Quick Verdict
An extraordinarily powerful workstation-class system-on-chip that delivers exceptional multi-threaded and GPU performance with remarkable power efficiency, though its locked ecosystem and discontinued status make the newer M2 Ultra or M3 Ultra worth considering.
Overview
Launch
2022
Status
DiscontinuedGeneration
1st Gen Apple Silicon Ultra
Market
High-End Workstation Desktop
The Apple M1 Ultra is the most powerful chip in Apple's first-generation Apple Silicon lineup, engineered by fusing two M1 Max dies through the proprietary UltraFusion interconnect. With 20 CPU cores, up to 64 GPU cores, a 32-core Neural Engine, and up to 128GB of unified LPDDR5 memory delivering 800 GB/s bandwidth, it targets the most demanding professional workflows in a compact desktop form factor.
The Apple M1 Ultra fuses two M1 Max chips using UltraFusion, creating a 20-core CPU with 16 performance and 4 efficiency cores, up to 64 GPU cores, and a 32-core Neural Engine. With up to 128GB of unified LPDDR5 memory at 800 GB/s, it handles 8K video editing, complex 3D rendering, and large machine learning models with ease. Available exclusively in the Mac Studio, it was the fastest Mac ever at launch and remains a potent workstation solution despite being superseded by the M2 Ultra.
Specifications
Performance
Exceptional multi-threaded performance for creative and professional applications, handling the heaviest workloads with ease.
Competent virtualization through Parallels and UTM, but ARM-based VMs have better compatibility than x86 VMs.
Capable of running many titles at high settings, but the macOS game library is limited and no external GPU support exists.
Remarkable performance-per-watt compared to competing x86 workstations, delivering more work per unit of energy consumed.
- •64-core GPU handles most macOS-native titles comfortably at 1440p
- •Limited game library on macOS compared to Windows
- •Rosetta 2 translation layer adds minor overhead for x86 games
- •No support for external GPUs via Thunderbolt
- •AAA titles running through CrossOver or Parallels may have reduced performance
- •32-core Neural Engine accelerates on-device machine learning inference at up to 22 trillion operations per second
- •800 GB/s memory bandwidth benefits large language model inference
- •128GB unified memory enables loading large AI models that exceed typical GPU VRAM
- •No dedicated tensor cores in the traditional NVIDIA CUDA sense
- •Apple Core ML and Metal Performance Shaders provide software-level acceleration
Architecture
TSMC 5nm
Process Node
Jade 2C Die
Codename
20C / 20T
Core Config
Architecture Overview
The M1 Ultra architecture represents Apple's first multi-die implementation of Apple Silicon, connecting two complete M1 Max dies through the UltraFusion interconnect. Unlike traditional multi-chip module approaches from AMD or Intel that use high-speed serial links, UltraFusion employs a dense silicon interposer with over 10,000 signals bridging the two dies, achieving 2.5 TB/s of bandwidth with latency low enough that the operating system and applications perceive the combined chip as a single unified processor. This seamless integration eliminates the software complexity typically associated with multi-die designs.
CPU Design
The CPU comprises 16 Firestorm performance cores and 4 Icestorm efficiency cores. Each performance cluster contains four cores sharing a 12MB L2 cache, while the efficiency cluster shares a 4MB L2 cache. The performance cores handle demanding computational workloads such as video encoding, compiling code, and 3D rendering, while the efficiency cores manage background tasks like mail syncing, music playback, and system services with minimal energy consumption. Apple's scheduler dynamically routes threads to the appropriate core type based on workload demands.
Memory Subsystem
The M1 Ultra features a 512-bit unified LPDDR5 memory interface running at 6400 MT/s, delivering 800 GB/s of bandwidth. Unlike traditional architectures where the CPU and GPU maintain separate memory pools with explicit data transfers, the unified memory architecture allows all processing units to access the same physical memory directly. This eliminates the latency and bandwidth penalty of copying data between CPU RAM and GPU VRAM, providing a significant advantage in memory-intensive workloads like video processing and GPU compute tasks.
PCIe & I/O
Apple Silicon does not expose PCIe lanes in the traditional x86 sense. The M1 Ultra routes its I/O through a custom on-die fabric, providing Thunderbolt 4 ports, NVMe storage connectivity, and other interfaces. The Mac Studio with M1 Ultra includes four Thunderbolt 4 ports, each offering 40 Gbps bandwidth, along with HDMI 2.0, 10Gb Ethernet, USB-A, and an SDXC card slot. Direct PCIe slot expansion is not supported.
