Benchmarking PCs, Capture Cards, and Software Mixers for HD Capture and Streaming

To determine how powerful a system it takes to produce and deliver live event streams, we tested a range of computers (old and new), capture devices, and multiple live streaming software programs, streaming to one or more streaming services, recording archive/ISO files in various formats, and measuring CPU utilization. Read on for the results.

When you’re producing a live event with a self-built, computer-based system, your focus narrows to a single relevant question: Is the system powerful enough to produce the streams required for the event?

To help find out, we assembled a range of computers (old and new), obtained several capture devices, and downloaded multiple live streaming software programs. Then we ran multiple tests, streaming to one or more streaming services, recording archive/ISO files in various formats, and measuring CPU utilization. On the Windows-based systems, we captured the results in Performance Manager and annotated the output to produce graphics such as those shown in Figure 1. On the Mac, we simply recorded CPU utilization numbers during the tests.

Before discussing the results, let’s touch on testing procedure. First, to normalize the results, I used the same high-motion, high-detail AVCHD clip for all tests. Each time, I started the streaming and/or recording activity, triggered the recorded video, let it play through, and then stopped recording/streaming. Then I changed the configuration and started again. To add a touch of complexity, I added a logo overlay to all videos.

I’ll detail each computer system and capture device used when reviewing the test results. For the mixing software, I used a prerelease version of Telestream Wirecast 7.1. I was glad I did, since CPU utilization was down significantly from version 7.0, which was way down from version 6.0. I tested vMix version 17.0.0.107 and version 0.15.4 of Open Broadcaster Software (OBS).

Let’s start with notebook testing.

Notebook Testing

I ran the first tests on an HP ZBook Studio G3 Mobile Workstation with a 2.8GHz Xeon quad-core (eight with Hyper-Threading Technology [HTT]) E3-1505M CPU with 32GB of RAM running Windows 10. The first capture device was a USB 3-based Epiphan AV.io 4K video grabber.

Figure 1 (below) details the results of tests run with Wirecast version 7.1, though tests varied slightly with the capabilities of the software mixer. I started with a 720p/2.25Mbps stream to Facebook Live, and then added a 1080p/8Mbps archive encoded with x264, which boosted CPU utilization to more than 55%. By way of comparison, with Wirecast version 7.0.1, CPU utilization for x264 was more than 70%. Test three was Facebook Live plus an archive encoded with the more efficient Intel Quick Sync Video codec.

Figure 1. CPU utilization for Telestream Wirecast on a 2.8GHz HP ZBook with an Epiphan AV.io 4K video grabber

As a rule of thumb, once CPU utilization pushes past 70%, you may start dropping frames in your live stream or recorded file. Though x264 exhibited slightly higher quality than Quick Sync, at 8Mbps the difference is very minor. If you notice that CPU utilization is consistently exceeding 60% or so, consider switching to Quick Sync (if available) for your archive version.

Note that I used x264 when available for all live streams to Facebook and Ustream. If you find CPU utilization uncomfortably high on your system, consider switching to Quick Sync (if available). I used x264 with the Very Fast preset since this was the default for all tested software programs. I ran some tests with the Medium preset, but this drove CPU utilization through the roof, and the quality difference was indiscernible.

Wirecast also offers an ISO archive, which produces a Motion JPEG (MJPEG)-based QuickTime file. While about as efficient CPU-wise as the Quick Sync codec, the Motion JPEG file is more than 11 times larger (91.7Mbps as compared with 8Mbps), which could put a strain on the storage capabilities of your notebook. As you’ll see later, I also found QuickTime MJPEG files slightly dingy, so I recommend using MP4 over MJPEG for the quality and file efficiencies. The final test involved sending 720p streams to two live streaming services, Facebook and Ustream, which required about as much CPU as Quick Sync and the ISO Archive.

What are the key takeaways? At 2.8GHz, an E3-1505M Xeon system, or a 2.8GHz i7 system, should have enough power to easily handle a 720p output stream, plus an archive version in x264 or Quick Sync formats. Slower versions of the same CPUs should deliver close to proportional performance, though the slowest CPU offered on ZBook Studio is a 2.6GHz i7, which should perform similarly to the tested model. Changing the CPU completely, say to a dual-core i5 (four with HTT), changes the formula completely, and I briefly look at an i5-based system later.

