|
Though the exact time and date are fuzzy, I guesstimate that I compressed my first video file in October 1993; I used a state-of-the-art 80386 Gateway system that took 45 minutes to render a 20-second AVI file to Indeo format. Whatever the exact time and date, about 2 minutes later was the first time I wished for a magic bullet to accelerate video compression.
Well, LSI Corp. is in the business of creating compression acceleration magic bullets, at least for Windows Media encoding. I was excited to try its new LSI Tarari Encoder Accelerator LCPX-6140 card, which is available for less than $6,000. The results were good
to great, depending on your encoding program, source files, and targets and workflows.
First, a little background. The San Diego-based fabless semiconductor company Tarari, Inc., was acquired by LSI Corp. in late 2007 for $85 million in cash. Tarari has emerged as a leader in content processor silicon to offload and accelerate computeintensive, complex algorithms used for content inspection in XML/web services, network security and digital media environments.
The 6140 is built around field-upgradeable, programmable processor chips from LSI. System requirements are stiff but not outrageous; you need a dual-processor, dualcore
Intel Xeon or AMD Opteron 2 GHz system or faster with 2GB RAM recommended. You must install the card in a 3.3 volt PCI-X bus slot, and, for the best performance, LSI recommends a 64-bit bus running at 133 MHz. The only operating systems currently supported are Windows XP and Windows Server 2003. Vista support is in the works. You can install ultiple cards in a single computer, which accelerates multiple simultaneous encoding
instances, which you could also do by opening multiple instances of the Window Media encoder. The card currently accelerates only VC-1/WMV 9 encoding, but it may support .264 encoding once Microsoft adds that to Expression Encoder. The card accelerates both live streaming and disc-based transcoding; I tested the latter.
I installed a single card on an HP xw8400 workstation running Windows XP. Installing the card was standard fare, as was installing the software. The only wrinkle was making anual registry changes, which turned out to be straightforward. After a quick trip to RegEdit and
a glance at the documentation, I was done. Once installed, you’ll see a little red diamond icon in the system tray on the bottom-right corner of Windows. This icon blinks happily when you are accelerating a Windows Media encoding run. If the encoding software turns out to be incompatible with the Tarari accelerator, the icon does not blink. Which leads us to the next question: Who’s in and who’s out?
Compatible Programs
According to LSI’s product brief, the “LSI Tarari encoder Accelerator provides seamless acceleration for any WMV and VC-1 encoding applications based on the Microsoft Windows Media Format SDK—without the need for any special scripting or digital glue.” Obviously, this means that there are programs that won’t be accelerated. I checked the status of the most commonly used Windows encoding programs with the LSI representative and then consolidated the information.
As you can see, most of the prominent encoding tools work today or will work very soon. In addition, note that if you’re encoding using scripts that call Microsoft’s Professional, nterprise, or even the older SDK 11 for WMV, the Tarari board will accelerate your rendering.
The Adobe situation stems from the company using a MainConcept API rather than a Microsoft API for the Adobe Media Encoder. Barring any changes, it may not be possible to make the Tarari accelerators compatible with the Adobe Media Encoder. On the other hand, LSI has been compatible with previous versions of Sorenson Squeeze, and the company promised the necessary fixes very soon. Still, if you’re buying the software to use it with Sorenson Squeeze, you should first verify that LSI has completed the necessary work.
With coprocessor cards such as Tarari, your results will vary based on the encoding program you’re using, input and output resolution, and the encoding setting you’re using (1-pass or 2-pass encoding). When comparing your results to baseline, computeronly encoding, the speed and the number of cores in the test computer will also affect the results. To try to illustrate this, I ran different tests on the programs that I had available.
Let’s start with Canopus ProCoder, a relentlessly efficient encoding program on multicore systems such as my test bed HP xw8400. Here, I encoded five 1-minute SD files to 640x480 resolution at a data rate of 500Kbps (468 video/32 audio) and three 720p files to 1280x720p resolution at a combined data rate of 928Kbps (800 video/128 audio). I produced all files using 2-pass encoding since that’s my general practice.
With the Tarari accelerator, it took 11:21 (min:sec) to produce the files; without the card, it took 45:46. On a percentage basis, this means that Tarari reduced encoding time by 75%. I compared the quality of the files produced with and without the card and found them visually
identical, though the final size varied slightly. Something was different, though it wasn’t noticeable.
