Processing DV Production
Posted Oct 22, 2003

As a small-shop digital studio, how much time will the latest hyperthreaded PC processors and dual-processor systems save you on your video editing work, and how fast will you recoup your investment on these speedy new systems?

In a previous life, I was VP of sales for a firm that sold restaurant computer systems. Called Squirrel, this is actually the company that makes those elegant touchscreen terminals you may see now and then in finer restaurants. Working with a rookie sales rep one day, we demonstrated the system to a restaurateur who had just purchased a run-down country club.

I had done my homework beforehand, identifying needs and savings, and once on-site we did a good job showing how the system would easily pay for itself in less than two years. The owner had nodded his acceptance during the demo, so I was looking forward to a smooth close.

When he declined to move forward, I pointed out that he had agreed that the system would pay for itself in two years or less. He replied, "That's what all the suppliers prove, Jan, including the PBX, rug, and furniture supplier. If I purchased everything that would pay for itself in under two years today, I'd be broke and out of business by next week. My priority is getting the customers in the door and guaranteeing a good experience. So it's phones and furniture first, then the computer system."

Perhaps if I had been Zig Ziglar or Tom Hopkins I would have had a glib reply, a surefire way to overcome this objection. But his logic made sense, so we wished him luck and promised to keep in touch.

Like the restaurateur in my story, video professionals are bombarded with offers of products that will improve the quality of their productions, streamline their efforts, or both. However, if you bought them all today, you would definitely be out of business next week.

Let me suggest that the priority for any one- or two-person video shop should be saving time. If you can find a product that will shave hours off your production cycle—time that you can then convert to more billable hours, higher margins, or both—you should buy it. If not, you should decline—politely, of course.

Here, we look at Windows computer systems with a strict focus on the payback period for video studios. We make some assumptions about billable hours and productivity up front, and then use these to determine how long it will take for a particular computer, or computer component, to pay for itself in terms of time saved.

Our Assumptions
Here are our critical assumptions, and you can scale our findings one way or the other depending upon how they relate to your actual situation. First, we assumed that you would not purchase a hardware-based editing solution, preferring to use the processor-based, real-time editing features offered in Premiere and most other current video editors.

In computing how long it would take to recoup your investment, we assumed that you were producing 30 minutes of video per week, output to DVD, using Adobe's new Premiere Pro (see sidebar, "Hyperthreading? That's What I'm Talking About!"). We assumed that you have more than one computer, and aren't totally dead in the water during final rendering. So if the equipment saved you an hour of rendering time, we figured you would be 75% productive during that time (compared to your productivity on your primary PC when it's not bogged down with rendering), and assumed that the equipment would boost productivity by 25%. Simply stated, if the computer rendered one hour faster than our baseline test bed, we assumed that you gained 15 minutes of productive time. sidebar mentioned here

Like most editors, Premiere Pro offers real-time previews for most effects, but we assumed that for most projects, you'd like to see actual rendered video, just to be sure. While you wouldn't necessarily preview every frame, we figured you would probably preview the most complex edits once if not several times. To simplify testing and our mathematics, we assumed that you would preview the entire project once during the course of production. Some sections several times, some sections never, but averaging out to the entire video once.

Unlike end-of-project rendering, however, preview speed significantly assists productivity, perhaps not 1:1, because you still need time to check email and get that 15th cup of coffee, but more than final rendering. So we assumed that you would be only 25% productive during that period; if the system saved you an hour of preview time, we assumed that this would convert to 45 minutes of additional productivity.

We assumed that you're currently working with a 2.4gHz Pentium 4 computer without hyperthreaded (HT) technology, with 500MB RAM and non-SCSI hard disk drives. Finally, we assumed that you'd like to make at least $75/hour while editing, and that you bill the client by the job, not the hour. Thus, the productivity added by this equipment should translate directly into higher margins and more billings.

Our test project was a 30-minute video with multiple segments, background audio, and a range of motion, chroma key, overlay, picture-in-picture, blur, and other effects. We tested on an HPxw4100 workstation with a 3.2gHz Pentium 4 processor and a dual 3.06gHz Xeon Dell Precision 650 workstation, changing processor and RAM configurations as needed for our tests. 

For system pricing, we averaged prices from Dell, Gateway, and Hewlett-Packard for the Pentium 4 system, and prices from Dell and Hewlett-Packard for the single and dual-processor Xeon systems.

Single- or Dual-Processor Computer
Table 1 looks at the ROI of upgrading from our baseline 2.4gHz Pentium 4 to one of the three benchmarked systems, in terms of the time saved in rendering performance as applied to other work, and the increased billable hours applied toward recouping the price of that new PC.

