Apple‘s M1 Ultra and AMD‘s Threadripper are two of the most powerful CPUs available today for demanding workloads like content creation, engineering simulations, data science, and high-end gaming. Both offer massive multi-core performance, but they take very different approaches to delivering it. In this in-depth comparison, we‘ll examine the strengths and weaknesses of each platform to help professionals and power users decide which one is right for their needs.
Tale of the Tape: M1 Ultra and Threadripper Specs Compared
Let‘s start with the raw specs. The M1 Ultra is Apple‘s top-end Apple Silicon chip, essentially fusing two M1 Max dies together using a novel packaging technology Apple calls UltraFusion. This enables the M1 Ultra to function as a single mammoth SoC with the following compute resources:
Spec | M1 Ultra |
---|---|
CPU Cores | 20 |
GPU Cores | 64 |
Neural Engine Cores | 32 |
Transistors | 114B |
Memory | 128GB Unified LPDDR5 |
Memory Bandwidth | 800 GB/s |
Process Node | TSMC 5nm |
The 20 CPU cores are split into 16 high-performance "Firestorm" cores and 4 efficiency "Icestorm" cores, a design Apple calls "Hybrid Architecture." Other standout specs include the massive 128GB of unified LPDDR5 memory and an incredible 800GB/s of memory bandwidth – more than many server CPUs.
On the Threadripper side, the current HEDT flagship is the 5995WX, part of the "Pro" designated line. It‘s based on AMD‘s Zen 3 architecture and manufactured on TSMC‘s 7nm node. Here‘s how the specs stack up:
Spec | Threadripper 5995WX |
---|---|
Cores/Threads | 64/128 |
Base/Boost Clock | 2.7/4.5 GHz |
Cache (L2+L3) | 288MB |
PCIe Lanes | 128 (Gen4) |
Memory Support | 8-channel DDR4-3200, up to 2TB |
TDP | 280W |
The 5995WX doubles the core count vs the M1 Ultra, with an industry-leading 64 cores and 128 threads. It also offers a staggering 128 PCIe Gen4 lanes and support for up to 2TB of DDR4 memory in a quad-channel configuration.
The differences in the underlying designs are stark. The M1 Ultra leverages a system-on-chip (SoC) architecture, with the CPU, GPU, memory, and I/O all integrated onto a single package. The 5995WX uses a more traditional chiplet-based CPU design, with the I/O and memory controllers on a separate die.
This reflects the fundamental difference in approach – Apple is focused on tight integration and efficiency, while AMD is all about modularity and brute force performance through core counts and bandwidth.
Performance Benchmarked
So how do these two heavyweights stack up in terms of real-world performance? Let‘s dive into some benchmark results.
In Geekbench 5, a synthetic test that measures raw single-core and multi-core performance, the M1 Ultra posts leading scores of 1793 and 24055, respectively. The Threadripper 5995WX is not far behind in multi-core at 22708, but lags significantly in single-core at 1213.
However, Geekbench is not always representative of real-world workloads. In the PugetBench suite of tests, which evaluate performance in content creation apps, we see the 5995WX pull ahead in some scenarios. In PugetBench for After Effects, the 5995WX scores a 2152 versus the M1 Ultra‘s 1718. In PugetBench for Premiere Pro, two leading video editing apps, the margin is even wider – the 5995WX scores 2040 compared to the M1 Ultra‘s 1285.
But the M1 Ultra strikes back in other creative apps. In PugetBench for Photoshop, the M1 Ultra outpaces the 5995WX by about 23% (1449 vs 1174). And in the relatively new PugetBench for DaVinci Resolve Studio, a video color correction and editing app, the M1 Ultra commands a significant lead of 2072 to 1705 – thanks in large part to optimizations for the M1 SoC‘s unique capabilities like unified memory.
The performance story is similar in productivity and scientific applications. The 5995WX is nearly untouchable in highly-parallelized CPU rendering workloads that scale well with core counts. In Cinebench R23, a popular 3D rendering benchmark, the 5995WX scores 71746 in the multi-core test compared to the M1 Ultra‘s still impressive 48485.
But the M1 Ultra makes up ground in benchmarks that tax both the CPU and GPU, or that can take advantage of the huge pool of unified memory. For example, in the Affinity Photo benchmark, which tests performance in the popular image editing app, the M1 Ultra is about 28% faster than the 5995WX.
In the Blender benchmark, which measures both CPU and GPU rendering performance, the M1 Ultra also comes out on top. In the Koro CPU rendering test, it achieves a score of 1542 versus the 5995WX‘s 1148. And in the Classroom GPU test, the 64-core GPU in the M1 Ultra completes the render in just 184 seconds, while a system with the 5995WX and an NVIDIA RTX A6000 GPU (a $4500 card) takes 254 seconds.
The performance crown ultimately depends on the specific workload. The 5995WX remains the absolute king of multi-threaded CPU performance, but the M1 Ultra is no slouch. And in applications that can leverage the M1‘s huge memory bandwidth, unified memory architecture, and powerful on-chip accelerators, it can actually outperform the highest-end Threadrippers.
Performance Per Watt
Where the M1 Ultra really distinguishes itself from Threadripper is in performance per watt. The M1 Ultra sips just 60-100W under load, while the 5995WX has a default TDP of 280W – and that‘s just for the CPU. A full Threadripper system under heavy load can draw 500-700W at the wall.
