Benchmarking the HP Z840 Workstation for Video, Part 2: Encoding

This is the second segment of my review of the HP Z840; in the first segment, I compared performance during editing and rendering. This time out, I analyze performance during encoding.

As before, I tried to answer multiple questions:

• First, how much faster is the Z840 compared to my older Z800 workstation in multiple encoding chores?
• Second, was performance faster with HTT enabled or disabled?
• Third, did the disk speed on the respective test systems significantly impact encoding performance. More specifically, does using SSD disks improve encoding performance over older HDD drives?

There was a wrinkle in this testing: Telestream Vantage, which like all enterprise class encoders, is best run on a clean system, wouldn’t install on the Z800, upon which I’ve installed nearly every encoding and video analysis program known to man. No surprise there. So I couldn’t test Vantage on that system.

However, Vantage ran fine on a dual-CPU HP Z600 I have in house, though it’s not nearly as powerful as the Z840 or Z800 in clock speed or number of CPU cores. Like the Z800, it’s equipped with traditional HDD drives.

Table 1 (below) summarizes the key components of each system. There are many more details that impact performance, including bus and memory speeds, but these are the key differences.

Table 1. Main components of the test systems.

Test Description

Let’s start with a brief description of my tests that I ran.

• Adobe Media Encoder (AME): Rendering time to output a 96-second 4K test file to five Apple iDevice presets. AME was installed on the boot drive and all files were saved to the C drive, with source files coming from the respective drives on the Z840.
• Squeeze: Rendering time to output a 5:13 (min:sec) 1080p AVCHD file to Squeeze’s multiple-bitrate preset, which converts the source file into four streams and formats them for delivery via HTTP Dynamic Streaming (HDS), Apple’s HTTP Live Streaming (HLS), Microsoft Smooth Streaming (MSS), MPEG-DASH, and MP4 files, converting the single input file to 20 outputs. Squeeze was installed on the boot drive and all files were saved to the C drive, with source files coming from the respective drives on the Z840.
• x265: Rendering time to simultaneously encode three 5-second 1080p source files to HEVC format. For this test, I created three folders on each source disk and ran x265 via command line with files saved to the source disk.
• VP9: Rendering time to simultaneously encode three 5-second 1080p source files to VP9 format. For this test, I created three folders on the test disk and ran VP9 via command line with files saved to the source disk.
• Vantage: Rendering time to output the 96-second 4K source file to six output streams at various configurations. Vantage was installed on the boot drive and all files were saved to the C drive, with source files coming from the respective drives on the Z840.

As before, I tested with HTT enabled and disabled, and rendered from all drives on the Z840. Table 2 (below) shows the overall fastest times for each computer, with the percentage decrease delivered by the Z840 on the right, in green when the difference exceeds 35%. Note that I used completely different encoding parameters for Squeeze, AME, and Vantage, as well as different source files for Squeeze, so the encoding times are not comparable.

Table 2. Total performance Z840 to the Z800 (Z840 to Z600 for Vantage). All times in seconds.

Vantage showed the most improvement, not surprising given that its results come from the slower Z600, but impressive nonetheless. The only encoder that didn’t cross the 35% threshold was WebM encoding from the command line; a brief look at Figure 1 (below) shows why. Specifically, Figure 1 is a frame grab from Windows Performance Monitor showing CPU utilization while encoding three streams of WebM and HEVC on the Z840. As you can see, even when producing three streams of WebM, utilization is well under 20%, compared to 60%-90% for HEVC.

Figure 1. CPU utilization while encoding three streams of WebM and HEVC.

Just for fun, I ran a quick test encoding seven WebM streams from the Turbo SSD G2 drive. Overall encoding time increased by only 4 seconds, with CPU utilization hovering around 19%. Given the lack of CPU headroom when encoding only three streams of HEVC, it’s doubtful that I could have encoded even a single additional stream without noticeably slowing performance. So though x265 had a faster time in Table 2, you should be able to produce many more simultaneous WebM streams on a single computer than HEVC, making WebM the faster encoder.

Overall, for most projects, a new Z840 should be able to deliver significant performance boosts over existing stations.