HD Video Playback Power Consumption Analysis

Download PDF

Download HD Video Playback Power Consumption Analysis [PDF 237KB]

Background

Platform Power Profile

As a precursor to our research, it is important to understand how energy is being consumed in the mobile computer. The power profile provides a model of various components on a mobile computer (mobile platform). Measurement results vary depending on the usage model. For example, the relative contribution of processor power to the overall platform power will be significant in a CPU-intensive workload, but it will not be a dominant factor while the platform is idling. Furthermore, it may also vary depending on whether hardware acceleration is enabled or disabled, as well as the type of codecs that are used in case of video playback. These cases are studied as scope of the paper.

Figure 1 shows how the power profile can vary during various usage models. For this particular profile, the CPU, memory, and file system tests were run using SiSandra benchmarks (http://www.sisoftware.co.uk)[1]. Note that the platform power in Figure 1 does not include LCD since we have excluded it from our analyses due to the fact the monitor has its own external power supply. (Others include WLAN, HD-Audio, mini-card, ICH, and other peripherals.)

Figure 1. Platform Power Profile

Testing Methodology

Three HD video playback applications were tested on our in house Software Development Platform (SDP) (Intel® Core™2 Duo Mobile Processor). The SDP was instrumented for power measurement/characterization. The playback applications were characterized while playing three different video titles, each with a different encoding format. We also tested with and without enabling Hardware Acceleration via Video Hardware Configuration mode within the applications for each workload. In addition, additional supporting tests were conducted on a Sony Vaio* OEM (also with Intel Core 2 Duo Mobile Processor, T7300 2.0 GHz, 2 GB RAM) using the same major video playback applications. The goals of this paper are:

To understand the impact of the optional HW acceleration feature on battery life while playing different encoding.

To identify the power gaps that must be filled to achieve two hours or more of playback.

The Workload

The workloads used for power analyses on both the Intel Software Development Platform SDP and the Sony Vaio were three common HD Video playback applications while each application was playing one of the following video titles:

• “RV” – encoded in MPEG-2
• “Planet Earth” – encoded in VC-1
• “Casino Royale” – encoded in H.264

Configuring the Experiment

The experiment configuration settings were as follows:

• Storage: Blu-ray discs*
• Optical Drive: Blu-ray
  • Pioneer SATA* Blu-ray drive in desktop form factor (Merom Dual Core engineering system)
  • VGN FZ-190* with Blu-ray drive (Sony Vaio)
• Operating System: Microsoft Windows Vista*
• Vista Power Plan: Balanced
• Screen: Full Screen
• Brightness: Full
• Battery Level: 100% at start of test
• Testing Time & Nature: 15 minutes per titles starting at Chapter 1 for each test run

HD Video Playback Applications

The graphs in each video playback application section below describe the power characterization results for the applications while running on Intel engineering system. Time measured in seconds. Power measurements were acquired using a Fluke NetDAQ system and corresponding software (v4.0), which reports average power (in watts [W]) which was then converted to total power using application run-time data.

HD Video Playback Application-1

Each study describes our analysis of a leading video playback application. Application #1 comes with a mobile specific pack. Tests were performed under default settings. The hardware acceleration mode in this particular software uses two configurations: 1) Hardware Decode Acceleration and 2) Color Acceleration. We found that operating with hardware acceleration on delivered the anticipated power benefits, particularly within the CPU, and that the savings varied by codec.

Figure 2 shows the relative average power consumption of various components in the platform with hardware acceleration on and off. There is clearly a reduction in the energy consumption of the CPU (resulting into platform power savings) with hardware acceleration turned on. The data also shows the impact of the various codecs on CPU power consumption. The specific data for Figure 2 is shown in Table 1.

