KB265775275: Imaging Engine - Which server gives best performance when ripping

Question

Which Imaging Engine Server gives best performance when ripping?

Although we have the system requirements for Imaging Engine, these are 'minimum' requirements. In this article, we will try to give some additional advise, but do understand that the ripping speed is influenced by many factors such as type of jobs and certain usage of settings (Image Resampling, Dot Cleanup, etc.)

Computer Specifications - What is Your Goal?

To have a good rip performance, customers are willing to have a custom made dedicated box for their Imaging Engine. In specifying the parameters for such a box, the first step is to choose your goal. Specifically, decide if your goal is to maximize peak performance (the time to RIP a single job) or total throughput (the total number of jobs that can be ripped over a period of time).

A dedicated box - A dedicated physical system (with desktop-class CPU like a Core i9 and SSD drives) may have higher peak performance, but is limited in number of workers and therefore may have lower total throughput over a period of time (such as a shift).

A virtual setup - A server (typically with Xeon-class CPU) will allow more workers on the same box, and therefore higher throughput over time, but likely at a lower peak performance per job.

The Optimum Number of Workers

An Imaging Engine worker processes files in sequential order. Multiple workers on a system compete for CPU, memory, and disk resources, which can slow individual job ripping times if multiple files are being ripped at the same time. If your goal is to maximize peak performance of a single job, you may choose to run a single worker. If your goal is to maximize throughput, you will want to increase the number of workers. This is configured in the Automation Engine Pilot as shown below (in the Configure panel).

The number of workers are defined by license and by the number of cores. Licenses for additional workers can be purchased from Esko.

Recommended Computing Specifications

The most important factors related to processing speed are (in order):

1. CPU Clock Speed
2. Number of cores (logical cores = same as number of physical cores when hyperthreading is disabled)
3. Disk Speed
4. RAM

CPU clock speed

The base clock speed is most important: 4 GHz is recommended (higher is better).

Especially important when using some features like 'Dot Cleanup'. See topic below.

Number of ‘cores’

Number of ‘cores’: at least 4 for the OS, and on top of that at least 4 per worker (more is better!).

Two processors are widely available to configure a box for your specific needs:

1. Intel Core i class
2. Intel Xeon class

The Intel Core i7 and i9 CPU’s are typically used in desktop and laptops PCs. The Intel Xeon is a server-class CPU. For a RIP, high base clock speed is more important than high number of cores. As such, in RIP applications, the i7 and i9 may actually provide higher performance than the Xeon. Again, it’s a trade off between peak performance and total throughput time.

The Intel Core (i7, i9) series have the highest base clock speeds (up to 4.2 GHz) but limited amount of “cores” (up to 8 for standard versions, up to 18 for “Extreme” versions).

The Intel Xeon have lower base clock speeds (up to 4.1 GHz for a 4-core version), and can have up to 56 cores (but then only have a 2.60 GHz base clock speed).

Disk Speed

• Solid State Drives (SSD) are much faster than HDD (“spinning”) drives.
  
  • Use SSD only if possible.
  • If you must use an HDD make sure it gets defragmented regularly.
• Imaging Engine takes 20 GB of disk space. In addition, provide at least 100 GB of “working disk space” per worker.
• Do not install Imaging Engine on the system drive (C Drive). Keep it on a separate drive where filling a disc with temp files will cause the least amount of harm.

Memory (RAM)

• At least 8 GB RAM for the OS, and on top of that at least 8 GB extra per worker (more is better!).
• Most jobs we see run easily in 8 GB, but we’ve seen exceptions that require up to 32 GB per job.

Less Important Factors

Hyperthreading: This doubles the amount of ‘cores’ available (e.g. from 8 “physical” cores to 16 “logical” cores).

• The extra “logical cores” are not as powerful as the real “physical cores.”
• Enabling hyperthreading does not allow you to double the amount of workers you can run.
• Enabling hyperthreading and keeping the same amount of workers will typically give a speed-up (more cores available per worker, if needed). Test this with your files to check.

Network: faster is better

• 1000 Base-T Ethernet is recommended. But 100 Base-T Ethernet is also supported.
• If the output files need to be on another system, RIP them to a local disk, then Copy/Move to the final destination (large files transfer faster than small chunks). You can use the Copy / Move task in the Automation Engine to automate the file copy to its final destination.

Other Factors influencing ripping performance

Some output options

• Resolution
  the higher the resolution the slower ripping
• 'Save print simulations (.view files)'
• 'Make all output options available for viewing'
  this slows down a little bit ripping, but do know that this option is more optimal than using the separate task 'Prepare for viewing'. Both options make sure you can view your output files in the Automation Engine Viewer

• Compression
  You should choose a compression method based on the compression speed and the compressed file size. Typically, faster compression methods compress the file less, so the output file is bigger. See the User Guide for more details. The Compression method offered by default for the output File Type you select (LZW for TIFF, PackBits for PDF and Flate for LEN or LENX) is generally the best for this file type. Do know that for LENX, the Flate compression is always used (even when you select 'none').

Image Resampling

A setting in the output menu of an Imaging Engine Ticket called “Image Resampling” has a dramatic effect on halftone dot quality. Prior to Imaging Engine 18.1.1, this setting caused intermittent issues in the screening process (out of memory problems). With Imaging Engine 18.1.1 the setting has been optimized. Still using this option will result in longer RIP times. You can choose between 'Auto Resampling' and 'Full Resampling', the last one gives the best quality but is slower.

Dot Cleanup (in highlights)

This option is license based (but included when having the Print Control Wizard).

When a job is RIP'ed with screening, it can happen that very small or broken dots (sometimes called "scum dots") appear in highlight areas. You can use Post-RIP settings to clean up areas of the output files but this will slow down ripping performance. We noticed it's important to have a high clock speed, so if you're using this option a lot (e.g. with Crystal Screens), it's advised to have a minimum clock speed of 3 GHz, if possible even 4 GHz or more.

Type of Screening

The type of Screening can influence the performance too. Typically advanced screenings (HD Flexo, Crystal Screening, ..) will slow down ripping.
First of all because they need more memory and result in larger output files. Secondly because the advanced effects make quality control harder, thus typically requiring an extra rip step to generate "print simulation files" (with .view extension)

Step & Repeat

Use Step & Repeat software that works from a single copy of each image and does not make a copy of the image for each step. Two examples are Plato and ArtPro+. The result of resampling a single image (versus each step) is much faster RIP speeds compared to applications such as Adobe Illustrator. You can check this by the file size of the stepped file. For example, for an 8 up file, if the stepped file is 8x the size of the one-up file, then optimal stepping is not being used and RIP times will be much longer than necessary.

Example

Hardware used (Dell Precision 7820):

• CPU : Intel Xeon Gold 6136 CPU @ 3.00 GHz
• 24 logical processors
• RAM : 32GB

Most important ticket settings:

• Output : LEN/LENX @ 4000 ppi
• Save print simulations (.view)  enabled
• Image Resampling : Auto Resampling
• Dot : CRS01 -> WSIPEMOD TP3 SUP3
• Dot Cleanup  enabled

Timings:

• Sequential execution (1 worker process)
  Job 1 : 25:13 minutes
  Job 2 : 36:42 minutes
  Job 3 : 26:19 minutes

• Parallel execution using 3 worker processes.
  Average time for the three jobs : 24:30 minutes