- Mac Studio M3 Ultra96 GB unified memory
Full specifications
- Operating system
- macOS
- Chip
- Apple M3 Ultra
- CPU
- 28-core (20 performance, 8 efficiency); 32-core option
- GPU
- 60-core; 80-core option
- Neural Engine
- 32-core
- Memory
- 96 GB unified memory; 256 GB option; 819 GB/s
- Storage
- 1 TB SSD (up to 16 TB options)
- Rear ports
- 4 × Thunderbolt 5, 2 × USB-A, HDMI, 10 Gb Ethernet, headphone jack
- Front ports
- 2 × Thunderbolt 5, SDXC card slot
- Wireless
- Wi-Fi 6E, Bluetooth 5.3
Good for
- Running 70B models fast, or 120B-class models with 256 GB
- Professionals combining local AI with heavy video, 3D or audio work
- A near-silent machine with many Thunderbolt 5 ports
Not ideal for
- Models up to 30B only: a Mac mini or a single graphics card is enough
- CUDA-only training code
- Buyers who want to upgrade memory or storage later
Apple
Apple Mac Studio (M3 Ultra)
The Mac Studio puts M3 Ultra, Apple's most powerful desktop chip, in a compact aluminium case. M3 Ultra pairs a very large unified memory pool with 819 GB/s of bandwidth, a combination that runs large language models quickly while staying close to silent.
AED 16,169Charged in US dollars at checkout: $4,404. Import duties and taxes included.In stock. Dispatched in 1 to 3 business days.
- Free shipping to the UAE
- Ships from our warehouse in Malaysia
- 30-day returns
- Import duties and taxes included
- Shipping, returns and warranty
macOS keeps part of the unified memory for the system, so plan for the GPU to use most of it, not all. Size the configuration with some headroom above the model.
Who it's for
Professionals who want to run the largest open models locally on one quiet machine, alongside heavy creative work.
Full specifications
- Operating system
- macOS
- Chip
- Apple M3 Ultra
- CPU
- 28-core (20 performance, 8 efficiency); 32-core option
- GPU
- 60-core; 80-core option
- Neural Engine
- 32-core
- Memory
- 96 GB unified memory; 256 GB option; 819 GB/s
- Storage
- 1 TB SSD (up to 16 TB options)
- Rear ports
- 4 × Thunderbolt 5, 2 × USB-A, HDMI, 10 Gb Ethernet, headphone jack
- Front ports
- 2 × Thunderbolt 5, SDXC card slot
- Wireless
- Wi-Fi 6E, Bluetooth 5.3
About the Apple Mac Studio (M3 Ultra)
The Mac Studio with M3 Ultra is Apple's most powerful desktop chip in a compact aluminium case: a 28-core CPU, 60-core GPU, 96 GB of unified memory and 819 GB/s of memory bandwidth. The 96 GB configuration runs 70B models at 4-bit, and the 256 GB option takes 120B-class models at 8-bit. Buy it to run large open models quickly on one quiet machine alongside heavy video, 3D or audio work.
Key features
- Apple M3 Ultra: 28-core CPU and 60-core GPU, or 32-core CPU and 80-core GPU
- 96 GB unified memory, configurable to 256 GB; 819 GB/s
- 32-core Neural Engine
- 6 × Thunderbolt 5 (4 rear, 2 front), up to 120 Gb/s
- Up to eight displays
- 10 Gb Ethernet, Wi-Fi 6E, Bluetooth 5.3
- 1 TB SSD, configurable up to 16 TB
What it runs
We count about 75% of unified memory as usable for models, because macOS keeps a share for the system. That gives these ceilings:
| Memory | Usable for models (approx.) | Largest tier that fits |
|---|---|---|
| 96 GB | 72 GB | 70B at 4-bit; 70B at 8-bit and 120B-class do not fit |
| 256 GB | 192 GB | 120B-class at 8-bit |
Pick the configuration with headroom above the largest model you plan to run, since memory cannot be added later. Our model fit tool checks any model size against both.