Figure 2. Application-1 HD Video Playback

Table 1. Power Consumption during Application-1 HD Video Playback

Component	Casino Royale (H.264)		Planet Earth (VC-1)		RV (MPEG-2)
	HW Acc On	HW Acc Off	HW Acc On	HW Acc Off	HW Acc On	HW Acc Off
Blu-Ray Drive	3.63	3.62	3.63	3.64	3.89	3.88
SDVD	0.00	0.00	0.00	0.00	0.00	0.00
HDD	1.54	1.55	1.54	1.55	1.58	1.54
CPU	6.39	18.64	6.55	8.57	3.23	3.19
Memory	0.33	0.60	0.67	0.76	0.38	0.34
GMCH	1.17	1.95	1.55	1.79	1.18	1.16
Other1	39.51	40.25	40.26	39.80	38.02	37.76
Total Platform	52.57	66.62	54.20	56.11	48.28	47.88
LCD Display	Not instrumented
Nvidia 8600 GTS	Instrumented as part of Total Platform

From the data in the Table 1, we easily see that CPU power and total platform power consumption are usually lower when hardware acceleration is on. The greatest CPU power consumption occurred with H.264 decoding (Casino Royale) at 18.4W with total platform power at 66.62W. In the case of VC-1 (Planet Earth) and MPEG-2 (RV), CPU power was at 8.57W and 3.19W respectively.

Observations from this study:

1. Various codecs differ considerably in their computational demands and corresponding power consumption. MPEG-2 is the easiest to decode, followed by VC-1, and then H.264, which is the most complex. We expect that H.265 is likely to be even more computing intensive than H.264.
2. The power savings due to hardware acceleration come almost completely from the power saved in the CPU. For example, with hardware acceleration on with H.264, the CPU power drain dropped by 12.25W and total platform power dropped by 14.05W – so most of the savings came from the CPU with some additional savings from the memory and chipset.

HD Video Playback Application-2

Application-2 is another industry leading video playback application. It also offers special mobile features to enhance battery life for video playback on mobile platforms. As indicated before, we did not change the default power configuration mode in this application as the primary focus is on the benefits of hardware acceleration.

Figure 3 shows the platform power consumption. As with Application-1, all decoding was power hungry and higher in particular with H.264 decoding. Once again, we can see significant differences between the energy demands of the decoders.

Figure 3. Application-2 HD Video Playback

Analysis done with this software indicated that compared to Application-1, this application had a somewhat more efficient H.264 decoder. The CPU power consumption with HW Acceleration off was only 14.93W, an improvement of over 3 watts as compared to Application-1. On the other hand, Application-2 had higher number for decoding VC-1 and MPEG-2. Application-2 appears to have been optimized chiefly for H.264. The details for Application-2 are shown in Table 2.

Table 2. Power Consumption during Application-2 HD Video Playback

Component	Casino Royale (H.264)		Planet Earth (VC-1)		RV (MPEG-2)
	HW Acc On	HW Acc Off	HW Acc On	HW Acc Off	HW Acc On	HW Acc Off
Blu-Ray Drive	3.59	3.54	3.62	3.63	3.28	3.28
SDVD	0.00	0.00	0.00	0.00	0.00	0.00
HDD	1.60	1.56	1.54	1.56	1.54	1.53
CPU	5.45	14.93	7.22	11.20	3.58	4.84
Memory	0.35	0.70	0.83	0.72	0.57	0.62
GMCH	1.31	1.92	1.76	1.83	1.40	1.58
Other1	40.01	39.59	41.17	40.31	37.58	39.29
Total Platform	52.36	62.23	56.15	59.25	47.96	48.46
LCD Display	Not instrumented
Nvidia 8600 GTS	Instrumented as part of Total Platform

HD Video Playback Application-3

The third study describes analysis done on one of the less commonly used applications for HD Video playback. Unlike the Applications-1 and -2, Application-3 showed some startling and unexpected results. Figure 4 shows the relative power consumption for all codecs.

