Configuring Your System for CS4, Part 2

In the last installment, I described buying considerations and costs for system RAM and CPUs for Apple''s Mac Pro and HP''s xw8600 workstation. In this installment, I''ll present the results of a series of benchmark tests with the two systems. As you probably recall, both systems use quad-core Intel Xeon processors, and both used 64-bit operating systems—OS X 10.5.5 for the Mac and 64-bit Windows Vista Business, Service Pack 1, on the HP workstation.

Related Links

Configuring Your System for CS4, Part 1 This month, I''ll review how the number of processors and the amount of RAM impact encoding performance, and then find other, non-CS4-related fish to fry... CS4 and 64-bit Systems, Part 1 So there I was, testing Adobe Creative Suite 4 (CS4)''s AVCHD compatibility. I created a simple project, about 4 minutes long, two picture-in-picture overlays with simple rotation and color correction... Test Drive: Nvidia Quadro CX and Adobe CS4, Part 1 This month, in our quest to identify the optimal CS4 configuration, we look at graphic chip manufacturer Nvidia''s Quadro CX technology...

The benchmark tests differ by format and computer. To test Windows performance, the DV test is the longest (about 4 minutes) and most comprehensive, and it includes color correction, chroma key via Adobe After Effects using Dynamic Link, chroma key via Adobe''s native Color Key filter, transparency, titles, logo insertion, and other effects. To test performance under the various test configurations, I rendered this file to DVD-compatible MPEG-2 in Adobe Media Encoder.

The other tests were all 1 minute long and included multiple tracks, with spinning, 1/4-screen picture-in-picture effects, color correction, and chroma key via After Effects using Dynamic Link. With all HD formats, I rendered to Blu-ray-compatible MPEG-2 format in Adobe Media Encoder. I describe how these tests differed from the tests I performed on the Mac below, but the numbers are not comparable.

Both sets of tests certainly could have been longer and more complex, but the number of tests (38 on the HP Windows test system alone) dictated that they be short and sweet. Even at the 1 minute duration, the 4K Red source files still took at least 18 minutes to render—and it doesn''t take long to eat into nights, weekends, and quality time with the kiddies.

In all instances, I kept Premiere Pro open while encoding, though I wasn''t actively editing. After running all the timed benchmark tests, I re-ran the DVCPRO HD-to-Blu-ray project again and tried some minor edits, including rendering a portion of the timeline, to gauge system responsiveness during the encoding cycle. I report the results of those tests below.

Table 1. Rendering times for a dual-processor HP xw8600 workstation with differing RAM configurations.

Let''s start by looking for the tipping point RAM configuration for both single- and dual-processor systems on the HP xw8600 workstation. For the record, all rendering times are in seconds.

Table 1 shows the results for the dual-processor configuration. As you can see, the tipping point for DV source footage appears to be 8GB, since dropping to 4GB increases rendering time by more than 50 percent from the 16GB configuration. The drop-off in performance for HDV, AVCHD, and DVCPRO HD projects was still reasonable at 4GB, though HP makes this kind of moot by pricing 8GB of RAM only $150 more than 4GB. If HP changes its pricing, or if you''re buying from another vendor with different pricing, you should consider starting at 4GB and only buying more if performance is unacceptable. Check out Table 3 below, however, before you make that decision.

For Red source footage, it appears that only 16GB will do, since dropping to 8GB increases rendering time a whopping 138 percent. Note that I didn''t render the Red project at 2GB of memory because the times at 4GB were already unreasonable.

Table 2. Rendering times for a single-processor HP xw8600 workstation with differing RAM configurations.

Single-processor tests revealed several anomalies, which I tested twice in all instances. First, in the DV encoding, the 4GB configuration was actually faster than both 8GB and 16GB—which makes little sense, but there it is. With a single processor, I''d be tempted to buy only 4GB of RAM for DV projects and add more RAM later if performance was dragging.

HDV shows very little performance degradation even down to 2GB, which is impressive. Still, though performance didn''t suffer, responsiveness did in other applications, so if you plan to continue editing while rendering—or even use the computer to surf the Web—consider at least 4GB. The 4GB configuration also seems adequate for AVCHD and DVCPRO HD, but you should consider Table 3 before making a final decision.

Table 3. Rendering time for previewing an After Effects chroma key effect (via Dynamic Link) in a DVCPRO HD project while rendering.

Table 3 shows the results of the aforementioned tests to determine system responsiveness while rendering. Again, in these tests, I started rendering the DVCPRO HD project to Blu-ray-compatible MPEG-2, and then timed how long it took to render a 15 second portion of the Premiere Pro timeline containing an After Effects chroma key effect inserted via Dynamic Link.

As you can see, in 2GB and 4GB configurations, rendering took more than 4 minutes, which dropped to slightly more than 2 minutes at 8GB and 16GB configurations. This proves that if you plan on actively using the computer while rendering, you''ll need to bump up your RAM configuration. That said, on a single-processor system, it feels like you''d be hard-pressed to put more than 8GB of RAM to good use—except, perhaps, for Red-based projects.