Performance in context
Bandwidth is what sets the M3 Ultra apart from other large-memory desktops. At 819 GB/s it is about three times the 273 GB/s of 128 GB mini PCs and the DGX Spark, and text generation speed on large models tracks memory bandwidth closely. A 70B model that fits on any of these machines will answer noticeably faster here. A single high-end graphics card is still faster for models that fit in its memory.
Ports
- Rear: 4 × Thunderbolt 5 (up to 120 Gb/s), 2 × USB-A (up to 5 Gb/s), HDMI 2.1, 10 Gb Ethernet, 3.5 mm headphone jack
- Front: 2 × Thunderbolt 5 and an SDXC (UHS-II) card slot
- Displays: up to eight at 6K 60 Hz, or four at 8K 60 Hz
Six Thunderbolt 5 ports leave room for fast external storage, capture devices and several displays at once. Internal storage starts at 1 TB and can be configured to 2, 4, 8 or 16 TB; like memory, it is fixed once the machine is built, so model libraries often live on an external drive.
Power, size and noise
Maximum continuous power is 480 W from an internal supply. The case measures 19.7 × 19.7 × 9.5 cm and the M3 Ultra model weighs 3.64 kg. It is built to sit on a desk next to the display.
Software
On macOS, local models run through Ollama, LM Studio and llama.cpp, and Apple's open-source MLX framework is designed for Apple silicon. There is no CUDA, so CUDA-only code needs an NVIDIA system.
What to pair it with
A Thunderbolt 5 SSD array for large projects and model libraries, and a 10 Gb switch to use the built-in Ethernet. If 64 GB is enough, the Mac mini with M4 Pro costs far less. Read Mac vs NVIDIA vs Ryzen AI Max+ to compare large-memory options.
In the box
- Mac Studio
- Power cord
Good for
- Running 70B models fast, or 120B-class models with 256 GB
- Professionals combining local AI with heavy video, 3D or audio work
- A near-silent machine with many Thunderbolt 5 ports
Not ideal for
- Models up to 30B only: a Mac mini or a single graphics card is enough
- CUDA-only training code
- Buyers who want to upgrade memory or storage later
What it can run
Up to 70B at 4-bit, 30–34B at 8-bit (approximate)
Based on 96 GB of unified memory, about 75% of which the GPU can use. See every machine by model size
Guides for this product
Local AI · 5 min read
How Much Memory Do You Need to Run a Local LLM?
How much memory a local LLM needs: 7B to 120B+ models at 4-bit and 8-bit, how unified memory differs from VRAM, and which machines hold each size.
Read the guide →Local AI · 5 min read
Mac vs NVIDIA vs Ryzen AI Max+ for Local AI
Which platform fits which work: CUDA on NVIDIA, unified memory on a Mac, or 128 GB on a Ryzen AI Max+ mini PC. Memory, software and trade-offs.
Read the guide →Reviews
Atomrig customer reviews
No customer reviews yet. Bought this from us? Tell us how it runs.
Related products

Apple
Apple Mac mini (M4 Pro)
- 24 GB unified
- 16-core; 20-core option
Runs up to 13–14B at 4-bit
Memory cannot be upgraded after purchase. If local AI is the goal, choose 48 GB or 64 GB now. With 24 GB, models above about 14B parameters are a squeeze.
AED 5,659
NVIDIA
NVIDIA GeForce RTX 5090
- 32 GB
- NVIDIA Blackwell (GB202)
Runs up to 30–34B at 4-bit
32 GB runs 30B-class models comfortably at 4-bit, but a 70B model will not fit on one card. For 70B, plan for two cards or a large unified-memory system.
AED 7,339
PowerColor
PowerColor AMD Radeon AI PRO R9700 32GB
- 32 GB
- AMD RDNA 4 (Navi 48)
Runs up to 30–34B at 4-bit
Check that your tools support ROCm on your operating system before buying. PyTorch and llama.cpp do; some niche tools still assume CUDA.
AED 4,769
ASRock
ASRock Intel Arc Pro B60 Creator 24GB
- 24 GB
- Intel Xe2 (Battlemage)
Runs up to 30–34B at 4-bit
This is a single-GPU card with 24 GB. It uses a PCIe 5.0 slot wired at x8 and also works in PCIe 4.0 slots; check ASRock's spec sheet for the exact length and power connector before you order.
AED 2,199