I then rendered a single HD file to the same parameters in Rhozet Carbon Coder, using single-pass rather than 2-pass encoding. Encoding without the card took a glacial 20:30, while encoding with the card took 2:41—a time savings of more than 87%. I then reran the tests using 2-pass encoding to see if Tarari accelerated the first analysis pass. With and without the card, the analysis pass took about 2:20.
This illustrates why single-pass encoding should produce higher-percentage speed increases than 2-pass. Essentially, the first pass is a “fixed cost” that remains the same whether the Tarari card is used or not, bringing down the percentage of time savings. However, the actual minutes saved will be about the same, which is probably the more important metric.
Let’s move on to the Microsoft encoding tools: the venerable Windows Media Encoder and the more current Expression Encoder 2. I considered running identical tests to those used for ProCoder, but I was concerned that this might compromise the broad applicability of my findings. So in each encoder, I rendered three HD source files (two 1080i and one 720p) to
720p presets.
Using the Windows Media Encoder and running the jobs in sequence, it took 30:17 to produce the files without the Tarari card, which dropped to 11:19 with the card—a time savings of about 63%. Next, I opened three instances of the Windows Media Encoder and encoded all three files simultaneously. Without the card, this consumed 11:11, which dropped to 6:28 with the card — a savings of about 42%.
Why did time savings drop so significantly? During the trials without the card, I noticed that when encoding only one file, Windows Media Encoder consumed only 40% of the overall processing power of my eightcore HP test system. With all three programs running simultaneously, CPU utilization often reached 100%, which translates to better efficiency without the card and less time savings from the card on both an absolute and percentage basis.
I produced similar results with Microsoft’s Expression Encoder 2. When running the three encodes in sequence, it took 21:41 without the card and 5:49 with the card—a drop of 73%. With three instances of the Encoder open, rendering time dropped to 10:28 without the card and 6:50 with it—a 35% reduction in time. So even if you’re taking steps to make your Microsoftbased encodings as efficient as possible, Tarari will still deliver a significant benefit.
My tests felt rather clinical, so I wanted to supplement my results with some user input. Tarari supplied several very eager and happy customers. One was Darren S. Person, chief technology officer of the CBS Television Stations Digital Media Group. Person’s group is responsible for producing video from CBS’s locally owned and operated stations around the country. He uses the Tarari accelerator in two very distinct roles.
First, his group provides the tools that allow the stations to edit in Windows Media format, so the Tarari cards help to convert video submitted by the stations as well as some user-generated video to Windows Media format. Second, in the more traditional role, the Tarari cards accelerate the group’s final VC-1 encoding with Expression Encoder 2 for UGC Silverlight deployment.
In their second role, the Tarari boards cut encoding times by up to 70%, so encoding that used to take an hour now takes less than 20 minutes. The Tarari card doesn’t affect quality one way or the other; according to Person, it was “fantastic” before they got the Tarari cards
and “fantastic” afterward. In his group’s New York offices, Person reports that the key benefit of the cards is that it lets him and his co-workers “scale up, rather than scale out,” doing more in less time with one computer rather than investing in space-consuming and powerhungry server farms.
Next, I spoke with Kevin Crumley, a senior video engineer at MSNBC who’s responsible for converting NBC’s satellite-based video feeds into streaming video for web distribution. Crumley noted that he’s used Tarari to accelerate encoding for several years, initially for the
video distributed to viewers over the web and now for producing video for archival purposes.
Working with QuickTime reference files produced by Avid editing systems, Crumley uses Telestream FlipFactory, accelerated by Tarari, to produce 500Kbps files at either 448x336 or 592x336 resolutions for archival. Though he hasn’t run recent benchmarks, he recollected that the boards accelerated encoding time by about 30%–40% when initially installed.
Finally, I traded several emails with Jay Naranjo, a media systems engineer with Technicolor Creative Services, who uses the Tarari card with Digital Rapids StreamZ software. In essence, Technicolor bought the cards to enable real-time VC-1 encoding. Working from a number of sources (including uncompressed SD and HD AVI, 30 Mbps–80 Mbps SD, and HD MPEG-2), Naranjo confirmed that the Tarari cards enabled him to meet this goal.
Overall, while your mileage will certainly vary, I was impressed with the results I achieved and the praise I heard from actual users of the card. Tarari seems to be a solid option for any high-volume producer who needs to accelerate his or her VC-1 encoding. It is also affordable compared to the fully burdened cost of buying, housing, and maintaining a separate computer or computers on a server farm to accomplish the same task. |
