Apple M1 Ultra

Apple M1 Ultra

Apple M1 Ultra: Power and Efficiency in One Chip

March 2025


1. Key specifications: Architecture, process technology, performance

Architecture and process technology

The Apple M1 Ultra, introduced in 2022, remains a benchmark of engineering excellence even in 2025. It is the world's first processor that combines two M1 Max chips into a single system (UltraFusion technology). The process technology is 5 nm (TSMC), with 114 billion transistors. The architecture is ARMv8.4-A featuring Firestorm (performance) and Icestorm (energy-efficient) cores.

Performance

- 20 cores (16P + 4E), 20 threads (Hyper-Threading is absent).

- Frequencies: 2.0 GHz (base for P-cores), up to 3.2 GHz in turbo mode.

- GPU: 64 cores, 21 TFLOPs.

- Geekbench 6: 2390 (Single-Core), 18219 (Multi-Core). For comparison, the Ryzen 9 7950X (2023) scores around ~2800/22000 but consumes twice as much power.

Key Features

- Unified Memory (up to 128 GB LPDDR5-6400): data is available to CPU, GPU, and Neural Engine simultaneously.

- Neural Engine (32 cores): 22 TOPS for AI tasks.

- Hardware acceleration for codecs (ProRes, H.265).


2. Compatible motherboards

Apple platform features

The M1 Ultra is part of a system-on-chip (SoC) integrated into Apple devices. There are no standalone motherboards for it. The chip is only installed in:

- Mac Studio (2022-2024).

- Mac Pro (2023), but with limited upgrade options.

Chipsets and sockets

- Integrated PCIe 4.0 controller (up to 128 lanes), but Apple uses proprietary connectors.

- For peripheral connections - Thunderbolt 4/USB4 (up to 6 ports).

Advice: For maximum M1 Ultra performance, choose a Mac Studio with enhanced cooling system.


3. Supported memory

Unified Memory instead of DDR

- Type: LPDDR5-6400.

- Capacity: 64 GB (base), 128 GB (optional).

- Bandwidth: 800 GB/s - four times higher than DDR5-6400.

Features

- Memory is soldered onto the chip: upgrades are impossible.

- For professional tasks (8K video, 3D rendering), it’s better to choose 128 GB.


4. Power supply recommendations

Power consumption

- TDP: 120 W (peak load - up to 200 W).

- Mac Studio is equipped with a 370 W power supply, which covers the chip and peripheral consumption.

Tips

- For stable operation of Mac Studio, do not use UPS with a power rating below 500 W.

- When connecting external GPUs (via Thunderbolt), consider their power consumption.


5. Pros and Cons

Pros

- Energy Efficiency: 3.2 GHz at 120 W vs. 5 GHz for competitors at 250+ W.

- Integration with macOS: optimized for Final Cut Pro, Logic Pro, Xcode.

- Quiet Operation: passive cooling in most scenarios.

Cons

- Price: Mac Studio with M1 Ultra ranges from $3500 (base) to $7000 (128 GB + 8 TB SSD).

- No upgrades: RAM and SSD are soldered.

- Limited compatibility: x86 software runs via Rosetta 2 with a 10-20% performance loss.


6. Use cases

Professional tasks

- Video editing: Real-time 8K ProRes rendering (for example, editing in DaVinci Resolve on Mac Studio takes half the time compared to a PC with RTX 4090).

- 3D modeling: Blender, Cinema4D - utilizing the Metal API.

- Machine learning: training models in TensorFlow using the Neural Engine.

Gaming

- Not a primary niche, but Baldur’s Gate 3 (ARM version) achieves 60 FPS at 4K.

- Windows emulation via Parallels - suitable for less demanding projects.

Multimedia

- 8K streaming without lags.


7. Comparison with competitors

AMD Ryzen Threadripper 7970X (2024)

- 32 cores, 64 threads, 350 W.

- Geekbench 6 Multi-Core: ~24000, but Single-Core - 2100.

- Price: $2500 (CPU only).

Intel Core i9-14900KS (2024)

- 24 cores (8P + 16E), 320 W.