Figure 4. Application-3 HD Video Playback

As shown in Table 3, CPU power consumption for Application-3 was surprisingly higher for H.264 decoding with hardware acceleration turned on by more than 2 watts. Application-3 also consumed more power than the other apps when decoding MPEG-2 and VC-1. A root-cause analysis of Application-3 yielded the reason. H.264 video playback on Application-3 is not fully supported with hardware acceleration. Multiple patches and plug-ins are required to playback HD videos most efficiently.

Table 3. Power Consumption during Application-2 HD Video Playback

Component	Casino Royale (H.264)		Planet Earth (VC-1)		RV (MPEG-2)
	HW Acc On	HW Acc Off	HW Acc On	HW Acc Off	HW Acc On	HW Acc Off
Blu-Ray Drive	1.66	3.56	1.52	3.67	3.66	3.66
SDVD	0.00	0.00	0.00	0.00	0.00	0.00
HDD	1.86	1.56	3.63	1.56	1.51	1.59
CPU	19.16	17.60	7.67	13.04	6.61	8.09
Memory	1.03	0.66	0.75	0.66	0.57	0.61
GMCH	2.20	2.01	1.75	1.98	1.52	1.90
Other1	43.80	41.34	42.24	41.59	41.28	40.45
Total Platform	69.70	66.74	57.55	62.50	55.15	56.30
LCD Display	Not instrumented
Nvidia 8600 GTS	Instrumented as part of Total Platform

Summary

This section summarizes our findings with particular focus on the impact of hardware acceleration on CPU power savings for the various applications and decoders.

Figure 5 and Table 4 focus on MPEG-2 and provide a relative comparison of the CPU energy consumed by the applications with HW acceleration On and Off. Application-1 is the most energy efficient overall. Applications-2 and -3 show clear benefit when hardware acceleration is used. We find it interesting that there is a difference of 5 watts between Applications-1 and -3 in the absence hardware acceleration. With that savings, Application-1 could provide up to 30 minutes more playback time on a standard notebook battery.

Figure 5. CPU Power Consumption during MPEG-2 HD Video Playback

Table 4. Power Consumption during MPEG-2 Playback Attributed to CPU

Hardware Acceleration	Application-1	Application-2	Application-3
HW Acc On	3.23	3.58	6.61
HW Acc Off	3.19	4.48	8.09

Figure 6 and Table 5 show the energy cost of decoding VC-1. Again, Application-1 is more power efficient than Applications-2 and -3. This holds true with hardware acceleration On or Off.

Figure 6. CPU Power Consumption during VC-1 HD Video Playback

Table 5. CPU Power Consumption during VC-1 Playback

Hardware Acceleration	Application-1	Application-2	Application-3
HW Acc On	6.55	7.22	7.67
HW Acc Off	8.57	11.20	13.04

With H.264 encoding (Figure 7 and Table 6), the story changes slightly. Application-2 is more power efficient than either Application-1 or Application-3 regardless of hardware acceleration settings.

Figure 7. CPU Power Consumption during H.264 HD Video Playback

In Table 6, the study shows that H.264 encoding is significantly consume more power when compared to MPEG-2 and VC-1. This is particularly true when hardware acceleration is disabled. It is important to stress that Application-3 has technical difficulties when playing HD video playback. This is particularly true with H.264 encoding.

Table 6. CPU Power Consumption during H.264 Playback

Hardware Acceleration	Application-1	Application-2	Application-3
HW Acc On	6.39	5.45	*19.16
HW Acc Off	18.64	14.93	17.60

NOTE: *A deeper root-cause analysis of Application-3 yielded the facts. H.264 video playback on Application-3 is not fully supported with hardware acceleration on.

In conclusion, Figure 8 and Table 7 show a full comparative analysis for all applications used in this power study. Application-1 is the most energy-efficient HD video playback software for MPEG-2 and VC-1 encoded content. However, Application-2 is more efficient-efficient for H.264 encoded content. Application-3 is the least energy-efficient software for high definition decoding.

