking is a post-decompression routine that consists of a filtering 4

6 between blocks to minimize any visual anomalies introduced by the motion estimation and the frequency transform. Deblocking is part of the VC-1 and H.264 decoding processes, but is not an MPEG-2 decoding requirement. Entropy Frequency Pixel Decode Transform Prediction Deblocking MPEG-2 VLC idct (Floating Point) Inter Frame (Bilinear) N/A VC-1 VLC idct (Integer) Inter Frame (+Bicubic) Inloop H.264/AVC CAVLC/CABAC Inverse Transform Inter & Intra Frame Inloop Sources: Moving Picture Experts Group (MPEG), Microsoft Both VC-1 and H.264/AVC require more processing during the decoding process than does MPEG-2 because of the deblocking step. In addition, entropy algorithms like CAVLC and CABAC cause dramatic increases in the complexity of the decoding process. In fact, using CABAC with H.264/AVC causes particular problems because it is, in effect, a bitstream process. That makes it impossible to parallelize, which means it cannot be distributed across the wide processing pipeline of a GPU and can only use one core of multi-core CPU (unless the author specifically encodes it with time-splicing, which is an unfriendly breach of a CPU s branch prediction processes). MPEG-2 AVC CPU utilization SD HD CAVLC CABAC CABAC 8 Mbps 20 Mbps 20 Mbps 20 Mbps 40 Mbps Entropy Decode 1.6% 4.2% 8.9% 26.5% 47.8% Freq. Transform 0.7% 2.0% 1.2% 1.5% 2.6% Pixel Prediction 0.5% 4.1% 11.0% 11.5% 11.6% Deblocking 0.0% 0.0% 8.2% 8.5% 8.2% 2.8% 10.3% 29.3% 48% 70.2% System: Pentium 4 2.8GHz CPU 1GB RAM The above processes highlight the CPU utilization of each stage. Many of the processes are relatively trivial for a modern CPU when decoding an SD MPEG-2 video, and the processor utilization increases when moving to an HD resolution MPEG-2 video but the total is still a little under 10%. In comparison, a ~20Mbps H.264/AVC video using CAVLC entropy decoding requires many multiples of CPU utilization under the Entropy Decode and Pixel Prediction stages, and the additional Deblocking stage places another significant burden on the processor. Moving to higher bit-rates doesn t significantly impact the CPU utilization for H.264/AVC in most cases, however CABAC entropy decoding really shows up as a processor intensive task with nearly 50% of the available processor cycles dedicated to this stage of the decoding process alone for a ~40Mbps encoded video the maximum supported by Blu-ray disks. 5

7 Achieving smooth playback of HD videos is not only processor intensive; it is also memory bandwidth intensive. There are many processes in the decode stage, and the decoding of a single frame requires multiple references from other frames. Thus multiple reads to and from memory are required necessitating relatively high performance RAM. Due to the high processing demands necessitated by the encoding schemes of Blu-ray and HD DVD, there are ramifications in terms of cost, power and heat. For instance a PC that is tasked with multiple operations, especially those in a Home Theater (HTPC) / Personal Video Recorder (PVR) capacity may be encoding and storing broadcast programming while playing back stored or optical media. High performance CPU s will be required to ensure that all operations are able to perform acceptably, which has ramification on the overall cost of the system. High processing requirements on a general purpose CPU will also translate into higher power draw requirements. Consumers in general are becoming more aware of the power consumption of devices and will be less keen to adopt solutions that do not have favorable power consumption in relation to dedicated CE devices. Power draw, though, does have more immediate concerns for notebook devices where, with DVD playback, it is reasonable to expect to be able to play a movie on a single battery charge, which might not be the case with movie playback via Blu-ray or HD DVD. Alternatively, higher power draw in desktop or HTPC scenarios can also translate into noise due to the increased levels of heat generated that has be moved via processor and/or case fans; increased audibility of the playback system itself may result in a negative experience to the user. 6

8 The GPU Advantage The Current Solution As the PC evolved so too did the development of graphics processors and the level of support for video processing. Not only do graphics processors implement specific processing functionality for video decoding, but the high performance, parallel processing pipelines of modern shader enabled GPUs can also be utilized to provide high quality post processing in order to rival even expensive dedicated Audio/Video hardware with SD sources. However, with HD sources more data needs to be processed, and the general decoding has become more complicated with more advanced video encoders, requiring innovative approaches for GPUs to address HD playback challenges. AMD s ATI Radeon X1000 series of graphics processors have been able to cover much of the decoding process by implementing specific hardware to assist in the decoding as well as utilizing the wide graphics processing pipeline. Figure 1: ATI Radeon X1000 hardware accelerated video playback This process has the advantages of moving the majority of the video decoding off of the CPU, thus potentially enabling full frame rate and/or lower power HD playback on a wider range of CPUs. However, this does still leave the entropy decoding stage on the CPU. The utilization of the graphics pipeline for elements of the decoding means that there is a reliance on the graphics capabilities of the ASIC, with current entry level not being sufficient for full HD playback. Another consideration is that the more graphics processing that is required for the decoding of video, the less there is available for post processing routines that are designed to enhance the quality of the resultant decoded frames. While this is largely solved for SD content, HD content still demands the same frame rates but with up to 6x more 7