Overclocking
The M1 Ultra does not support overclocking. Apple's firmware tightly controls clock speeds, power management, and thermal behavior. Users have no access to multiplier adjustments, voltage controls, or manual frequency tuning. The chip's performance is governed entirely by Apple's dynamic power management algorithms, which balance performance and thermals automatically.
- Doubled CPU core count from 10 to 20
- Doubled GPU core count from 32 to 64
- Doubled Neural Engine cores from 16 to 32
- Doubled memory bandwidth from 400 GB/s to 800 GB/s
- Dual media encode and decode engines
Key Highlights
- Outstanding multi-threaded CPU performance with 20 cores
- Massive 800 GB/s unified memory bandwidth
- Up to 128GB unified memory accessible by CPU, GPU, and Neural Engine
- Extremely power-efficient compared to x86 workstations
- Dual ProRes encode and decode engines for video professionals
- 32-core Neural Engine for hardware-accelerated machine learning
- Quiet operation even under sustained heavy workloads
- Seamless multi-die operation transparent to software
- Only available in Mac Studio, no standalone or DIY option
- No support for external GPUs
- macOS has a limited game library compared to Windows
- Memory and storage are not user-upgradeable
- No hardware-accelerated ray tracing (introduced with M3 family)
- Discontinued and superseded by M2 Ultra
- No traditional PCIe expansion slots
- HDMI 2.0 instead of HDMI 2.1 limits external display options
History
Apple's journey to the M1 Ultra began with the company's historic decision in June 2020 to transition the entire Mac lineup away from Intel processors toward its own custom Apple Silicon. The original M1 chip debuted in November 2020, immediately demonstrating that ARM-based Macs could compete with and often surpass x86 counterparts in performance-per-watt. Apple expanded the architecture for professionals in October 2021 with the M1 Pro and M1 Max, introducing wider memory buses, more GPU cores, and dedicated media engines.
However, even the M1 Max left a performance ceiling for the most demanding desktop workstation users.</br></br>In March 2022, Apple unveiled the M1 Ultra at its Peek Performance event, revealing the UltraFusion interconnect that bonds two M1 Max dies into a single cohesive chip. This was a significant milestone in consumer semiconductor design, demonstrating that multi-die architectures could operate with low enough latency to be transparent to software.
The M1 Ultra shipped exclusively in the highest-end Mac Studio configuration, a new compact desktop form factor designed for professionals. At launch, it was the most powerful Mac ever produced, outperforming the Intel-based Mac Pro in many benchmarks while consuming a fraction of the power.</br></br>The M1 Ultra's legacy extends beyond raw performance.
It validated Apple's scalable chip design philosophy and established the UltraFusion technology that would underpin future Ultra-class chips. When the M2 Ultra succeeded it in June 2023, the multi-die approach was already well-established, and the concept of fusing two Max-class dies became an expected part of Apple's product cadence rather than a surprise innovation.
Improvements over Previous Generation
- Doubled CPU core count from 10 to 20
- Doubled GPU core count from 32 to 64
- Doubled Neural Engine cores from 16 to 32
- Doubled memory bandwidth from 400 GB/s to 800 GB/s
- Dual media encode and decode engines
Alternatives & Competitors
Should You Buy It?
Recommended for the right buyer
Professional content creators and workstation users who need massive multi-threaded performance and unified memory within the Apple ecosystem, particularly on the refurbished market.
Avoid if…
- You need regular access to Windows-only professional software
- You require PCIe expansion slots for add-in cards
- Gaming is your primary use case
- You want user-upgradeable RAM or storage
- A Mac Studio with M2 Ultra is available at a comparable price
Use Cases
Interesting Facts
The M1 Ultra contains approximately 114 billion transistors, making it one of the largest consumer chips ever manufactured at the time of its release.
UltraFusion's 2.5 TB/s interconnect bandwidth exceeded AMD's Infinity Fabric bandwidth at the time of launch.
The M1 Ultra can play back up to 18 streams of 4K ProRes 422 video simultaneously.
Software sees the M1 Ultra as a single chip, requiring no special optimization to use all 20 CPU cores.
Each M1 Max die on the M1 Ultra has its own dedicated media engine, effectively doubling video encoding throughput.
The Mac Studio M1 Ultra was the fastest Mac ever sold at its launch, outperforming the Intel Mac Pro in many real-world benchmarks.