TABLE 1: The ROI of Buying New

P4
2.4gHz
P4
3.2gHz
with HT

Xeon
3.06gHz
with HT

Dual Xeon
3.06 gHz
Billing rate$75$75$75$75
Preview Time (minutes)132677244
Time saved per week (75%)n/a48.754566
Rendering Time11188.481.547.7
Time saved per week (25%)n/a5.657.3815.83
Time saved per week (25%)n/a54.452.3881.83
System pricen/a$2,273$2,882$4,381
Weeks to recoup investmentn/a334443

Notes:
Abbreviations: P4=Pentium 4; HT=Hyperthreading
Times are in minutes and percentages of minutes, not seconds
System prices include 500MB RAM, Windows XP Professional, the least expensive 64MB graphics option, an 80GB hard drive, FireWire card, and the least expensive DVD recorder option offered by the vendor.

As the table shows, in terms of pure weeks to payback, the hyperthreaded Pentium 4 is the most attractive option, though the Dual Xeon system with its staggeringly fast rendering times is very alluring. Obviously, if your current system is much slower than the 2.4gHz Pentium 4 system, converted time savings will be greater and weeks to recoup will be shorter. 

Another assumption, apropos of this audience-flattering age: all you video professionals out there get things right the first time. On the other hand, if you're like me and typically render each project two or three times before final because you have ample reason to believe you didn't get it right the first or second go-round, payback times get much shorter.

Assuming (here we go again) that you know where you fit in the "how many full renders per project?" continuum, let's move on to the RAM side of the equation.

500MB or 1GB?
Here we start with the assumption that you purchased the dual-processor system and test to determine whether you need to go with 1GB RAM, or if 500MB will do. Note that Premiere Pro's minimum system requirement calls for 256MB of memory, but recommends 1GB.

We performed three additional tests on the dual Xeon system:
• Test 1: Configure system with 1GB RAM, run Premiere, and re-run original rendering and preview tests.
• Test 2: Configure system to 500 MBRAM, load Adobe PhotoShop, Encore, Audition, and After Effects to push total memory requirements to above 500MB, then run Premiere and re-run original rendering and preview tests.
• Test 3: Configure system with 1GB RAM, load Adobe PhotoShop, Encore, Audition, and After Effects to push total memory requirements to above 500MB, then run Premiere and re-run original rendering and preview tests.

What these tests show is that if you're running Premiere standalone, with no other programs loaded, upgrading to 1GB provides no real performance boost. On the other hand, if you run the program with a host of other supporting applications open, forcing Premiere to page to disk frequently during preview and encoding, you slow performance by about 25%. You recoup that entire loss and more if you add 500MB RAM, and the upgrade will pay for itself in 15 weeks.

If you're running Premiere, upgrading to 1GB of memory seems like a no-brainer decision, whether you're buying a new computer or simply retro-fitting your current machine. Now let's look at the graphics side of the equation.


TABLE 2: Upgrading to 1GB RAM

500MB RAM
Premiere only
(original test)
Test 1
1GB RAM
Premiere only
Test 2
500MB RAM
other apps loaded

Test 3
1GB RAM
other apps loaded

Billing rate$75$75$75$75
Preview time (minutes)444457.4144.1
Time saved per week (75% of savings)n/an/a0.009.98
Rendering time (minutes)4746.0862.1646.08
Time saved per week (25% of savings)n/a.2504.02
Total time saved per week (minutes)n/a.25014
Upgrade pricen/a$255n/a$255
Weeks to recoup investmentn/a816n/a15

Graphics for Editing
To test whether a high-performance graphics card could produce additional hard productivity gains, we tested with both an NVIDIA Quadro FX 2000, a $2,000 option, and NVIDIA QuadroFX 500, a $149 option. The results in both preview and rendering were virtually identical, indicating that Premiere, like most video editors, doesn't stress the computer's graphic subsystem.

The only exception is Pinnacle Edition, which actually uses the 3D graphics processing unit on some boards to accelerate 3D effects. However, Premiere Pro's system requirements call only for a 1,024x768 32-bit color video display adapter, with 1,280x1,024 or dual monitors recommended.

The obvious lesson: buy the features you need and little else. If you have dual flat-panel monitors, opt for a card that can support them. In general, however, any 64MB card will do the job, so unless you're performing 3D-intensive work in other applications, don't invest significantly in graphics.

Hard Disk Drives
Both Dell and HP charge a high premium for SCSI-based systems and peripherals. For example, had we configured our single-processor system with a SCSI hard disk, it would have cost around $3,200—about $1,000 more than the Serial ATA-based system.

Moreover, to upgrade to a 120GB SCSI drive would cost around $900, $650 more than an IDE or Serial ATA drive. Do these faster drives produce enough speed boost to justify the expense?

To test this, we configured the single-processor HP system with one drive, ran the preview and rendering trials, and recorded the results. Then we added a 15,000RPM SCSI drive to the system, recaptured the video to that drive, and retested.

We experimented with several iterations of placing the temporary and output files in different locations, even adding another 10,000RPM SCSI drive to the mix to free the system drive from all rendering-related work. The best combination shaved less than two minutes off preview and rendering times.

TABLE 3: Hard Disk Speed Comparison

Single-Drive System15K RPM SCSI Disk
Billing rate$75$75
Preview time (minutes)6765.3
Time saved per week (75% of savings)01.28
Rendering time (minutes)88.486.8
Time saved per week (25% of savings)00.4
Total time saving per week (minutes)01.68
System pricen/a$1650
Weeks to recoup investment0788

Assuming we had purchased the SCSI-based system and an additional 120GB drive, our additional cost would be about $1,650, which would take around 15 years to recoup—obviously not a good investment.

