Monday, January 10, 2022

How fast are Apple M1 computers for video?

Since Apple introduced their new M1 "system on a chip" there's been incredible hype over the capabilities of their laptops. In this article I will summarise information provided on the Blackmagic forums, in order to judge the worth of these systems for video editing. The results may surprise you. 

I adopt a sceptical approach to manufacturers' marketing copy. Unfortunately, YouTube does the opposite, acting as an amplifier for any fanbase, providing a platform for breathless videos that fail to critique new products. This applies twice over to Apple, who have particularly fervent fans.

The M1 is a revolutionary processor chip. It has two main characteristics that justify this statement. First, it integrates CPU (processor) and GPU (graphics) capabilities to a degree previously unseen. In particular, this allows memory to be freely allocated between these functions. Second, it has very low power requirements, making it eminently suitable for portable computing. 

Stripped of headlines, Apple's own copy goes no further than claiming that the M1 provides the best performance to power-consumption of any processor. Claims are also made of the performance increases of Apple silicon versus Intel. In the example screenshot, pertaining to DaVinci Resolve, it's claimed that the MacBook Pro 16" M1 Max has a performance increase of 1.9x versus the previous model [1]. Since Apple computers were never known for their video performance, this claim actually seems reasonable. 

Note that there's no comparison to hardware supporting other operating systems. An absolute metric is of greater importance than relative comparisons, at least to those of us who simply want to get work done.

In November 2021 a thread began on the Blackmagic forums entitled "New Apple MBP M1 + Resolve 17 - Is it worth the price?" In order to evaluate performance, one needs a real-world testbed. Uli Plank created a standard DaVinci Resolve project to act as a stress test. This utilised two particularly demanding VFX: speed warp and noise reduction. To run this yourself, you first need to grab the DRP file from the original post. Then get the two "James Perse - Robertson House" video clips from the Blackmagic URSA Mini Gallery here.

The test is simple: render the project and report the time it takes. On a Mac, you should target ProRes 422 HQ. On a PC, which lacks ProRes due to Apple's licensing, the project should instead be rendered to DNxHR. It should be noted that this gives the M1 Macs an (unfair?) advantage, since the Mac has on-chip ProRes encoding.

Many readers contributed benchmark numbers using this test. I've compiled a spreadsheet to summarise the results from that forum thread. The table includes four MacBook Pro M1 Max systems, ten older Apple computers, and three Windows desktop computers.

Testing Heldscalla

In recent posts I've documented my computer build, christened Heldscalla. This system includes a Ryzen 9 5900X processor, 64 GB RAM, and a GeForce RTX 3060 Ti video card. There are two SSD drives (1 TB and 2 TB) plus a 10 TB hard drive. This is not a workstation class computer, but is a moderately powerful build using consumer components. 

Attempting to use Plank's project, I immediately get the dreaded error message "Your GPU memory is full". This is because the 8GB of VRAM on my video card is insufficient to support the UHD resolution footage. This is entirely expected, since I bought the RTX 3060 Ti to support a maximum resolution of 4K.

One solution would be to create proxy files, but this project would then not be directly comparable to the other results. Instead, I transcoded the media to 4K, as Plank suggested. The project now rendered smoothly. During this process the GPU utilisation went up to 100%, but the CPU hovered around only 3-4%. While I know that Resolve is GPU-bound for these sorts of effects, I was still surprised that the CPU was doing so little. 

Since I'd designed this computer around a quiet air-flow design, there was only a purr to be heard, even under this graphics load. Internal temperatures maxed out at a very comfortable 60° C.

The render took only 4:17.

But is it fair to compare this test, using 4K footage, with the other results that use UHD resolution? Since the CPU and disk were not stressed in this test, they are not factors. Can we assume that the graphics processing cores scale appropriately with higher Nvidia model numbers? It appears so. Matt Cole used an RTX 3090 (24 GB VRAM) to produce a benchmark identical to my own. This supports the contention that it won't matter significantly which Nvidia RTX 3000 model you use. The main factor is obtaining the VRAM needed to support your desired timeline resolution.