Figure 8. CPU Power Consumption during HD Video Playback Applications

Table 7. CPU Power Consumption during HD Video Playback Applications

Hardware Acceleration	Application-1			Application-2			Application-3
	MPEG-2	VC-1	H.264	MPEG-2	VC-1	H.264	MPEG-2	VC-1	H.264
HW Acc On	3.23	6.55	6.39	3.58	7.22	5.45	6.61	7.67	19.16
HW Acc Off	3.19	8.57	18.64	4.48	11.20	14.93	8.09	13.04	17.60

Recommendations

The study above shows the significant effects of hardware acceleration for various codecs used for high definition video playback, and particularly the energy saved in the CPU.

This section makes some recommendations for software developers, original equipment manufacturers (OEMs), and users that will help reducing energy consumption and extend playback time on mobile computers.

Develop Software to use Hardware Accelerators

For software developers, this study recommends taking advantage of video decode hardware accelerators to reduce CPU and overall platform power usage.

Enhancing and/or Upgrading Hardware

For OEMs and ultimately for consumers, Intel continues to develop more energy-efficient computers. Next generation Intel consumer platforms, both mobile and desktop, will include energy-efficient 45nm processor technology (Penryn) and dedicated hardware accelerators for high-definition video decode. These new platform architectural changes will reduce CPU and total platform power and boost performance. In addition to microarchitecture enhancements, Penryn processors have two cores to support multi-threading and Intel® SSE4 instructions to enhance performance. In addition, the introduction of Deep Power Down Technology (C-6) significantly reduces CPU power consumed in an idle state, as illustrated in the Figure 9[2].

Figure 9. C-States Residency Idle power vs. Responsiveness

Optimize the Software

In the near future, high definition content will be pervasive on the internet and yet many of the millions of computers owned by consumers do not have hardware accelerators. We therefore recommend that developer’s video playback applications take steps to optimize software decoders to take advantage of hyperthreading, multiple cores, and special instruction sets such as Intel SSE4. There are also data efficiency techniques such as memory buffering and caching that can further improve energy-efficiency[3].

Increase OS Settings Awareness

One way for consumer to alter CPU performance and energy-efficiency is to set a platform power profile. For example, selecting “Max Performance” or similar profile will set the CPU at the highest operating frequency. Selecting “Max Battery” or similar profile will place the CPU at the lowest operating frequency. Max Battery is energy-efficient, but it may also reduce the quality of video playback. The best policy is to set the “Balanced” profile and allow Intel SpeedStep™ technology and the OS to dynamically change the CPU frequency on demand. The graph in Figure 10 shows the energy consumption of various power profile on a Sony Vaio with MS Vista while playing H.264 content with Application-1. LCD brightness was at 50%[4]. Playback in “Power Saver” mode resulted in dropped frames and faltering playback. The “Balanced” profile used less than 1 watt more and provided flawless playback.

Figure 10. Vista Power Profile Burn Rate

Conclusion

It is clear from our analysis that hardware accelerated video decode saves energy over doing the same task completely in software. We have also shown that there are differences in the software decoders. For the same encoding technology, some software decoders are more energy-efficient than others. Software decode energy-efficiency can be improved by using advanced instructions and multi-threading. With these approaches, playback time on mobile devices can be extended enough to play an entire HD movie on a single charge from a standard battery.

As a side-note, we performed some experiments on a Blu-ray equipped Sony Vaio with a 52WHr battery. Under the best circumstances, we were able play a movie for about 90 minute. The system was drawing about 36 watts on average over the 90-minute period. To get two hours of playback, we will need to reduce the 36 watts to 26 watts – a reduction of 10 watts. From the component measurements we have taken, it does not appear we will get this reduction from the optical drive, the hard drive, the display, the memory, or the chipset. We will have to get it from the CPU or the graphics card – the systems doing the video decoding. Next generation Intel systems with hardware accelerated decode will eliminate the need for a graphics card for video decode, save energy, and extend mobile system battery life.