The Software Factor

How does software change the equation? Have a look at Figure 2 (below), which shows vMix Pro running a similar suite of tests. As you can see, vMix was more efficient than Wirecast across the board, though the advantage dropped from 15–20 percentage points with 7.0.0 to 5–10 percentage points with 7.1. As mentioned previously, the vMix tests are slightly different due to the different archival formats offered by the two programs, including a high-quality but high-data rate YUV format. If you’re running on a slower notebook and CPU utilization is running high, vMix is definitely a viable option. Another option is the free OBS, which is fine for a simple project such as this. It consumed about as much CPU as vMix in live streaming and archiving trials.

Figure 2. CPU utilization for vMix Pro on a 2.8GHz HP ZBook connecting via the Epiphan AV.io 4K video grabber. Click the image to see it at full size.

The other takeaway from working with a USB-based capture system is the need to manage your ports carefully. The ZBook comes with three USB ports, and if you add a portable USB 3.0 drive to store the archive version, placing it in the wrong port can flood the connection. Overall, you’re best off buying a system with sufficient internal storage for a completed event to avoid contention issues.

For those interested in 1080p streaming, I tested streaming to Facebook Live at [email protected] using Wirecast 7.1. Streaming-only consumed about 25% of total CPU, up from about 18%. With vMix, the bump was similar, from about 15% to 25%.

I ran a similar suite of tests on a MacBook Pro powered by a 2.8GHz Intel Core i7 CPU with Wirecast 7.0.1 and OBS (vMix doesn’t have a Mac-based mixer), but I had to return the MacBook Pro before I received the Wirecast 7.1 upgrade. Mac performance was similar to that achieved on the ZBook in Windows, and I would expect the Wirecast performance improvements found in Windows to extend to the Mac.

On the other side of the coin, I tried duplicating these tests on a Dell Inspiron notebook powered by an i5-4210U CPU, with a base frequency of 1.7GHz and a top speed of 2.4GHz. Note that the i5 has two cores (four with HTT), rather than four on the i7 (eight with HTT), so in addition to the slower speed, there are fewer cores. On the Dell, streaming only at 720p flatlined the CPU at 100%, and even dropping resolution to 360p still resulted in peaks well over 90%. Though it might be feasible to stream from an i5 with a faster clock speed, if you’re buying new, go for an i7 or four-core (eight with HTT) Xeon-based system.

Mixing It Up With the AJA Io 4K

Epiphan’s AV.io 4K is an affordable single-stream product, but what about multiple-camera productions on a notebook? While you can hang multiple single-stream capture devices from your notebook, you’ll get the most efficient results with a product like AJA’s Io 4K, which supports up to four 1080p inputs and connects via Thunderbolt 2.

To test the Io 4K, I used the Osprey SDARD-4 Distribution Amplifier, which inputs a single HD-SDI stream and outputs four, connected to the Io 4K. I used the same AVCHD source video, converting that to HD-SDI with a Blackmagic Design Mini Converter HDMI to SDI. The procedure was the same, except during playback. I cycled through the four inputs in Wirecast and vMix every 5 seconds or so. As shown in Figure 3 (below), the results were pretty phenomenal, with vMix supporting live four-camera mixing and x264 archiving while consistently below 60% CPU utilization. Even compared with the new Wirecast version 7.1, vMix proved about 10% more efficient than Wirecast in all operations. Again, performance on the MacBook Pro matched that in Windows with Wirecast version 7.0.1.

Figure 3. CPU utilization for vMix on a 2.8GHz HP ZBook connecting via the AJA Io 4K video grabber

These results seemed almost unrealistically positive, so I hunted around to see if I could find confirming results. I found one producer on the Telestream boards who reported 80%-plus CPU utilization on a 3GHz i7-based 13″ MacBook Pro Retina, though actual details about the production are limited and the producer was very positive (“CPU was 80%+ but that was with 4 cameras in one shot. This is a great solution …”). There’s also a YouTube video available at bit.ly/2bDcZ6P that shows the MacBook Pro/AJA setup working, though CPU utilization details are lacking.