Table 4. Decrease in encoding time from adding a second processor.

Let''s quickly look at the economics of adding a second CPU to a particular RAM configuration. As you can see, adding a second processor to the 16GB RAM configuration is all good, with 20 percent-plus drops in encoding time for all formats except for Red, which appears to need memory more than a second CPU. Nowhere near the 50 percent drop you''d love to see by adding the second processor, but substantial nonetheless.

At 8GB, the decrease in rendering time afforded by the second CPU drops for every format, with only an 11 percent performance boost in AVCHD and an increase in rendering time of 71 percent for the Red project. As you''ll see in our next analysis, if you''re producing with Red source video, having adequate RAM is more important than dual processors.

You can scan the 2GB and 4GB columns and see the results there. Clearly, a second processor delivers the most benefit at higher RAM configurations, and it can actually increase the rendering time for some formats in lower-memory configurations.

Considering both price and performance, what system configuration is the best choice for each project format? To determine this, I ranked the performance of each test configuration from fastest to slowest and calculated the processor- and RAM-related cost differential from the best-performing system. This lets you gauge the performance decrease and cost savings together to find the optimal system for your needs.

Table 5. Comparing system costs vs. performance for DV projects (ordered by rendering time).

In all instances, the dual-processor, 16GB system was fastest. However, with the DV test project, the second fastest system was a single-processor configuration with 4GB of memory, which was only 1 percent slower but cost 86 percent less than the fastest system. No doubt in my mind which configuration I would start with. Just to be clear, the total cost shown is the additional cost of the CPU and RAM for the configurations shown, not the total cost of the system.

Table 6. Comparing system costs vs. performance for HDV projects (ordered by rendering time).

With the HDV test project, the SP-8 (single processor, 8GB) configuration jumps out at me, offering a savings of 83 percent from the best-performing system with only a 25 percent decrease in performance. If I were producing HDV on a budget, that''s where I would start.

Table 7. Comparing system costs vs. performance for AVCHD projects (ordered by rendering time).

The same system (SP-8) offers a good blend of affordability and performance with AVCHD source footage, with an 83 percent price reduction and 24 percent decrease in rendering time.

Table 8. Comparing system costs vs. performance for DVCPRO HD projects (ordered by rendering time).

With DVCPRO HD, the dual-processor systems dominate from 16GB down to 4GB, with the 8GB system offering nearly the same performance as the fastest system with a 22 percent reduction in cost.

Table 9. Comparing system costs vs. performance for Red projects (ordered by rendering time).

With the 4K Red video format, memory is obviously much more important dual processors. If I were on a budget, I''d zero in on the SP-16 system.

OK, I think we''ve beaten the xw8600 Windows-related test conclusions to death (or at least a stupor), so let''s move on to the Mac tests—which, thankfully, are much less involved.

Table 10. Rendering with a Mac Pro.

As mentioned in the first installment, I used different tests with the Mac and Windows workstations because the systems were configured with different processors and I didn''t want the results compared. More specifically, with the Windows test projects, I added After Effects components to the Premiere Pro timeline and rendered via Adobe Media Encoder. On the Mac, I substituted native color key overlay effects for the After Effects chroma key, sent the sequence off to Encore via Dynamic Link, and rendered a DVD or Blu-ray folder to hard disk, depending upon the source footage.

I performed all tests in the dual-processor configuration because I didn''t have the guts to perform the major surgery necessary to access and remove one of the two processors that shipped with the system. Since the system shipped with 8GB of RAM, this was my maximum configuration. I also tested in 2GB and 4GB configurations, which are the only other configurations offered by Apple. All render times are in seconds.

As you can see, both the HDV and Red formats tanked in the transition from 8GB to 4GB, indicating that you should stick with 8GB for both of these formats. Interestingly, at least from a pure rendering standpoint, the other three formats remained fairly peppy all the way down to 2GB.

Table 11. Comparing RAM performance with a Mac Pro.

How responsive was the system while rendering in these memory configurations? Again, to test this, I rendered a 15-second portion of the Premiere Pro timeline containing a native color key effect while rendering the DVCPRO HD project to Blu-ray-compatible MPEG-2.

As you can see, in the 2GB configuration, rendering took more than 7 minutes, which dropped to just less than 4 minutes at 4GB and 8GB configurations. This tells me that if you plan to use the system for other activities while rendering, you should buy at least 4GB of RAM.

How to use this data given that your projects will differ widely from my test projects? Two platitudes come to mind, which I''ll use as a substitute for original thought. First is that your mileage will certainly vary. The suggested configurations could prove more than adequate, or they could bog down like a city car in the country mud.

However, (here comes the second) in the land of the blind, the one-eyed man is king. Any information is better than none at all. In your shoes, I would start at the suggested configurations—remembering that you can always add RAM and a second processor, but you can seldom give it back if you end up buying more than you need.