This is an important point, since it's often assumed that the cost of a high-end RTX 3090 is necessary for a Windows-based PC to perform well. This test demonstrates that an RTX 3060 Ti is more than good enough for 4K resolutions.

Discussion

Apple's advertised 1.9x boost appears to be conservative. This is not surprising, since older Apple computers were often poor at video, despite any claims made while those models were current. 

It's also clear that the fastest computers are all Windows/Linux machines. This has always been true, and remains true even with the M1 chip in the equation. Claims that the M1 "blows away all others" are unfounded. It should be enough that the MacBook Pro M1 is the fastest portable computer currently available.

However, speed isn't everything. These products suffer in other respects. 

1. Apple laptops cannot be upgraded or serviced by the user. Repairs easily made by a Windows laptop user, even something as simple as replacing a battery, are forbidden. The increasing visibility of the "right to repair" movement demonstrates that many people care about such things. 

2. MacBooks have only a single internal drive. I generally recommend a video/media build that reserves an SSD for OS and apps, a HDD for project storage, and a second SSD as your Resolve cache. External drives should be used only for backup, since their bandwidth is compromised (even on the fastest ports). Certain Windows laptops support two internal drives, but Apple have chosen to limit this feature on their models.

3. The lack of ports prevent MacBooks from being desktop replacements. At my desk I daily use two video monitors, an audio interface, a control surface, card readers, external pointing device and keyboard (ergonomics insists), plus one or more backup drives. That's eight ports. I could perhaps get away with two less on a laptop (using the monitor and built-in keyboard). The first generation of M1 laptops had a grand total of two ports. The MacBook Pro models have four USB ports plus HDMI. (This includes power, so a fair comparison requires subtracting one from the count of free ports.)

4. Even with inflated GPU pricing, Heldscalla cost me €3150 for 13 TB storage (no monitor). The MacBook Pro 16" M1 Max 64 GB costs €4770 with only 2 TB of internal storage, or €6840 for 8 TB. As I write, the incremental cost (at retail) of this upgrade should be €600, not €2070. Can anyone feel good about getting ripped off like this?

A MacBook Pro with six ports, two internal drives, and market-pricing on expansion options would be a nice machine. Unfortunately, that would require a company with very different ethics from Apple, whose primary imperative is to fleece their customers to line the pockets of share-holders [2].

Conclusions

The current MacBook Pro laptops with M1 Max chip are at least twice as fast as the previous Intel-based laptops. But a consumer-grade desktop computer (running Windows or Linux) is twice as fast again. You give up portability in exchange for many expansion possibilities and the freedom to upgrade in the future. You also get to save half your money, which you can put towards something more productive. You might even buy a laptop for the work you need to do in the field [3].

Of course this all depends on your requirements. Apparently there exist video editors who must do all their high-end post-production away from the desk. Such users will no doubt be able to justify any expenditure to accomplish this task. In the majority of cases, however, a desktop solution is preferable.

Nonetheless, I look forward to the next generation of portable AMD chips (Ryzen 6000), which will enable high-performance Windows laptops with full-day battery life. Those might be competitive with the M1 chip for video production, but only time will tell.

Footnotes

[1] Apple are clear that about the conditions: "Tested with pre-release DaVinci Resolve Studio 17.4 using lens flare effect and a 10-second UHD project at 3840x2160 resolution and 24 frames per second."

[2] Yes, all corporations do this. But the unprecedented profits of Apple support the contention that they are the worst culprit.

[3] A MacBook Air M1 with 16GB RAM and a 1 TB drive is €1860. You can buy this plus a system like Heldscalla and still save €1830!

RELATED POSTS

2 comments:

robin said...

This comparison from Zebra Zone uses last-gen Windows hardware but is nonetheless interesting. It's not nearly so one-sided as other discussions.

https://youtu.be/RjmDMuM70Lk

robin said...

This article was significantly updated on 11 January 15:00 UDT.

Post a Comment