ZoomISO v2 Benchmarks

ZoomISO v2 is optimized for high performance on Apple Silicon. To demonstrate these enhancements, and to provide context for how ZoomISO may perform in your workflow, this page presents a variety of test results and graphs.

The following data will be continuously updated as new results are collected, and should not be interpreted as an endorsement, certification, performance claim, or official recommendation of any particular product or configuration, including ZoomISO. The information presented on this page is not a scientific, peer-reviewed study, but rather an effort to provide real-world testing data to prospective customers and system designers.

Testing Configurations

ZoomISO v2.0.5 was benchmarked on first-generation Apple Silicon processors (the "M1" product-line). 
 

Devices used:

  • M1 - Mac Mini (base model)

  • M1 Pro - MBP 16” ‘21

  • M1 Max - Mac Studio (base model)

  • M1 Ultra - Mac Studio (top model)

 

Testing was conducted in a Zoom meeting with High Bandwidth Mode (HBM) enabled in a breakout room via the Breakout 100 add-on. The 1080p add-on was also enabled. All outputs were tested at 1080p 29.97. 16 PCs played back unique 1080p participant placeholder video loops with virtual mic audio into the testing breakout room.

 

SDI testing was accomplished using 2x Blackmagic DeckLink Quad 2 cards in Sonnet Echo Express SEI enclosures (unless otherwise specified). NDI was decoded on an external PC using the NDI Studio Monitor tool. Syphon was decoded using the Simple Syphon Client.

 

Bandwidth use was reported by Meraki network tools. Bandwidth tests may represent some interference as it was not possible during the test to observe network use exclusively to the Zoom Cloud from specifically the ZoomISO application. The tools represented the WAN downlink, and the starting value for each test was subtracted from subsequent figures.

Testing concluded when 16 outputs were reached, or when the CPU reached 85%, whichever came first, unless otherwise specificed. CPU usage was measured using the ZoomISO Zoom Client's Statistics page for Overall CPU use.  At 85% CPU usage, Zoom may drop the resolution of the incoming video, which would distort the testing results. 

Regarding Apple Silicon's interactions with Thunderbolt, all devices tested except for the M1 Mac Mini utilize at least two dedicated hardware controllers for Thunderbolt. Sonnet Enclosures were plugged into ports backed by dedicated controllers to avoid rapid performance degradation due to bandwidth limitations imposed by sharing a controller. The M1 Mac Mini has a single controller for its two Thunderbolt ports. As a result, the M1 Mac Mini has been deemed incapable of producing more than 8 outputs to SDI, so only one Sonnet Enclosure was attached, limiting its export quantity to 8 vs 16. While the reduced initial load decreases all CPU load when compared to other chips, had the other chips only used 8 outputs, they would likely perform better than the reported M1 results. 

 

The Syphon protocol requires that the frame encoder and decoder run on the same computer. As a result, the reported numbers for Syphon represent both ZoomISO’s load plus Simple Syphon Client’s decode in order to generate an actual video chain load on the system. SDI and NDI outputs were decoded by external hardware because they support that configuration and are commonly used in that way.

By creating all outputs first, and then assigning participants to the outputs, the actual load on the system represents the impact of decoding additional incoming video feeds. If the outputs were added one-at-a-time, the CPU impact would represent the combination of initializing an output and decoding the Zoom video feeds. As a result, the decision was made to create outputs first, and then assign participants to the outputs with each test pass. This method represents an over-estimate of CPU impact, as most users do not activate more outputs than they plan to use in a given workflow.

Testing Procedure

  1. Join breakout, enable engine, create output quantity, assign to device

  2. Activate decoder hardware or software

  3. Disable all outputs

  4. Assign participants to outputs

  5. Set Gallery View to 12 participants by shrinking window if necessary

  6. Record “starting load” for CPU and network

  7. Activate an output

  8. Record CPU and network load

  9. Repeat for all outputs

  10. Stop test when 16 outputs are achieved or when 85% CPU use is reached, whichever comes first

Results

Untitled presentation (5).png
Untitled presentation (6).png
Untitled presentation (8).png
Untitled presentation (4).png