I asked AJA about minimum system requirements for the Io, and the company deferred to the requirements of the software program. I checked the vMix website, which recommends a 3GHz+ i7 with 8GB of RAM, while Telestream recommends a similar configuration for 1080p streams. My tests on the 2.8GHz systems indicated that there was little CPU headroom, and I wouldn’t recommend producing on anything slower than our tested systems.

On the other hand, I’m sold on Thunderbolt-based mobile productions for systems that support it. Note that the ZBook offers two Thunderbolt 3 ports, but no Thunderbolt 2 ports. The price for this future-proofing is that you’ll need a Thunderbolt 3-to-Thunderbolt 2 dongle. I used the StarTech Thunderbolt 3 to Thunderbolt Adapter, which costs about $75. While the dongle worked fine with the AC-powered AJA unit, it did not work with my Thunderbolt 2 external SSD drive, though the specs say it does pass power through to bus-powered devices.

Desktop Testing

Though notebooks have obvious appeal for mobile producers, desktops are cheaper, particularly if you already have them around. They’re as convenient as a notebook if you produce from a single location. The first set of tests I ran was to determine if a circa-2011 HP Z600 workstation with two 2.66GHz Intel Xeon CPUs, each with four cores (eight with HTT), can serve as an adequate platform for live event production. I tested with two cards from Osprey Video, the dual HDMI input Osprey 821e and the Quad HD-SDI (or dual input 3G SDI or DVB-ASI) Osprey 845e. I tested with Wirecast and vMix, but not OBS, assuming that a multiple-camera event producer wouldn’t use shareware.

It seems that every time you install a capture card, you learn a new lesson, and so it was with the Osprey 821e. Specifically, despite being a four-lane PCI Express (PCIe x4) card that could physically fit into one of the four-lane PCIe ports on the Z600 motherboard, the 821e needed the bandwidth available only in a PCIe x16 slot (Figure 4, below). I discovered this after installing the board into an x4 slot and spying intermittent color bars in the mixer’s preview window. I called the always-helpful Osprey tech support line, and the first question the tech asked was in which slot I installed the card.

Figure 4. Even though the Osprey 821e could fit into a PCIe x4 slot, it needed the bandwidth available from a PCI x16 slot. Click the image to see it at full size.

When I told him it was the PCIe x4 slot, he suggested that I install it into the PCIe x16 slot instead, which resolved the problem. This brings up two key points. First, if you’re attempting to convert an existing workstation into a live capture station, make sure it has a free PCIe x16 slot, particularly for a video capture card with two or more inputs. Second, as you’ll see with the four-port Osprey 845e card, a live capture station needs both a fast CPU and a speedy bus.

Dual HDMI Inputs (Osprey 821e)

Once I had the system up and running, performance with dual HDMI capture proved more than adequate, particularly with the vMix software. As you can see in Figure 5 (below), the system easily processed a 720p live stream plus an MP4 archive, plus all the other tests we threw at it. With Wirecast, the results were about 10 percentage points higher in all tests save x264, where CPU utilization pushed to about 50%. Quick Sync is not an option for the older Xeons on the Z600.

Figure 5. CPU utilization for vMix on an HP Z600 with two 2.66GHz quad-core Xeons. Click the image to see it at full size.

Unfortunately, the system peaked with the dual HDMI input card and proved inadequate for the quad HD-SDI 845e card. Interestingly, it wasn’t CPU utilization that disqualified the unit—it was bus throughput. I ran the same tests with the 845e as I did with the 821e, and CPU utilization for both programs was very similar. However, periodically during the production, color bars would replace the video in the preview and output windows, indicating that all video data was not getting through.

While the CPU power may be sufficient on these older systems, you’ll need the bus speeds available on newer systems to support four-channel input cards. Would two dual-input HD-SDI cards have worked? Interesting question: That’s definitely an option I would explore if I had to make the Z600 work.

Matrox VS4 and an i7-Based System

Over the last few generations of processors and motherboard chipsets, bus and memory throughput has improved along with the number and speed of CPU cores. This takes us to the last system that we tested: a custom-built i7-based system supplied by Matrox with its VS4 Quad HD capture card. Specifically, the computer was powered by a 3.4GHz i7-4770 CPU with 16GB of RAM and an Nvidia Quadro K620 graphics card running Windows 7 with 16GB RAM. The chipset was an Intel Z87, which falls into the Haswell family.