- Geekbench 6: 3100/23000.

- Price: $700.

Conclusion: The M1 Ultra excels in energy efficiency and tasks optimized for Apple Silicon but falls short in multithreading compared to high-end desktop CPUs.


8. Practical assembly tips

What to consider

- Choosing a device: Mac Studio vs. Mac Pro. The former is more compact and cheaper, while the latter is for those needing PCIe slots.

- Configuration: 128 GB of memory is essential for working with neural networks or heavy 3D tasks.

- Peripherals: Thunderbolt drives (like SanDisk Extreme Pro 1600 MB/s) for fast data access.

Example assembly (Mac Studio):

- M1 Ultra (64-core GPU), 128 GB RAM, 4 TB SSD - $6500.

- Monitor: Pro Display XDR (32″, 6K) - $5000.


9. Final conclusion: Who is the M1 Ultra for?

Ideal user:

- Professionals in video, music, and 3D graphics using macOS.

- Developers working with iOS applications and AI.

- Those who value quiet operation and minimalism.

Not recommended for:

- Gamers (better off with a PC with RTX 5080).

- Enthusiasts who enjoy upgrading hardware.

Why is it still relevant in 2025?

Apple has yet to release a successor to the M1 Ultra, and its balance of power and efficiency remains unparalleled for niche tasks.


Prices are stated for new devices as of March 2025. The relevance of the data is confirmed by open sources.

Basic

Label Name
Apple
Platform
Desktop
Launch Date
March 2022
CPU Architecture
Apple Firestorm + Apple Icestorm
CPU Name
Apple M1 Ultra
Model Name
?
The Intel processor number is just one of several factors - along with processor brand, system configurations, and system-level benchmarks - to be considered when choosing the right processor for your computing needs.
Apple M1 Ultra
Foundry
TSMC
Generation
Apple M1 series

CPU Specifications

Interconnect Bandwidth
2.5 TB/s
Interconnect Technology
UltraFusion
Performance Cores
16
Total Cores
?
Cores is a hardware term that describes the number of independent central processing units in a single computing component (die or chip).
20
Total Threads
?
Where applicable, Intel® Hyper-Threading Technology is only available on Performance-cores.
20
Efficient-cores
4
Efficient-core Max Turbo Frequency
?
Maximum E-core turbo frequency derived from Intel® Turbo Boost Technology.
2.064 GHz
Instruction Set Extensions
NEON
Performance-core Max Turbo Frequency
?
Maximum P-core turbo frequency derived from Intel® Turbo Boost Technology.
3.228 GHz
Extended Instruction Set
ARMv8-A, NEON
L1 Cache
P-cores: 192 KB instruction + 128 KB data per core; E-cores: 128 KB instruction + 64 KB data per core
L2 Cache
P-core clusters: 48 MB total; E-core clusters: 8 MB total
L3 Cache
96 MB system level cache
Technology
?
Lithography refers to the semiconductor technology used to manufacture an integrated circuit, and is reported in nanometer (nm), indicative of the size of features built on the semiconductor.
5 nm
Instruction Set
?
The instruction set is a hard program stored inside the CPU that guides and optimizes CPU operations. With these instruction sets, the CPU can run more efficiently. There are many manufacturers that design CPUs, which results in different instruction sets, such as the 8086 instruction set for the Intel camp and the RISC instruction set for the ARM camp. x86, ARM v8, and MIPS are all codes for instruction sets. Instruction sets can be extended; for example, x86 added 64-bit support to create x86-64. Manufacturers developing CPUs that are compatible with a certain instruction set need authorization from the instruction set patent holder. A typical example is Intel authorizing AMD, enabling the latter to develop CPUs compatible with the x86 instruction set.
ARMv8-A
Transistor Count
114 billion