References

1. Bob Steigerwald, Rajshree Chabukswar, Karthik Krishnan and Jun De Vega, Creating Energy-Efficient Software http://softwarecommunity.intel.com/articles/eng/1458.htm
2. Rajshree Chabukswar, DVD Playback Power Consumption Analysis http://softwarecommunity.intel.com/articles/eng/1089.htm
3. Aleksandr Budik, 45-nm Penryn and Nehalem: architectural details, http://www.digital-daily.com/cpu/intel_penryn_nehalem/
4. WikiPedia web resources, http://en.wikipedia.org

About the Authors

Tareq Darwish is a Software Engineer working on Platform Power Enabling as part of client enabling in the Software Solutions Group. His current focus is on defining tools and technologies to support the development of energy-efficient software for Intel-based mobile platforms. Prior to working at Intel, He worked for 9 years with Lexmark Int. in Lexington Kentucky as a Software Development Engineer. He earned his MS degree in Applied Computing and Software Engineering at Eastern Kentucky University. His email is tareq.h.darwish@intel.com

Rajshree Chabukswar is a Software Engineer working on client enabling in the Software Solutions Group that enables client platforms through software optimizations. Prior to working at Intel, she obtained a Masters degree in Computer Engineering from Syracuse University, NY. Her email is rajshree.a.chabukswar@intel.com.

Kiefer Kuah is a software engineer in the Software Solution Group. He has worked on video codecs, video editors, and games as part of client enabling team. His email is kiefer.kuah@intel.com

Bob Steigerwald is an Engineering Manager in the Software Solutions Group at Intel in Folsom, California. He received his B.S. degree in Computer Science from the US Air Force Academy, Masters in CS from the University of Illinois, and Ph.D. in CS from the Naval Postgraduate School where his research was in software engineering and software reuse. Currently his team works on defining tools and technologies to support the development of energy-efficient software for Intel-based mobile platforms. His email bob.steigerwald@intel.com.

Appendix A: Definition and Technology

Encoding: The process of taking original information and transforming it into another. Video encoding typically includes compression of the original format.

Decoding: The process of transforming information from one format into another. (The reverse of encoding). Video decoding occurs when you play the video or when you are preparing a video stream for encoding into another digital format.

Transcoding: The process of decoding a video source and then encoding it into another format, e.g. transforming an MPEG-2 encoded source into H.264.

Codec: A device or program capable of performing encoding and decoding of a digital data stream or signal. The word codec may be a combination of any of the following: 'compressor-decompressor', 'coder-decoder', or 'compression/decompression algorithm'.

Video codec: A device or software that enables digital video compression and/or decompression

MPEG-2: Motion Picture Experts Group as well as the process of capturing (digitizing) or converting video and/or audio to one of several standardized video and/or audio formats for distribution (Internet, LAN) or for archiving to optical disc (CD, DVD).

VC-1: The informal name of the SMPTE 421M standard initially developed by Microsoft. It was released on April 3, 2006 by SMPTE. It is now a supported standard for HD DVDs, Blu-ray Discs, and Windows Media Video 9.

H.264: A standard for video compression. It is also known as MPEG-4 Part 10, or MPEG-4 Advanced Video Coding (AVC). It was written by the ITU-T Video Coding Experts Group (VCEG) together with the ISO/IEC Moving Picture Experts Group (MPEG-2) as the product of a partnership effort known as the Joint Video Team (JVT).

HD DVD or High-Definition DVD: A high-density optical disc format designed for the storage of data and high-definition video. HD DVD was designed to be the successor to the standard DVD format and is derived from the same underlying technologies. Since all variants except the 3x DVD employ a blue laser with a shorter wavelength, it can store about 3¼ times as much data per layer as its predecessor (maximum capacity: 15 GB per layer instead of 4.7 GB per layer).

