HP 12-core Z800 Test Drive, Part 1
Mar 24, 2010 12:00 PM, By Jan Ozer
I still do some shoots in DVtypically freebies for local artists or when I know the project is going straight to YouTube or SD DVD.
Loose Strings is a local kids' band that was auditioning for America's Got Talent. The project was a 7:28 single stream of 16:9 DV, which I rendered to H.264 using Adobe Media Encoder's YouTube Widescreen SD template. In this test, the 12-core computer was 47 percent faster than the older 8-core unit and 217 percent faster than the xw8600.
The next project was a 7:20 video that I edited for a local hunting guide and rendered to MPEG-2 for an SD DVD and also to H.264 for uploading to YouTube. Though primarily DV source, there was also some H.264 source and this was where the older Z800 started to go wiggy, particularly when rendering to MPEG-2, producing a result that was actually slower than the older xw8600. Both Nehalem-based systems were substantially faster than the xw8600 when rendering to H.264.
Since editing isn't all about rendering, I threw in some "preview" comparisons as well, here rendering the last 40 seconds of closing credits on the hunting video. As you can see on the table, the 12-core Z800 rendered the preview in 9.2 seconds; the xw8600 was well behind in 25 seconds.
I do most of my work in HDV, so I had some longer, multicamera projects to test here. First was a 4:35 (min:sec) single-camera HDV shoot output to H.264 for upload to YouTube. As you can see in Table 2, the xw8600 took 10:03 to produce the file, the eight-core Z800 took 6:14, and the 12-core system, 4:54.
Next was an hour-and-40-minute two-camera HDV shoot mixed down to SD DVD. Here the eight-core system took 23 percent longer than the 12-core, with the older-style Xeon taking 213 percent longer. Here, the 12-core Z800 was actually three times faster than the xw8600, boding well for HDV producers who are considering buying the new box.
The final project was rendering a 5-minute song from a three-camera shoot to Blu-ray-compatible H.264 format. The source files had an interesting heritage: two plain HDV, the other AVCHD (H.264). Originally, I had produced the project on a Mac and converted the AVCHD source to ProRes 422 in Apple Final Cut Pro during ingest, because I prefer to work with ProRes over native AVCHD.
ProRes worked fine on my XP and Vista systems, but not on my Windows 7 system, so I had to convert it back to 720p H.264. This punked out the eight-core system, which simply stopped encoding during long periods of time during this render. Though the 12-core system still saved 47 percent over the older-style Xeon, I was beginning to think that there was a bottleneck in how Adobe handled H.264 source formats. We'll put that directly to the test with AVCHD projects in a minute.
To test preview speed in HDV, I added a dynamic slow-motion effect to a 20-second segment of a clip, first slowing it down to 50 percent speed over the course of 2 seconds, then speeding back up to 100 percent over the same duration. I saw very little difference in preview time here, saving an insignificant second between the 12-core system and the older-style 8-core.
Though it's tough to understand what's going on under the hood, the similarity in preview time likely means that the speed-change function in Premiere simply isn't multithreaded. I did check processor use with Windows Task Manager during preview, and it was very low on both machines.
Now on to AVCHD.
Continue the discussion on “Crosstalk” the Millimeter Forum.