Apple launches M5 Ultra Mac Studio and M6 Mac Mini with quad-die architecture

The new desktop lineup focuses on high-bandwidth memory for local LLM execution and introduces the first 2-nanometer Apple Silicon.

August 25, 2026 · Apple, Mac Studio, Mac Mini, AI, Silicon

Apple has expanded its desktop lineup with the debut of the M6 chip and the M5 Ultra, positioning its hardware as a primary platform for local AI development. The new Mac Studio models are now available for preorder, shipping on September 22, 2026.

A shift to quad-die architecture

The flagship M5 Ultra chip represents a significant architectural pivot from the M3 Ultra. While previous Ultra chips connected two Max dies via “Ultrafusion” bridge technology, the M5 Ultra is Apple’s first “quad-die” design, fusing two dual-die M5 Max chips into a single unit. According to Engadget, this allows for a sixfold increase in connection density and boosts interconnection speeds to 4.4TB/s, a substantial jump from the 2.5TB/s speed of the M3 Ultra.

The M5 Ultra provides up to 36 processing cores—comprising 12 “super” cores and 24 performance cores—compared to the 32 cores on the M3 Ultra. These hardware improvements result in a 25 percent increase in single-threaded performance and up to 30 percent faster multithreaded speeds, according to Engadget.

Local AI and memory scaling

A core focus of these updates is the ability to run large language models (LLMs) locally without relying on cloud infrastructure. Apple highlights its RAM architecture as a way to eliminate memory bottlenecks for developers using tools like OpenClaw, according to CNBC Technology.

The M5 Ultra supports up to 512GB of unified memory with a 50 percent increase in bandwidth over the M3 Ultra, reaching 1.2TB/s. Apple states this allows users to “run huge LLMs with hundreds of billions of parameters entirely on device.” To facilitate large-scale research, Apple revealed that multiple Mac Studios can be linked via Thunderbolt 5 to pool memory and run frontier-level models exceeding one trillion parameters.

Additionally, the M5 Ultra includes a redesigned GPU featuring a Neural Accelerator in each of its 80 cores. This setup reportedly boosts AI compute speeds up to 4.5 times compared to the M3 Ultra, according to Engadget. For content creators, the chip introduces an updated Media Engine capable of playing 33 concurrent streams of 8K 30 fps ProRes video.

The M6 debut and Mac Mini updates

The Mac Mini is receiving its first M6-generation chip, marking Apple’s transition to TSMC’s 2-nanometer process technology. This new processor features a 12-core CPU—a first for the company—and a dual 16-core Neural Engine.

According to Wired, the M6 is 40 percent faster than the M4 and 20 percent faster than the M5 in multithreaded tasks. The Mac Mini also gains higher throughput with bandwidth up to 170 GB/s and a baseline of 2.5Gb Ethernet. While the M6 model starts at $899, Apple is also offering an M5 Pro version for this chassis starting at $1,699.

Pricing and connectivity

Apple has increased prices across the desktop lineup to account for rising memory costs.

• Mac Mini: The M6 model starts at $899 (a $300 increase over the 2024 M4 base model).
• Mac Studio (M5 Max): Starts at $2,499.
• Mac Studio (M5 Ultra): Starts at $5,499 for a configuration with 96GB of RAM and a 1TB SSD.

The top-end M5 Ultra configuration featuring 512GB of unified memory is scheduled for an October release; Engadget estimates this specific model could cost approximately $15,000.

Connectivity across the new lineup has been modernized to support Wi-Fi 7, Bluetooth 6, and Thunderbolt 5. The M5 Ultra also introduces PCIe Gen 6 internal SSD architecture. While Apple claims these speeds reach up to 128GB/s for x16 lane devices, Engadget notes that current hardware remains limited by external storage, with the fastest Micron SSD currently topping out at 28GB/s.

Performance benchmarks

Apple provided several specific performance lifts when comparing the M5 Ultra to the M3 Ultra:

• CopyCat ML training (Foundry Nuke): Up to 3.3x faster.
• LLM prompt processing (LM Studio): Up to 4x faster.
• Text-to-image LLM performance: Up to 4.3x faster.
• Scene rendering (Maxon Redshift): Up to 1.7x faster.

For professional workflows, the M5 Max version of the Mac Studio is cited by CNBC Technology as capable of running LLMs nearly four times faster than its predecessor.