Blu-ray Disc (also known as Blu-ray or BD) is a high-density optical disc format for the storage of digital information, including video. The name is derived from the blue-violet laser used to read and write this type of disc. Because of its shorter wavelength (405 nm), substantially more data can be stored on a Blu-ray Disc than on the original DVD format, which uses a red (650 nm) laser. A Blu-ray Disc can store 50 GB, almost six times the capacity of a standard DVD.

Hardware Acceleration: In computing, is the use of hardware to perform some function faster than in software running on the normal (general purpose) CPU. Examples of hardware acceleration include performing acceleration functionality in graphics processing units (GPUs) and instructions for complex operations in CPUs.

Hardware Accelerator: The hardware that performs the acceleration, when in a separate unit from the CPU, is referred to as a hardware accelerator, or often more specifically as graphics accelerator. A video card also referred to as a graphics accelerator card.

Appendix B: Power Measurement Methodology

Measuring power usage of individual components in a mobile platform is not a trivial task. Various tools exist to provide a high-level estimate of the power consumed by a particular mobile platform, but they do not provide the granular details on specific components. A more accurate but invasive way to measure power will be to use data acquisition (DAQ) tools where specific hardware components are instrumented and a more granular power measurement can be logged. The following lists the platform details we used for our analyses, along with the power-measurement methodology.

Hardware

• Fluke NetDAQ* 2686A
• Target PC: Intel Core Duo/2.2 GHz Merom, Matanzas CRB, 2 GB (2x1 GB) DDR2 667 MHz, 80 GB SATA 7200 rpm , CD-RW/DVD-RW drive,(SATA Blu-ray drive in DT form factor, NVIDIA 8600 GTS*, ATX Power Supply*[5]), Microsoft Windows Vista Ultimate
• Host PC: Any IA32 system

Software

• Test Applications (different HD Video playback applications used)
• NetDAQ Logger: Fluke DAQ Software v4.0

Test Setup

• The Host PC can be any IA32 system with Microsoft Windows* XP and the NetDAQ* logger software. The logger collects the measured current and voltages and lets the user calculate the average power (W). The sampling interval we used for our entire analysis was 25 milliseconds. The platform power measurement does not include the LCD display of the mobile device.
• The NetDAQ has modules that are attached (individual wires) to the Target PC and measures the current and voltage drop across the sense resistors. The NetDAQ is connected to the Host PC via a crossover network cable.
• The Target PC: (Santa Rosa/Merom) has a special motherboard (Matanzas CRB) with built-in sensors that each target component (i.e., the CPU, Memory, HDD & etc), all sense resistors.

Appendix C: LCD Brightness

In Figure 11, Maximum Brightness consumes 44.5W and Minimum Brightness consumes 35.98W, which is 8.5 watt difference between Min and Max brightness. This difference can be used for ~15 minutes more playback.

Figure 11. Monitor Brightness Setting Burn Rate

In addition, another test was conducted to check the battery drain rate with regard to brightness level with setting as follow brightness at 50%, title was Casino Royale (H.264) using Application-2, Vista Balanced Power Profile, Acceleration on (NVIDIA), on OEM Sony Vaio. Standard Battery starts at 100%, drains until system dies.

Table 8 shows the average is 1 hr, 17 min, 39 sec with a std deviation of 48 seconds. So since 95% lay within ± 2 std dev, we can predict the movie time accurately within ± 2 minutes – very consistent.

Table 8. Application-2 H.274 Codec Battery Drain Rate

Battery Drain Rate on Application-2 with H.264 CODEC Brightness at 50%
Test No.	Battery Time	Remaining Capacity
1	1:18:29	6%
2	1:17:53	5%
3	1:17:18	5%
4	1:18:50	4%
5	1:17:11	6%

[1] Performance tests and ratings are measured using specific computer systems and/or components and reflect the approximate performance of Intel products as measured by those tests. Any difference in system hardware or software design or configuration may affect actual performance. Buyers should consult other sources of information to evaluate the performance of systems or components they are considering purchasing.