Despite 20 CPU cores and 64 GPU cores, the Mac Studio remains significantly quieter than most 6-core Intel workstations under load.
The M1 Ultra was the first Apple Silicon chip to offer 128GB of unified memory in a Mac.
Apple achieved the multi-die design by using a custom silicon interposer rather than a traditional package-on-package or EMIB approach.
The 512-bit memory bus is wider than most desktop CPUs and even many server-grade processors.
People Also Ask
Is the M1 Ultra faster than the M2?
In multi-core workloads, the M1 Ultra is significantly faster than the base M2 due to its 20 cores versus 8. However, the M2 Ultra outperforms the M1 Ultra across the board.
How does the M1 Ultra compare to Intel processors?
The M1 Ultra competes with high-end desktop Intel processors in multi-threaded tasks while consuming substantially less power, but Intel chips offer better Windows compatibility and gaming performance.
Can the M1 Ultra run Windows?
Only through virtualization software like Parallels Desktop or UTM. Boot Camp is not supported on Apple Silicon Macs, and x86 Windows performance in VMs is limited.
What is the power consumption of the M1 Ultra?
Apple does not publish official TDP figures. The Mac Studio with M1 Ultra has a 370W power supply, and under maximum combined CPU and GPU load, the system draws approximately 280-350W.
Is the M1 Ultra better than the M1 Max for video editing?
Yes. The M1 Ultra offers double the media engines, more GPU cores, and higher memory bandwidth, which benefits multi-stream 4K and 8K ProRes workflows.
How does UltraFusion work?
UltraFusion uses a custom silicon interposer with over 10,000 signals connecting two M1 Max dies at 2.5 TB/s bandwidth, presenting them to the operating system as a single unified processor.
Can you buy the M1 Ultra as a standalone processor?
No. The M1 Ultra is exclusively available as part of the Mac Studio desktop computer and cannot be purchased or installed separately.
Does the M1 Ultra support hardware ray tracing?
No. Hardware-accelerated ray tracing was introduced with the Apple M3 family. The M1 Ultra relies on software-based ray tracing approaches.
Is the M1 Ultra good for machine learning?
Yes. The 32-core Neural Engine and 800 GB/s memory bandwidth make it strong for on-device ML inference, and the 128GB unified memory can accommodate large models that exceed typical GPU VRAM.
What is the difference between the 48-core and 64-core M1 Ultra GPU?
The 64-core variant has 16 additional GPU cores active, delivering approximately 25-30% higher graphics and compute performance compared to the 48-core variant.
Frequently Asked Questions
Which Mac models use the M1 Ultra?
The M1 Ultra is exclusively available in the high-end configuration of the Mac Studio (2022). No other Mac model has shipped with this chip.
Can the M1 Ultra Mac Studio be upgraded?
No. The M1 Ultra is soldered to the logic board, and both the unified memory and internal storage are not user-upgradeable. External storage via Thunderbolt is the only expansion option.
How many displays can the M1 Ultra support?
The Mac Studio with M1 Ultra supports up to five displays: up to four ProDisplay XDRs at 6K over Thunderbolt and one 4K display over HDMI.
Does the M1 Ultra support external GPUs?
No. macOS on Apple Silicon does not support external GPU acceleration, making the integrated 48 or 64-core GPU the only graphics option.
What is the Neural Engine in the M1 Ultra?
A 32-core dedicated machine learning accelerator capable of up to 22 trillion operations per second, handling tasks like image recognition, natural language processing, and video analysis.
Is the M1 Ultra good for programming and software development?
Yes. The 20-core CPU compiles large codebases quickly, and the unified memory handles memory-intensive development tools and containers efficiently.
Does the M1 Ultra support AV1 decoding?
No. AV1 hardware decoding was introduced with the M3 family. The M1 Ultra supports hardware decoding of H.264, HEVC, and ProRes formats.
How does the M1 Ultra handle heat?
The Mac Studio uses a large copper thermal module with a single fan that keeps the M1 Ultra cool and quiet even under sustained heavy workloads, rarely exceeding 45-50 dB.
Can the M1 Ultra run Docker containers?
Yes, but only ARM64 containers run natively. x86 containers can run through QEMU emulation with a performance penalty, which may be significant for compute-heavy workloads.
Is the M1 Ultra still worth buying in 2025?
Yes, particularly on the refurbished market. It remains one of the most powerful Macs available for professional creative workloads, though the M2 Ultra and M3 Ultra offer newer technology and improved efficiency.