Memory Specifications

Memory Bus Width
1024-bit
Memory Type
?
Intel® processors come in four different types: Single Channel, Dual Channel, Triple Channel, and Flex Mode. Maximum supported memory speed may be lower when populating multiple DIMMs per channel on products that support multiple memory channels.
Unified LPDDR5-6400
LPDDR5 Speed
LPDDR5-6400
Max Memory Size
?
Max memory size refers to the maximum memory capacity supported by the processor.
128 GB
Memory Channels
?
The number of memory channels refers to the bandwidth operation for real world application.
8
Max Memory Bandwidth
?
Max Memory bandwidth is the maximum rate at which data can be read from or stored into a semiconductor memory by the processor (in GB/s).
800 GB/s
Maximum Memory Speed
6400 MT/s

GPU Specifications

GPU APIs
Metal, OpenCL
GPU Name
Apple M1 Ultra GPU
Media Engine
Hardware-accelerated H.264, HEVC, ProRes, and ProRes RAW
ProRes Encode/Decode Engines
4
Video Concurrency
Up to 18 streams of 8K ProRes 422 video
Video Decode
H.264, HEVC, ProRes, ProRes RAW
Video Decode Engines
2
Video Encode
H.264, HEVC, ProRes, ProRes RAW
Video Encode Engines
4
Video Processing Unit
Apple media engine with ProRes acceleration
GPU Base Frequency
450 MHz
GPU Max Dynamic Frequency
1296 MHz
Graphics Core Count
64
Number of Displays Supported
Up to 5 displays
Graphics Performance
Up to 21.2 TFLOPS FP32
OpenCL Support
?
OpenCL (Open Computing Language) is a multi-platform API (Application Programming Interface) for heterogeneous parallel programming.
Yes

AI Specifications

AI Engine
32-core Apple Neural Engine
Neural Engine Core Count
32
NPU Name
Apple Neural Engine
NPU Performance
22 TOPS

Interfaces and Ports

Thunderbolt Support
Yes, Thunderbolt 4 up to 40 Gb/s
USB Version
USB4
USB4 Support
Yes, USB4 up to 40 Gb/s

Miscellaneous

Hardware-Verified Secure Boot
Yes, Apple silicon secure boot chain of trust
Runtime Anti-Exploitation
Kernel Integrity Protection, Pointer Authentication Codes, Fast Permission Restrictions
Security Processor
Secure Enclave

Benchmarks

Cinebench R23
Single Core Score
1534
Cinebench R23
Multi Core Score
24189
Geekbench 6
Single Core Score
2398
Geekbench 6
Multi Core Score
18458
Geekbench 5
Single Core Score
1793
Geekbench 5
Multi Core Score
24055
Passmark CPU
Single Core Score
3860
Passmark CPU
Multi Core Score
41380
Cinebench 2024
Single Core Score
114
Cinebench 2024
Multi Core Score
1624
Cinebench 2024 GPU
Score
3331
Blender
Score
379

Compared to Other CPU

Cinebench R23 Single Core
1898 +23.7%
1749 +14%
1534
1186 -22.7%
945 -38.4%
Cinebench R23 Multi Core
45651 +88.7%
24189
15852 -34.5%
13066 -46%
10027 -58.5%
Geekbench 6 Single Core
2605 +8.6%
2488 +3.8%
2398
2286 -4.7%
2212 -7.8%
Geekbench 6 Multi Core
30800 +66.9%
21456 +16.2%
18458
16754 -9.2%
Geekbench 5 Single Core
2036 +13.6%
1875 +4.6%
1793
1705 -4.9%
1644 -8.3%
Geekbench 5 Multi Core
86607 +260%
36476 +51.6%
24055
19393 -19.4%
16064 -33.2%
Passmark CPU Single Core
3915 +1.4%
3860
3803 -1.5%
3735 -3.2%
Passmark CPU Multi Core
48288 +16.7%
45268 +9.4%
41380
38100 -7.9%
35626 -13.9%
Cinebench 2024 Single Core
126 +10.5%
123 +7.9%
114
110 -3.5%
Cinebench 2024 Multi Core
1767 +8.8%
1624
1360 -16.3%
1203 -25.9%
Cinebench 2024 GPU
5968 +79.2%
4514 +35.5%
3331
M3
3135 -5.9%
2423 -27.3%
Blender
1154 +204.5%
379
192 -49.3%
139 -63.3%
51 -86.5%