9 pixels to process and the more intensive decoding routines, thus available processing becomes more limited. Unified Video Decoder (UVD) In order to enable full HD video decoding on a complete range of PC graphics solutions AMD is introducing the Unified Video Decoder, or UVD. UVD is a dedicated video decoding block, initially developed by AMD s Digital TV group, that facilitates the full, bit accurate, decoding process of VC-1 or H.264/AVC encoded video. Figure 2: AMD s Unified Video Decoder By utilizing dedicated hardware for the decoding of VC-1 and H.264 encoded video, the CPU is completely alleviated from the decoding process, including the processor intensive entropy decoding stage. An additional benefit is freeing up the graphics pipeline. The net result of removing the entire decoding process off of the CPU to the graphics processor is that it frees up significant CPU cycles, as illustrated below: 8

10 % CPU Ultilization (Lower is better) CPU Decode GPU Assisted Decode UVD Decode 10 0 Yozakura (H.264 HD DVD) System : Intel Pentium D (Dual Core) 3.2GHz 1GB RAM Figure 3: Example of CPU utilization during H.264 playback The documentary Yozakura 1 is one of the highest bit-rate HD optical HD titles produced thus far. It is encoded as a 1080 interlaced video, and we can see that a mid-level dual core CPU is highly occupied with the decoding process for a video of this kind. CPU assisted decoding certainly alleviates some CPU cycles, but for a video of this bit rate encoded with H.264 the CPU is still spending the majority of its processing cycles decoding the video. UVD decoding sends the CPU utilization to much lower levels, in this case there is only a 12% utilization equating to a removal of 70% of the CPU cycles that would have been spent decoding the video under the CPU decode case. In fact, the remaining 12% CPU utilization is the OS and application overhead required when playing back an HD optical source. With UVD, VC-1 decoding will benefit from the same low CPU utilization. One result of switching processing from a CPU based software decoding to a dedicated processing solution on the GPU is that of lower overall system power utilization. An area where this can be of particular benefit is in a notebook PC with HD playback processing placing a greater burden on the battery. 1 Yozakura 2006 Fuji Television Network Inc., Japan 9

11 60 % Average System Power Draw in Watts Yozakura (H.264 HD DVD) CPU Decode UVD Decode System : Intel Core 2 Duo 2.0GHz 1GB RAM Figure 4: Example of power consumption during H.264 playback The above test shows the power draw of a notebook platform playing the Yozakura clip. In this instance we can see an average total system power draw reduction of 15.8W, equating to a 30% reduction in system power draw. This 30% power reduction obviously equates to significantly improved battery life while the notebook is playing HD optical content. Reducing overall system power draw under HD optical content playback is obviously a benefit for desktop and HTPC s as well, but there are secondary benefits as well. With the processors (CPU and GPU) spending fewer execution resources on the decoding process, which is being done more efficiently in dedicated hardware, there is less overall heat produced and that can result in a quieter system as fans are not required to spin as fast and move as much air to dissipate the heat from the primary system components. Under a GPU assisted decoding scenario, where the rendering pipeline is partially relied on for some decode stages, there is the problem that current entry level graphics processors just do not have sufficient rendering capabilities to be able to facilitate the decoding of HD VC-1 or H.264/AVC media. By implementing a specific decode block this not only alleviates the CPU from any of the decoding process, but also the GPU rendering pipeline, thus enabling full HD decoding with just entry level graphics processors. This can reduce entry costs for a system to support full HD playback by not needing high performance CPUs and GPUs as would be required under the software or GPU assisted decoding situations. ATI Avivo Video Post Processor (AVP) The resultant frames of video following the decoding process can go through further post processes in order to improve the quality of the image. While SD video poses the same challenges in terms of post processing, the computational resources required for HD video 10

