Benchmarking the HP Z840 Workstation for Video, Part 3: Analysis

This is the third segment of my review of the HP Z840; in Part 1, I analyzed performance during editing and rendering; in Part 2, performance during encoding. This time out, I analyze performance for video analysis, specifically, performance using the Moscow University Video Quality Measurement Test (VQMT) and a file conversion necessary to run the VQMT.

Why generalize one test to an entire category? Here’s my thinking. One of the key benefits of the Z840 is its Turbo SSD G2 drives, which are more than 20 times faster than traditional hard disk drives (HDD) in some operations (see Table 1 in Part 1).

There are myriad applications that involve the analysis of data contained in exceptionally large data sets, and I wanted to learn how the Z840’s fast drives, combined with the system’s 40 CPU cores, could work through that data.

The VQMT is a video quality measurement tool that compares a source and encoded file and produces a number of objective metrics, including Peak Signal-to-Noise Ratio (PSNR), the Structured Similarity Index (SSIM), and the Video Quality Metric (VQM). In order to apply the test, you first have to convert the compressed files into YUV format, which I perform in FFMPEG.

I tested with 96-second 4K video files which are 33 GB in size in YUV format. During the file analysis operation, VQMT retrieves a frame, compares it to the original, and calculates the objective metric.

You can see the graphics output from the tool in Figure 1 (below); each point on the graph shows the VQM score for that individual frame.

Figure 1. The visual output from the VQMT. Click the image to see it at full size.

I used the VQM metric in my comparisons because it’s the most compute-intensive of the three mentioned, and also the most useful. It’s also a task I performed hundreds of times last summer for a consulting project, so I was curious to see how the Z840 could have chewed through this data. As you’ll see, it did not disappoint.

In case you missed the initial installment, I compared the 40-core Z840 with a circa 2010 24-core Z800. HP sent the Z840 with multiple types of drives, including two HP Turbo SSD G2 drives, one for boot/program, one for data, a traditional SATA SSD drive for data, and a traditional HDD drive for data. The Z800 has three HDD drives, including a 15K boot/program drive, and I installed an older SSD drive on the Z800 specifically for this test. I ran the VQMT via its command-line tool, so the speed of the boot disk played only a minor role, if any, in the overall results.

As before, I look at a multiple issues:

• First, how much faster was the Z840 than my older Z800?
• Second, was performance faster with hyper threaded technology (HTT) enabled or disabled?
• Third, did the disk speed on the respective test systems significantly impact encoding performance?