In terms of raw efficiency, the M1 Ultra delivers nearly 5x the single-threaded performance per watt and over 2.5x the multi-threaded performance per watt of the 5995WX. That‘s a testament to the incredible efficiency of the M1‘s ARM-based architecture and design.
This efficiency advantage becomes even more pronounced when considering the power consumption of the surrounding platform. The M1 Ultra achieves its chart-topping performance within the power envelope of a standard desktop PC. The Mac Studio has a maximum power draw of 370W – for the entire system.
In contrast, Threadripper systems have monstrous power requirements. A single 5995WX CPU can draw close to 400W with PBO enabled, and high-end motherboards have VRMs capable of delivering over 1000W. When combined with power-hungry GPUs, memory, and cooling systems, full-spec Threadripper workstations can draw 1500W or more under heavy load.
This has significant implications for real-world use. The lower power consumption of the M1 Ultra enables a radically smaller and quieter system design, with lower operating costs. A Mac Studio will add about $60 to a typical yearly electricity bill, while a Threadripper system can easily add $250 or more.
Moreover, the minimal cooling requirements allow the Mac Studio to take up less than 8 inches square on a desk. Most Threadripper systems are massive full towers packed with fans and liquid cooling radiators. They take up a lot of space and generate a lot of noise.
For many professionals, the power efficiency and portability of the M1 platform will be a game changer. It enables serious workstation-class performance in a form factor that can fit in a home office, recording studio, or even a backpack. The benefits to a company‘s bottom line in power costs alone can also be substantial at scale.
Ecosystem and Total Cost of Ownership
Of course, performance is only one part of the equation when considering a professional workstation. The software ecosystem and total cost of ownership are equally important factors.
Threadripper benefits from the broad compatibility of the x86-64 architecture and established software support in the professional market. Windows is still the dominant OS in many enterprise environments, and the vast majority of engineering and content creation apps are certified for Threadripper.
The M1 ecosystem is more of a mixed bag. There is a growing list of applications optimized for Apple Silicon, including mainstays like the Adobe Creative Suite and Maxon‘s Cinema 4D. For the apps that are native for M1, the performance is often transformative.
But there are still gaps in application support, especially for niche professional software. Many specialized tools in VFX, CAD/CAM, and scientific computing do not yet have native M1 versions. They have to either run through Apple‘s translation layer, Rosetta 2, or rely on cloud-based virtualization solutions.
This is changing rapidly as more developers port their applications to ARM, but it remains an advantage for the Threadripper platform in the near term. Professionals will need to carefully evaluate application support for their specific use case and workflow.
In terms of total cost of ownership, comparisons can be a bit difficult given the differences in system configurations between Mac Studio and Threadripper workstations. But in general, the M1 Ultra systems have a significant up-front cost advantage.
A fully-loaded Mac Studio with the M1 Ultra 20-core CPU/64-core GPU, 128GB RAM, and 8TB SSD costs $7999. Building an equivalent Threadripper Pro 5995WX workstation with 128GB RAM, 8TB NVMe storage, and a professional GPU like an NVIDIA RTX A6000 would cost well over $15000 – and that‘s before adding a premium for the pre-built system and support.
Considering the included Afterburner accelerator card, unique I/O capabilities, and significantly lower operating costs of the M1 Ultra, the TCO advantage becomes even more appealing for companies outfitting an entire workforce with new machines.
Of course, the calculation can change for workflows that require maximum CPU core counts, extremely large memory capacities, or highly-specialized I/O and accelerators. In these edge cases, Threadripper Pro can still be the platform of choice.
But for the vast majority of professional users, who prioritize strong single and lightly-threaded performance, GPU acceleration, and memory bandwidth, the M1 Ultra is arguably the better overall value in terms of features and performance per dollar.
Conclusions and Future Outlook
Ultimately, the choice between the M1 Ultra and Threadripper comes down to the specific requirements of the workload and the state of software support for each platform. Both offer incredible levels of performance for serious professionals.
The Threadripper 5995WX remains the undisputed leader in multi-threaded CPU performance and offers the most configuration flexibility. For highly-parallel workloads that scale with core counts, high-end Threadripper systems are tough to beat.
The M1 Ultra, on the other hand, delivers a better balance of single and multi-threaded performance, GPU compute, and power efficiency. The unified memory architecture and purpose-built accelerators give it a unique edge in applications optimized for Apple Silicon.
Looking forward, the competition between Apple and AMD in the workstation space is only going to intensify. Apple has already previewed the M2 as the next-generation of Apple Silicon, fabbed on TSMC‘s 3nm node. Rumors are already swirling of an even higher-end M2 Extreme model that could make its debut in the Mac Pro.
On the Threadripper side, AMD has announced the "Storm Peak" generation set to launch in 2024. It will feature the Zen 4 core architecture, a new SP5 socket, and support for DDR5 memory and PCIe 5.0. This could bring substantial IPC and bandwidth improvements to the platform.
But the real wildcard is how the software ecosystem evolves and which platform developers prioritize for optimizations. As Apple Silicon continues to gain market share in the professional market, more ISVs are likely to treat it as a first-class citizen. The tables could turn quickly if key applications begin to run better on M1 than x86.
As it stands today, both the M1 Ultra and Threadripper 5995WX offer uncompromising performance for the most demanding professional use cases. The choice comes down to individual needs and preferences. But one thing is for certain – the pace of innovation in this space is extraordinary and shows no signs of slowing down. It‘s an exciting time to be in the market for serious computing power.