12 places a much greater burden on a the graphics pipeline, such that current mainstream GPUs do not have sufficient performance to enable them. On higher performance GPUs, functionality may be enabled, but this can come at the cost of higher power utilization, which may not be desirable on certain configurations, such as those targeted for notebooks. To this end, next generation GPUs from AMD feature the ATI Avivo Video Post Processor (AVP). As with UVD, AVP is a specific block on the GPU s ASIC designed with enough processing capability for HD video resolutions and bit-rates. The functionality of AVP includes: Deinterlacing (with edge enhancement) Horizontal and vertical up and downscaling Programmable color space conversion and gamma correction AVP s pipeline is a memory-to-memory process, which means that the functionality of AVP can be used exclusively, or it can be a part of the overall post-processing routine by offloading some of the processing that was done on the GPU rendering pipeline thus freeing up GPU cycles. The advantage here is that for entry level GPUs it is enabling high quality HD postprocessing. As AVP also removes these post processing functions from the rendering pipeline this too can have some additional power benefits, thus enabling the capability for high quality video playback while still leaving a longer battery lifetime in a notebook scenario. Although high definition media playback represents an issue to the PC in terms of the processing it requires, a UVD solution largely solves these issues. In conjunction with UVD, AVP can provide complementary benefits, while also raising the HD playback image quality bar across the range of graphics processors. In summary, UVD and AVP can provide: Lower system power consumption in relation to systems that utilize CPU for GPU assisted decoding, benefiting: Notebook battery life Heat output System noise Enables HD video decoding capabilities across the range of graphics cards, even at the entry level graphics solutions, and reduces the bar for CPU processing power required to facilitate HD playback Provides high quality HD video playback across the range of graphics processors. 11

13 HD content protection challenges The advent of new HD optical discs, such as Blu-ray and HD DVD, as well as digital cable on PCs, mandates certain requirements to playback protected content at HD resolutions. These new technologies introduce the need to comply with content protection mechanisms established by content providers, like movies studios or digital cable operators. Hardware manufacturers and software suppliers need to implement these content protection mechanisms to allow the abovementioned content to be played back at HD resolutions. Content providers have mandated that their content, when stored on Blu-ray and HD-DVD discs, can only be displayed at HD resolutions if HDCP is supported on both the PC and the monitor, i.e. when the interface is protected by HDCP. The same applies to digital cable on PC: to display content received through digital cable on a monitor, and then the interface is required to be protected by HDCP, for the content to be shown. It is noteworthy that not every Blu-ray and HD DVD disc requires HDCP. It is only when the content stored on it has information indicating that it requires protection, such as certain movies, that HDCP comes into play. There are different digital interfaces supporting HDCP: Digital Visual Interface (DVI), High Definition Multimedia Interface (HDMI) and DisplayPort. HDCP compliant source HDCP protected link HDCP compliant Display HD Video Player / PC HD Display Figure 5: Elements in an HD protected content playback Figure 6: DVI and HDMI connectors 12

14 HDCP support on ATI Radeon HD 2000 In order to provide a full HD experience, ATI Radeon HD 2000 series have been designed with a comprehensive approach to provide not only exceptional quality playback for HD video, but also seamless HDCP support to be ready for playback of protected HD content from Blu-ray and HD DVD. As part of the encryption mechanism of HDCP, there are unique keys that are stored on each graphics card allowing it to decrypt and decode protected content. Some implementations of graphics cards store HDCP keys on an external CryptoROM; this adds additional costs and complexities to the solution, as it requires board manufacturers to manage different configurations. ATI Radeon HD 2000 series follow a different approach: The HDCP keys are stored directly onchip, ensuring that all ATI Radeon HD 2000 ASICs are HDCP ready 2, thus enabling end-users to playback protected content. This implementation helps in lowering the cost of the total solution for HDCP compliant boards, and provides an easier and more reliable offering. Non ATI Radeon HD Graphics ASIC Transmitter HDCP Protected Link ATI Radeon HD ASIC Transmitter HDCP Keys HDCP Protected Link HDCP Keys CryptoRom HD Display HD Display HDCP keys stored on CryptoROM HDCP keys stored on ATI Radeon HD ASIC Figure 7: HDCP support on graphics solutions Many graphics cards today do not support HDCP on dual-link DVI monitors; they only support it on single-link DVI interfaces or monitors. In such a scenario, when an end user attempts to playback protected content on dual-link DVI monitors, the video playback software will report that HDCP is not supported, forcing the user to drop the desktop resolution to a single-link DVI rate; in some cases it could be down to only a quarter of the native resolution of the monitor. If the monitor has a scalar, it would scale it back to its native resolution; however, image artifacts may be introduced by the scaling function. 2 Playing HDCP content requires additional HDCP ready components, including but not limited to an HDCP ready monitor, disc drive, multimedia application and computer operating system. While ATI Radeon HD 2000 series ASICs have with HDCP keys integrated, AMD s add-in board manufacturers may choose not to enable HDCP on their specific ATI Radeon HD 2000 based products. 13