The VS4 is a quad HD-SDI capture device with hardware ISO recording capabilities that makes it efficient to capture ISO streams from each of your four source devices. In this series of tests, shown for vMix in Figure 6 (below), both software programs performed well, though Wirecast used about 5 percentage points more CPU than vMix in most tests.

Figure 6. CPU utilization for vMix on a generic i7-based system and the Matrox VS4

As you can see in Figure 6, with the i7-based system, vMix was able to stream to Facebook Live and capture four ISO streams in MPEG-2 I-frame format while staying well below 60%. Adding a live stream to Ustream boosted CPU to almost 70%, though ISO files captured by both programs were free of dropped frames.

The other point of Matrox’s hardware-based ISO recording is that it produced better quality than the equivalent MJPEG-based QuickTime output. You can see this in Figure 7 (below), which shows Matrox on the left and QuickTime footage on the right. You can certainly correct some of the dinginess on the right with brightness and contrast adjustments, but the Matrox frame is much sharper as well. If ISO recording is important to you, you should strongly consider buying a board with hardware-based ISO output.

Figure 7. Matrox’s hardware-based ISO output was superior to Motion JPEG-based QuickTime.

Summing Up

You may be wondering, “Why didn’t you try the quad-input Osprey 845e card on the i7-based system?” Well, I did, but the system didn’t recognize the card. This happens sometimes, according to Osprey’s tech support staff, which is a good intro into our final conclusions.

First, not all capture cards work with all systems. If you’re buying a capture device for an existing computer or notebook, make sure it’s returnable because there’s a good chance it won’t work. If you’re buying a new computer to convert to a live switching station, you should strongly consider a prepackaged system. Telestream just entered this market with multiple systems under the Wirecast Gear label, while vMix has offered turnkey systems for a while. You can also buy from third parties such as 1 Beyond or Paladin, and Osprey Video also offers base systems you can buy for guaranteed compatibility, performance, and stability.

Along the same lines, recognize that not all capture cards work equally well with all software video mixers. All capture products have a generic DirectShow interface (or the Mac equivalent) that software vendors can use to work with the device. However, they also usually offer software development kits (SDKs) for improved performance and more stable operation. It’s hard to tell which level of support the software video mixer affords each capture device, but it’s a safe bet that if the vendor offers an integrated system, the software mixer supports the capture card’s SDK directly. This is another reason to buy a turnkey system.

If you’re looking for a notebook-based system, you’ve seen that the different software programs consume varying levels of CPU. You’re best off experimenting with multiple programs to find the best mix of features and performance. This is also true if you’re building your own desktop system from an existing workstation or computer. Beyond producing and watching live streams, be sure to check your archived streams for drop frames.

Finally, if you’re building your own system from a notebook or desktop, note that some problems don’t appear until 20–30 minutes into a production, or even longer, so you should test the system for the expected duration of your longest planned live production. Remember to test with footage similar to the actual live event, since a system that performs well when encoding a static scene may fail when encoding the actual soccer game it was designed to produce.

About Jan Ozer

Avatar photo
I help companies train new technical hires in streaming media-related positions; I also help companies optimize their codec selections and encoding stacks and evaluate new encoders and codecs. I am a contributing editor to Streaming Media Magazine, writing about codecs and encoding tools. I have written multiple authoritative books on video encoding, including Video Encoding by the Numbers: Eliminate the Guesswork from your Streaming Video (https://amzn.to/3kV6R1j) and Learn to Produce Video with FFmpeg: In Thirty Minutes or Less (https://amzn.to/3ZJih7e). I have multiple courses relating to streaming media production, all available at https://bit.ly/slc_courses. I currently work as www.netint.com as a Senior Director in Marketing.

Check Also

NAB Session on AI in Video Streaming

Like most encoding professionals, I’ve followed AI-related streaming advancements for the last few years. I’m …

Technical Emmy won by NETINT.

Jan Ozer Becomes Unrestricted Free Agent

Watch out Saquon! Sorry for the spring training, pre-combine sports reference, but it seems appropriate.  …

Rating techniques that cut bandwidth costs.

Five Codec-Related Techniques to Cut Bandwidth Costs

The mandate for streaming producers hasn’t changed since we delivered RealVideo streams targeted at 28.8 …

Leave a Reply

Your email address will not be published. Required fields are marked *