The tests documented in Table 3 indicate that when performance is slower than real time, as in rendering and preview, disk speed isn't the gating factor; processor power is. So we wondered whether a more demanding application would leverage the advantages of the SCSI drives.

To test this, we set up a project with multiple picture-in-picture effects, and then previewed, which should occur in real time. Starting with video files stored on the Serial ATA C: drive, we added additional streams until Premiere started dropping frames. From the Serial ATA drive, this started when we added the third stream.

Then we started over, pulling streams from the SCSI drive, finding that the 15,000RPM SCSI drive could pull four streams without dropping frames, and started failing when we added the fifth. Accordingly, if your projects involve a lot of multiple-performance, multiple-stream effects, and real-time preview is a key factor for you, you may want to consider spending the extra money on SCSI.

Clearly, for video server and other similar applications where retrieval speed and efficiency is critical, SCSI drives are essential. On the other hand, for most other projects, Serial ATA and IDE will do just fine, with much less hit on the pocketbook and very little impact on performance.

The ROI of HT, DP, and Otherwise Accelerated DV
To summarize, make your most significant investment in the processor, definitely pursuing a hyperthreaded or dual-processor solution. Of course, as we see in the sidebar, if you're not working with Premiere, your results will definitely vary.

Overdoses of RAM provide little benefit, so going over 1GB probably isn't worth your while. Buy a hard disk for capacity not speed, and stay pretty close to the minimum requirements for graphics, since extra money here will definitely not pay for itself.

And you may not thank me now—after I've thrown all these non-biased, real-world numbers your way, and still come out recommending that you pour money into new systems—but next time a sales rep gets in your face and tells you that the product will pay for itself it two years or less, now you know one objection that he or she may not have heard before.

SIDEBAR
Multi-Threaded? That's What I'm Talkin' About!

A very reasonable question would be why we used a beta copy of Premiere Pro for these tests? After all, it's only beta, and we really, really liked Pinnacle Edition in our last review.

Well, we'll save our review of Premiere Pro for next issue, but the reason we used it here is demonstrated in the processor utilization report: how effectively it takes advantage of hyperthreaded processors.

You can access this screen on your Windows XP desktop by clicking ctrl-alt-del and clicking the performance tab. This one is from the dual-processor (DP) 3.06 Xeon processor Dell Precision 650 workstation.  To be clear, this computer has two Xeon processors, each with hyperthreaded (HT) technology, so each looks like two processors to the operating system. Two times two, in this case, is the predicable four.

What's unique and exciting about Premiere, is the extent to which it utilizes all available resources. As you may know, programs have to be specifically written to support HT technology and multiple processors. It doesn't sound easy, and it's not, since you have to figure out a way to divide the rendering and encoding task into four separate streams without adding so much administrative overhead that you slow the process.

Obviously, this was a very high priority for the Adobe engineers, because they performed magnificently, with all four threads utilizing close to 100% capacity. If you were to draw a picture of the ideal multi-threaded efficiency, this would be it. In contrast, the other three prosumer editors utilized only about 30% of total resources, leaving lots of otherwise unutilized processor cycles laying there on the table.

For our purposes, that means that if you throw extra hardware at the rendering and encoding process, Premiere Pro will fully and completely take advantage of it. Without naming names, several of the other editors showed little or no improvement when rendering on the dual-processor as opposed to a single-processor system. Obviously, this means that their code hasn't been modified to take advantage of HT/DP systems, or that it's simply not as efficient as Premiere's.

In these trials, Premiere Pro was our fastest performer, producing the 30-minute project in just under fifty minutes (47:43), compared to MediaStudio Pro at 50:15, Edition at 69:14, and Vegas at 2:56:59. The good news for these competitors—especially Ulead and Pinnacle—is that if they can produce the kind of processor efficiency that Premiere has delivered in their next versions, they should blow by Adobe in performance. The bad news for the other contenders for the software NLE crown is that this is beta code for Adobe, and with debug code taken out, and all optimizations entered, Premiere may become even faster.

In the short run, if you're using an editor other than Premiere, you may not see the performance gains described in this review, particularly from adding processing power. Obviously, however, now that HT systems are common, DP systems affordable, and comparative performance clearly measurable, efficient HT/DP code should become a priority for all companies.

In the meantime, a hearty well done to Adobe for showing the way.

COMPANIES MENTIONED IN THIS ARTICLE
Adobe Systems, Inc., www.adobe.com
Dell Computer, Inc., www.dell.com
Hewlett-Packard Company, www.hp.com
Gateway, Inc., www.gateway.com
Intel Corporation, www.intel.com
NVIDIA Corporation, www.nvidia.com
Pinnacle Systems, Inc., www.pinnaclesys.com
Sony Pictures Digital/Screenblast (Vegas), www.screenblast.com
Ulead Systems, Inc., www.ulead.com