15 HDCP support over Single-link DVI ATI Radeon HD s HDCP over Dual-link DVI Figure 8: HDCP over DVI ATI Radeon HD 2000 series can support HDCP on all of is DVI interfaces, be it single-link DVI or dual-link DVI. For example, if an ATI Radeon HD 2000 graphics card offers two dual-link DVI interfaces, then both are capable of supporting HDCP. This enables end-users to display HDCPprotected content on dual-link DVI monitors at their native resolution; this offers the best image quality possible, because it eliminates the need for any scaling to be done on the monitor, which could potentially degrade quality. HDMI & Audio support on ATI Radeon HD 2000 To be compliant with Windows Vista Premium Logo requirements, a system with enabled HDMI output must include two distinct audio sources one main system source and one for HDMI output. In this scenario the system builder has to purchase a second PCI sound device; that adds cost and complexity to the system. In a configuration with only one audio source used for HDMI, the user will not be able to use digital audio output for the system sounds. In both cases, the audio is routed to the graphics card via a physical cable (S/PDIF) in order to be included with the video on the HDMI output (Figure 9). ATI Radeon HD 2000 series provide a comprehensive approach to provide audio support for HDMI by including an HD-audio controller on the ASIC, creating a protected audio path separate from the system s audio, compliant with Windows Vista Premium requirement and without any additional connection or cable. Protected audio path HDMI Digital out Analog out Line(s) in HD-Audio link 14

16 Alternative HDMI and audio support ATI Radeon HD s HDMI and audio support Figure 9: HDMI and audio support Combined with the ATI Radeon DVI-to-HDMI adapter, ATI Radeon HD 2000 solutions enable end users to get full video and audio support via standard DVI for an easy connectivity and the best HD experience. Current DVI-to-HDMI connectors do not carry audio, thus do not enable HDMI. ATI Radeon adapter handshakes with the board to enable audio via the DVI port; when the adapter is connected to the output of an ATI Radeon HD 2000 solution, it becomes a seamless HDMI output. Figure 10: ATI Radeon DVI-to-HDMI adapter ATI Radeon HD 2000 series innovate in the support of HDMI, bringing several benefits to the end users: Plug-n-play HDMI solution, with no internal or external cabling required Full audio experience, preserving the system sounds on S/PDIF output Flexibility through seamless support of different interfaces using ATI Radeon DVI-to- HDMI adapter when needed 15

17 Summary The emergence of HD has created a lot of expectations from consumers looking for a unique video experience, including ones using PCs as their entertainment centers. But as a new technology, HD also brought new challenges related to the large quantity of data to be processed in HD, computational complexity, power consumption and content protection. The GPUs can play a critical part in providing consumers with high quality and seamless HD playback. Unfortunately, most of the current graphics solutions do not provide the performance for full quality HD playback, and in some cases, lack of proper support for HDCP, audio and HDMI output prevented end users from fully enjoying the HD experience. By offering unique video playback technologies which are UVD and AVP technologies, providing a comprehensive approach to HDCP requirements and a seamless support for audio and HDMI interface, the ATI Radeon HD 2000 series truly address HD playback challenges. 16

18 Disclaimer The information presented in this document is for informational purposes only and may contain technical inaccuracies, omissions and typographical errors. AMD MAKES NO REPRESENTATIONS OR WARRANTIES WITH RESPECT TO THE CONTENTS HEREOF AND ASSUMES NO RESPONSIBILITY FOR ANY INACCURACIES, ERRORS OR OMISSIONS THAT MAY APPEAR IN THIS INFORMATION. AMD SPECIFICALLY DISCLAIMS ANY IMPLIED WARRANTIES OF MERCHANTABILITY OR FITNESS FOR ANY PARTICULAR PURPOSE. IN NO EVENT WILL AMD BE LIABLE TO ANY PERSON FOR ANY DIRECT, INDIRECT, SPECIAL OR OTHER CONSEQUENTIAL DAMAGES ARISING FROM THE USE OF ANY INFORMATION CONTAINED HEREIN, EVEN IF AMD IS EXPRESSLY ADVISED OF THE POSSIBILITY OF SUCH DAMAGES Advanced Micro Devices, Inc. All rights reserved. AMD, the AMD Arrow logo, ATI, the ATI logo, Avivo, Catalyst, Radeon, and combinations thereof are trademarks of Advanced Micro Devices, Inc. Vista is a trademark, and Microsoft and Windows are registered trademarks, of Microsoft Corporation in the United States and/or other jurisdictions. Other names are for informational purposes only and may be trademarks of their respective owners.