Wednesday, March 9th, 2022 Posted by Jim Thacker

Apple unveils the M1 Ultra

Apple’s Peek Performance event. Skip to 00:23:30 to see the announcement of the new M1 Ultra processor.

Apple has unveiled the M1 Ultra, the latest chip in its new line of Apple Silicon processors, which it describes as the “most powerful … ever built for a personal computer”.

The processor, which combines two of Apple’s previous top-of-the-range M1 Max processors in a single unit, will ship later this month in the Mac Studio, Apple’s new professional desktop workstation.

The firm is targeting Mac Studio workstations equipped with M1 Ultra processors at a range of high-end graphics workflows, including 3D design and rendering, professional photography, and video editing.

According to Apple, a Mac Studio equipped with a M1 Ultra processor “can play back up to 18 streams of 8K ProRes 422 video [simultaneously] — a feat no other chip can accomplish”.

The firm also featured a range of CG applications, including After Effects, Cinema 4D and Redshift, while announcing the M1 Ultra at yesterday’s Peek Performance event. So how well will it perform with them?

Apple Silicon processor
M1 Ultra M1 Max M1 Pro M1
Unified memory (1) 128GB 64GB 32GB 16GB
Memory bandwidth 800GB/s 400GB/s 200GB/s Not stated
CPU cores (1) 20 10 10 8
GPU cores (1) 64 32 16 8
GPU compute (1, 2)
21 10.4 5.2 2.6
Neural Engine cores 32 16 16 16
Video decode engines 2 1 1 1
Video encode engines 4 2 1 1
ProRes encode/decode engines 4 2 1
Release date 2022 2021 2021 2020

(1) Maximum configuration available
(2) Figure for the M1 processor is estimated

Two M1 Max processors in a single system on a chip
Although tech news sites had speculated that Apple would announce a next-generation M2 processor during yesterday’s user event, the announcement eventally proved to be “one last chip [in] the M1 family”.

Or rather, two chips in one: the M1 Ultra connects two of Apple’s existing M1 Max processors to create a single system on a chip (SoC) with 20 CPU cores, up to 64 GPU cores and up to 128GB of unified memory.

The dies are connected using Apple’s new UltraFusion architecture, which it claims provides “twice the connection density of any [other] technology available”, at 2.5TB/s of inter-processor bandwidth.

Doubling GPU core counts gives a corresponding increase in compute power for tasks like GPU rendering and video encoding, placing the M1 Ultra on a par with even high-end discrete GPUs for Windows systems.

Its quoted FP32 compute performance of 21 Teraflops is higher than any of Nvidia’s previous-generation Quadro RTX workstation GPUs – the Quadro RTX 8000 came in at 16.3 Teraflops – and higher than some of Nvidia’s current-gen Ampere-based RTX Axxxx cards.

In addition, the 128GB of unified memory available to a M1 Ultra inside a maximally specced Mac Studio exceeds the on-board memory of even the current top-of-the-range Nvidia RTX A6000.

In the case of the M1 Ultra, that memory has to be shared between CPU and GPU, but it still potentially increases the size of the 3D scenes that can be rendered inside graphics memory.

Faster and less power-hungry than “even the highest-end PC GPU”?
According to Apple, the performance-to-power-consumption ratio of the Apple Silicon processors is also considerably higher than those for discrete GPUs.

The firm’s press release describes the M1 Ultra as “delivering faster performance than even the highest-end PC GPU available while using 200 fewer Watts of power”.

Faster performance in what is an open question: in a footnote, Apple only states that performance was measured using “select industry‑standard benchmarks”.

The M1 Ultra was also tested against a system featuring a GeForce RTX 3090, Nvidia’s top-of-the-range gaming card, which online databases suggest has a lower FP32 performance than that for the RTX A6000.

Speed boosts for creative software
So how will the specs of the M1 Ultra translate into real-world performance in CG software?

Unlike with the announcement of the M1 Max, Apple hasn’t provided detailed performance comparisons to previous-generation systems for specific creative applications.

The video above shows staff from firms including Adobe, Boris FX and Maxon discussing the performance of the M1 Ultra, but they all essentially just say that it’s faster than previous-generation hardware.

The closest it comes to a specific figure is Maxon trainer Elly Wade saying that “scene rendering is up to 4x faster”, presumably in Cinema 4D and Redshift, but it isn’t clear to which previous-gen processor the M1 Ultra is being compared.

So which CG applications natively support Apple Silicon?
It’s also worth noting that for a DCC application to take full advantage of the new hardware, its developer needs to have implemented native Apple Silicon support.

Other applications can run on M1 Macs via Apple’s Rosetta 2 translation environment, but the performance gains aren’t usually as significant.

At the time of writing, Autodesk and Foundry have yet to release native Apple Silicon builds of their software, and Adobe has yet to release a native build of After Effects or the Substance 3D tools.

US reseller Toolfarm has a regularly updated webpage that shows the current status of common post-production tools, and you can find our own list of CG apps that have announced native M1 support here.

Pricing and availability
M1 Ultra processors are due to ship in new Mac Studio workstations, available from 18 March 2022.

Pricing for a Mac Studio equipped with a 48-GPU-core M1 Ultra processor and 64GB of unified memory starts at $3,999, with a 64-GPU-core processor and 128GB of unified memory costing $1,800 extra.

Read more about the M1 Ultra processor on Apple’s website